Initial upload: evidence, control matrix, incidents, claims, benchmark runs and results (study cut-off 2026-10-15)
7c58952 verified Download raw/matrix/controls.csv from thanatchon36/harnesswatch: direct link, hf CLI and curl.
- Browser
- Download file 268 kB
-
https://huggingface.co/datasets/thanatchon36/harnesswatch/resolve/main/raw/matrix/controls.csv
- Command line
-
hf download hf://datasets/thanatchon36/harnesswatch/raw/matrix/controls.csv
-
curl -L -o controls.csv https://huggingface.co/datasets/thanatchon36/harnesswatch/resolve/main/raw/matrix/controls.csv
268 kB
| harness,tier,version,dimension,score,na_reason,rationale,evidence,rater | |
| aider,A,0.86.2,D01,1,,"Shell suggestions need explicit yes, but there are no allow/deny rules, edits and auto-commits happen without asking, and repo-controlled .aider.conf.yml can set test/lint commands that run without prompt.",[from-official-docs] https://github.com/Aider-AI/aider/blob/main/aider/io.py; [from-official-docs] https://github.com/Aider-AI/aider/blob/main/aider/args.py; [from-official-docs] https://github.com/Aider-AI/aider/blob/main/aider/main.py,rater1 | |
| aider,A,0.86.2,D02,0,,No sandbox.,[from-official-docs] https://github.com/Aider-AI/aider/blob/main/README.md,rater1 | |
| aider,A,0.86.2,D03,1,,Adding a URL to chat asks for confirmation (auto-yes under --yes-always); no control for shell traffic.,[from-official-docs] https://github.com/Aider-AI/aider/blob/main/aider/coders/base_coder.py,rater1 | |
| aider,A,0.86.2,D04,0,,Repo .aider.conf.yml and .env (override=True) load silently from git root/cwd; no trust concept.,[from-official-docs] https://github.com/Aider-AI/aider/blob/main/aider/main.py,rater1 | |
| aider,A,0.86.2,D05,N/A,N/A (feature absent: no MCP client in aider source).,N/A (feature absent: no MCP client in aider source).,,rater1 | |
| aider,A,0.86.2,D06,N/A,N/A (feature absent: no plugin/extension mechanism).,N/A (feature absent: no plugin/extension mechanism).,,rater1 | |
| aider,A,0.86.2,D07,1,,"Keys in env/.env/plaintext config, history files in repo root; no redaction or read protection documented.",[from-official-docs] https://github.com/Aider-AI/aider/blob/main/aider/args.py; [from-official-docs] https://github.com/Aider-AI/aider/blob/main/aider/main.py,rater1 | |
| aider,A,0.86.2,D08,1,,"Scripted mode (--message/--yes-always) keeps the explicit-yes gate for shell, but there is no official CI integration with documented minimal permissions or actor checks.",[from-official-docs] https://github.com/Aider-AI/aider/blob/main/aider/args.py; [from-official-docs] https://github.com/Aider-AI/aider/blob/main/aider/io.py,rater1 | |
| aider,A,0.86.2,D09,N/A,N/A (feature absent: no persistent agent memory; chat-history restore is opt-in).,N/A (feature absent: no persistent agent memory; chat-history restore is opt-in).,,rater1 | |
| aider,A,0.86.2,D10,N/A,N/A (feature absent: no sub-agents).,N/A (feature absent: no sub-agents).,,rater1 | |
| aider,A,0.86.2,D11,1,,Local markdown chat history and optional LLM/analytics logs; no structured tool-call log or export.,[from-official-docs] https://github.com/Aider-AI/aider/blob/main/aider/args.py,rater1 | |
| aider,A,0.86.2,D12,1,,"Analytics are opt-in (sampled prompt), but no retention period is documented, which is the score-1 clause.",[from-official-docs] https://github.com/Aider-AI/aider/blob/main/aider/analytics.py,rater1 | |
| aider,A,0.86.2,D13,1,,"PyPI and curl|sh installer without documented checksum, signature or provenance.",[from-official-docs] https://github.com/Aider-AI/aider/blob/main/aider/website/docs/install.md; [from-official-docs] https://github.com/Aider-AI/aider/blob/main/aider/versioncheck.py,rater1 | |
| aider,A,0.86.2,D14,N/A,N/A (not documented: no managed-policy facts collected; the rubric's 0-for-no-enterprise-offering clause needs a documented absence).,N/A (not documented: no managed-policy facts collected; the rubric's 0-for-no-enterprise-offering clause needs a documented absence).,,rater1 | |
| aider,A,0.86.2,D15,0,,No SECURITY.md and no published advisories found.,[from-official-docs] https://github.com/Aider-AI/aider/blob/main/,rater1 | |
| amp,B,0.0.1792036844-g99022c,D01,1,,No approval prompts by default; per-call approval only via a user/admin policy plugin.,[from-official-docs] https://ampcode.com/docs/markdown/tools; [from-official-docs] https://ampcode.com/docs/markdown/customize/plugins,rater1 | |
| amp,B,0.0.1792036844-g99022c,D02,1,,Isolation only via opt-in remote orbs (--executor orb); no local sandbox.,[from-official-docs] https://ampcode.com/docs/markdown/cli; [from-official-docs] https://ampcode.com/docs/markdown/cli/settings,rater1 | |
| amp,B,0.0.1792036844-g99022c,D03,0,,No agent egress allowlist documented; only the client's own required domains are listed.,[from-official-docs] https://ampcode.com/security,rater1 | |
| amp,B,0.0.1792036844-g99022c,D04,1,,"Workspace MCP servers need approval, but project plugins and AGENTS.md load without a trust prompt.",[from-official-docs] https://ampcode.com/docs/markdown/customize/mcp; [from-official-docs] https://ampcode.com/docs/markdown/customize/plugins; [from-official-docs] https://ampcode.com/docs/markdown/customize/agents-md,rater1 | |
| amp,B,0.0.1792036844-g99022c,D05,1,,"Workspace servers need approval and an enterprise registry allowlist exists, but amp.mcpPermissions defaults to allow and is per server, not per tool; no change detection.",[from-official-docs] https://ampcode.com/docs/markdown/cli/settings; [from-official-docs] https://ampcode.com/docs/markdown/enterprise/mcp-registry-allowlist; [from-official-docs] https://ampcode.com/docs/markdown/customize/mcp,rater1 | |
| amp,B,0.0.1792036844-g99022c,D06,1,,"Plugins are arbitrary JS/TS run in the user's environment, project plugins load without prompt; no vetting or signing.",[from-official-docs] https://ampcode.com/docs/markdown/customize/plugins; [from-official-docs] https://ampcode.com/docs/markdown/cli/settings,rater1 | |
| amp,B,0.0.1792036844-g99022c,D07,1,,"Plaintext ~/.local/share/amp/secrets.json, mode not documented; best-effort redaction. Score 2 needs keychain or verified 0600.",[from-official-docs] https://ampcode.com/security; [from-official-docs] https://ampcode.com/docs/markdown/cli/execute-mode,rater1 | |
| amp,B,0.0.1792036844-g99022c,D08,1,,Execute mode inherits the no-approval default; docs recommend a policy plugin or isolated environment for untrusted input; no official CI action hardening.,[from-official-docs] https://ampcode.com/docs/markdown/cli/execute-mode; [from-official-docs] https://ampcode.com/docs/markdown/tools,rater1 | |
| amp,B,0.0.1792036844-g99022c,D09,N/A,"N/A (no agent-written memory feature documented; AGENTS.md guidance files are user-authored instructions, threads are server-side history).","N/A (no agent-written memory feature documented; AGENTS.md guidance files are user-authored instructions, threads are server-side history).",[from-official-docs] https://ampcode.com/docs/markdown/customize/agents-md,rater1 | |
| amp,B,0.0.1792036844-g99022c,D10,1,,Subagents documented without any permission or environment scoping.,[from-official-docs] https://ampcode.com/docs/markdown/tools,rater1 | |
| amp,B,0.0.1792036844-g99022c,D11,2,,"Server-side thread storage of tool calls/results with a data-management API export; enterprise audit logs cover auth events only, full audit logs on request.",[from-official-docs] https://ampcode.com/security,rater1 | |
| amp,B,0.0.1792036844-g99022c,D12,2,,"Training opt-in (never on Enterprise), deletion within 30 days, minimal provider retention. Thread storage is the service function (see e1_notes). Not 3: no residency/zero-retention option or admin enforcement documented.",[from-official-docs] https://ampcode.com/security; [from-official-docs] https://ampcode.com/docs/markdown/enterprise/minimal-data-retention,rater1 | |
| amp,B,0.0.1792036844-g99022c,D13,2,,"Checksum files served with installers; auto-update can be disabled (AMP_SKIP_UPDATE_CHECK, updates.mode).",[from-official-docs] https://ampcode.com/docs/markdown/cli/settings; [from-official-docs] https://ampcode.com/security,rater1 | |
| amp,B,0.0.1792036844-g99022c,D14,2,,Local and workspace managed settings with documented precedence; MCP registry allowlist; gap: no permission model to lock.,[from-official-docs] https://ampcode.com/docs/markdown/cli/settings; [from-official-docs] https://ampcode.com/docs/markdown/enterprise/mcp-registry-allowlist,rater1 | |
| amp,B,0.0.1792036844-g99022c,D15,1,,"Disclosure contact and cash bounty, but no published advisories/CVEs.",[from-official-docs] https://ampcode.com/security,rater1 | |
| antigravity,B,CLI 1.3.5 / app 2.23.1 / IDE 2.5.5,D01,1,,"CLI default request-review asks and deny>ask>allow lists exist, but repo-controlled files can approve tool calls: .agents/hooks.json PreToolUse hooks may return auto_approve and repo custom agents declare their own commandExecutionPolicy. No central disable of --dangerously-skip-permissions.",[from-official-docs] https://antigravity.google/docs/cli/reference; [from-official-docs] https://antigravity.google/docs/cli/headless; [from-official-docs] https://github.com/google-antigravity/antigravity-cli/blob/main/CHANGELOG.md; [from-official-docs] https://antigravity.google/docs/hooks; [from-official-docs] https://antigravity.google/docs/subagents,rater1 | |
| antigravity,B,CLI 1.3.5 / app 2.23.1 / IDE 2.5.5,D02,1,,Namespaces/Seatbelt sandbox is on only under the app Default preset; CLI enableTerminalSandbox and the IDE sandbox are off by default; workspace-only write scope not documented. Patched native-tool bypass (Pillar) noted.,[from-official-docs] https://antigravity.google/docs/sandbox; [from-official-docs] https://antigravity.google/docs/settings; [third-party-report] https://www.pillar.security/blog/prompt-injection-leads-to-rce-and-sandbox-escape-in-antigravity,rater1 | |
| antigravity,B,CLI 1.3.5 / app 2.23.1 / IDE 2.5.5,D03,2,,"URL fetch asks by default (CLI 1.1.28+), sandboxed commands have no network by default with read_url allowlist, browser allowlist starts at localhost. Not 3: no central allowlist management; unsandboxed commands unrestricted. Earlier webhook.site allowlist entry (3P) fixed.",[from-official-docs] https://antigravity.google/docs/sandbox; [from-official-docs] https://github.com/google-antigravity/antigravity-cli/blob/main/CHANGELOG.md; [from-official-docs] https://antigravity.google/docs/ide/allowlist-denylist; [third-party-report] https://www.promptarmor.com/resources/google-antigravity-exfiltrates-data,rater1 | |
| antigravity,B,CLI 1.3.5 / app 2.23.1 / IDE 2.5.5,D04,1,,"CLI trust prompt exists and gates workspace hooks, but the docs do not show repo MCP config, custom agents or plugins as gated, and the hooks page documents no trust requirement.",[from-official-docs] https://antigravity.google/docs/getting-started; [from-official-docs] https://antigravity.google/docs/hooks; [from-official-docs] https://github.com/google-antigravity/antigravity-cli/blob/main/CHANGELOG.md; [from-official-docs] https://antigravity.google/docs/subagents; [third-party-report] https://mindgard.ai/blog/google-antigravity-persistent-code-execution-vulnerability,rater1 | |
| antigravity,B,CLI 1.3.5 / app 2.23.1 / IDE 2.5.5,D05,1,,"Unconfigured MCP tools default to Ask with per-tool patterns, but no project-server approval, managed MCP allowlist or change detection is documented.",[from-official-docs] https://antigravity.google/docs/mcp; [third-party-report] https://embracethered.com/blog/posts/2025/security-keeps-google-antigravity-grounded/,rater1 | |
| antigravity,B,CLI 1.3.5 / app 2.23.1 / IDE 2.5.5,D06,1,,"Plugins bundle MCP servers, hooks and binaries; workspace plugins in .agents/plugins activate for the project; no signing or vetting documented.",[from-official-docs] https://antigravity.google/docs/cli/gcli-migration; [from-official-docs] https://antigravity.google/docs/plugins,rater1 | |
| antigravity,B,CLI 1.3.5 / app 2.23.1 / IDE 2.5.5,D07,1,,Session tokens in the OS keyring but file fallback without D-Bus; MCP OAuth tokens in a plaintext JSON file (3P: mode 0644); no log retention documented; .env read block bypassed via terminal (3P).,[from-official-docs] https://antigravity.google/docs/cli/gcli-migration; [from-official-docs] https://antigravity.google/docs/mcp; [from-official-docs] https://github.com/google-antigravity/antigravity-cli/blob/main/CHANGELOG.md; [third-party-report] https://www.promptarmor.com/resources/google-antigravity-exfiltrates-data; [third-party-report] https://github.com/google-antigravity/antigravity-cli/issues/815,rater1 | |
| antigravity,B,CLI 1.3.5 / app 2.23.1 / IDE 2.5.5,D08,1,,"Headless soft-denies unapproved actions (safe default), but no official CI integration with minimal permissions or actor checks is documented, which score 2 requires.",[from-official-docs] https://antigravity.google/docs/cli/headless,rater1 | |
| antigravity,B,CLI 1.3.5 / app 2.23.1 / IDE 2.5.5,D09,1,,"Global rules files and 'knowledge items' persist in app-data dirs; no write confirmation, project scoping or off switch documented.",[from-official-docs] https://antigravity.google/docs/rules; [from-official-docs] https://antigravity.google/docs/agent-settings,rater1 | |
| antigravity,B,CLI 1.3.5 / app 2.23.1 / IDE 2.5.5,D10,1,,"Per-subagent commandExecutionPolicy (default sandbox), tools and mcpServers exist, but subagents inherit accept-edits and propagation of bypass is not documented; workaround of denied requests fixed in CLI 1.2.15.",[from-official-docs] https://antigravity.google/docs/subagents; [from-official-docs] https://antigravity.google/docs/cli/modes; [from-official-docs] https://github.com/google-antigravity/antigravity-cli/blob/main/CHANGELOG.md,rater1 | |
| antigravity,B,CLI 1.3.5 / app 2.23.1 / IDE 2.5.5,D11,1,,Local diagnostic log only; enterprise request/response logging to a GCP project exists but structured local tool-call logs are not documented (score-2 clause unmet).,[from-official-docs] https://antigravity.google/docs/enterprise,rater1 | |
| antigravity,B,CLI 1.3.5 / app 2.23.1 / IDE 2.5.5,D12,1,,Telemetry on by default with opt-out; Terms use interactions to improve products/ML by default with opt-out; retention not documented.,[from-official-docs] https://antigravity.google/docs/cli/reference; [from-official-docs] https://antigravity.google/docs/enterprise; [from-official-docs] https://antigravity.google/terms,rater1 | |
| antigravity,B,CLI 1.3.5 / app 2.23.1 / IDE 2.5.5,D13,2,,"Installer verifies SHA-512 against a manifest; auto-update can be disabled (AGY_CLI_DISABLE_AUTO_UPDATE). Not 3: no signatures, provenance, SBOM or pinning.",[from-official-docs] https://antigravity.google/docs/cli/install; [from-official-docs] https://antigravity.google/cli/install.sh; [from-official-docs] https://antigravity.google/docs/cli/troubleshooting,rater1 | |
| antigravity,B,CLI 1.3.5 / app 2.23.1 / IDE 2.5.5,D14,1,,"Enterprise identity, licensing and VPC-SC governance only; no managed settings or admin-enforced permission policy.",[from-official-docs] https://antigravity.google/docs/enterprise; [from-official-docs] https://antigravity.google/docs/ide/overview; [from-official-docs] https://antigravity.google/docs/sdk/policies,rater1 | |
| antigravity,B,CLI 1.3.5 / app 2.23.1 / IDE 2.5.5,D15,0,,"Docs-level facts show no SECURITY.md/security.txt and no advisories or CVEs; the Google Bug Hunters VRP is cited only by third-party reports, which cannot raise a score.",[from-official-docs] https://antigravity.google/.well-known/security.txt; [from-official-docs] https://github.com/google-antigravity/antigravity-cli/security/advisories; [third-party-report] https://www.pillar.security/blog/prompt-injection-leads-to-rce-and-sandbox-escape-in-antigravity,rater1 | |
| augment,B,0.36.0,D01,1,,"toolPermissions allow/deny rules exist (most restrictive wins), but unmatched tools follow implicit behaviour and docs say Auggie runs commands and tools automatically; rules not enforced in the IDE extension.",[from-official-docs] https://docs.augmentcode.com/cli/permissions; [from-official-docs] https://docs.augmentcode.com/cli/integrations,rater1 | |
| augment,B,0.36.0,D02,N/A,N/A (not documented: the only fact is an unverified statement of absence).,N/A (not documented: the only fact is an unverified statement of absence).,,rater1 | |
| augment,B,0.36.0,D03,0,,No destination allowlist of any kind; only whole tools can be removed.,[from-official-docs] https://docs.augmentcode.com/cli/reference,rater1 | |
| augment,B,0.36.0,D04,0,,Project .augment/settings.json including hooks loads automatically for anyone in the repo; no trust concept (only an indexing confirmation).,[from-official-docs] https://docs.augmentcode.com/cli/hooks; [from-official-docs] https://docs.augmentcode.com/cli/reference,rater1 | |
| augment,B,0.36.0,D05,1,,"MCP servers need user setup in ~/.augment/settings.json, but no approval, allowlist, managed policy or change detection is documented.",[from-official-docs] https://docs.augmentcode.com/cli/integrations,rater1 | |
| augment,B,0.36.0,D06,1,,Skills load automatically from workspace .augment/skills and .claude/skills; marketplace vetting not documented.,[from-official-docs] https://docs.augmentcode.com/cli/reference,rater1 | |
| augment,B,0.36.0,D07,1,,Session token stored locally with storage mechanism and mode undocumented; no redaction or secret-file protection documented.,[from-official-docs] https://docs.augmentcode.com/cli/setup-auggie/authentication,rater1 | |
| augment,B,0.36.0,D08,1,,--print falls back to implicit (auto-run) behaviour unless rules are set; no official CI action hardening documented.,[from-official-docs] https://docs.augmentcode.com/cli/automation/overview; [from-official-docs] https://docs.augmentcode.com/cli/permissions,rater1 | |
| augment,B,0.36.0,D09,N/A,N/A (no agent-written memory feature documented; rules files are user-authored instructions).,N/A (no agent-written memory feature documented; rules files are user-authored instructions).,[from-official-docs] https://docs.augmentcode.com/cli/rules,rater1 | |
| augment,B,0.36.0,D10,N/A,N/A (not documented: no sub-agent facts collected).,N/A (not documented: no sub-agent facts collected).,,rater1 | |
| augment,B,0.36.0,D11,1,,Unstructured log file (default temp) only; no structured tool-call log or export documented.,[from-official-docs] https://docs.augmentcode.com/cli/reference,rater1 | |
| augment,B,0.36.0,D12,1,,"No training on customer code, but telemetry opt-out and retention period are not documented (score-1 clause).",[from-official-docs] https://www.augmentcode.com/security; [from-official-docs] https://docs.augmentcode.com/cli/overview,rater1 | |
| augment,B,0.36.0,D13,1,,npm registry install without signature or provenance; auto-update on by default (can be disabled).,[from-official-docs] https://docs.augmentcode.com/cli/reference; [from-official-docs] https://registry.npmjs.org/@augmentcode/auggie,rater1 | |
| augment,B,0.36.0,D14,2,,System-wide /etc/augment/settings.json has highest precedence and cannot be overridden; coverage of MCP/plugins/telemetry not documented and not verified.,[from-official-docs] https://docs.augmentcode.com/cli/hooks,rater1 | |
| augment,B,0.36.0,D15,1,,Security contact and GHSA reporting route; 0 published advisories; no bounty.,[from-official-docs] https://github.com/augmentcode/auggie/blob/main/SECURITY.md; [from-official-docs] https://github.com/augmentcode/auggie/security/advisories,rater1 | |
| azure-ai-foundry-agent-service,C,GA (docs page updated 2026-10-15); voice agents and agent guardrails in preview,D01,1,,Tier C: tool-call guardrails (annotate/block) exist in Preview and apply only when assigned; human approval defaults for MCP tools could not be documented (404).,[from-official-docs] https://learn.microsoft.com/en-us/azure/foundry/guardrails/guardrails-overview,rater1 | |
| azure-ai-foundry-agent-service,C,GA (docs page updated 2026-10-15); voice agents and agent guardrails in preview,D02,3,,Tier C: hosted agents run each session in a provider-managed VM-isolated sandbox with BYO VNet and (Preview) egress controls.,[from-official-docs] https://learn.microsoft.com/en-us/azure/foundry/agents/overview; [from-official-docs] https://learn.microsoft.com/en-us/azure/foundry/guardrails/guardrails-overview,rater1 | |
| azure-ai-foundry-agent-service,C,GA (docs page updated 2026-10-15); voice agents and agent guardrails in preview,D03,3,,Tier C: managed runtime with private networking/BYO VNet and egress policy (Preview).,[from-official-docs] https://learn.microsoft.com/en-us/azure/foundry/guardrails/guardrails-overview; [from-official-docs] https://learn.microsoft.com/en-us/azure/foundry/agents/overview,rater1 | |
| azure-ai-foundry-agent-service,C,GA (docs page updated 2026-10-15); voice agents and agent guardrails in preview,D04,N/A,N/A (cloud service; reads no repository-level configuration).,N/A (cloud service; reads no repository-level configuration).,,rater1 | |
| azure-ai-foundry-agent-service,C,GA (docs page updated 2026-10-15); voice agents and agent guardrails in preview,D05,1,,"Tier C: curated toolbox endpoint with central auth exists, but a default-on tool-call approval is not documented (MCP approval page 404) and unauthenticated MCP access is allowed.",[from-official-docs] https://learn.microsoft.com/en-us/azure/foundry/agents/overview,rater1 | |
| azure-ai-foundry-agent-service,C,GA (docs page updated 2026-10-15); voice agents and agent guardrails in preview,D06,N/A,N/A (not documented: no extension facts collected).,N/A (not documented: no extension facts collected).,,rater1 | |
| azure-ai-foundry-agent-service,C,GA (docs page updated 2026-10-15); voice agents and agent guardrails in preview,D07,2,,"Tier C: managed identity (federated credential, no stored secret) documented; only MCP/A2A tools support agent identity and redacted traces are not documented.",[from-official-docs] https://learn.microsoft.com/en-us/azure/foundry/agents/concepts/agent-identity,rater1 | |
| azure-ai-foundry-agent-service,C,GA (docs page updated 2026-10-15); voice agents and agent guardrails in preview,D08,N/A,N/A (not documented: no facts on unattended-run enforcement).,N/A (not documented: no facts on unattended-run enforcement).,,rater1 | |
| azure-ai-foundry-agent-service,C,GA (docs page updated 2026-10-15); voice agents and agent guardrails in preview,D09,1,,"Memory is a built-in tool; scoping, visibility and off switch not documented.",[from-official-docs] https://learn.microsoft.com/en-us/azure/foundry/agents/overview,rater1 | |
| azure-ai-foundry-agent-service,C,GA (docs page updated 2026-10-15); voice agents and agent guardrails in preview,D10,1,,Tier C: all unpublished agents in a project share one identity (vendor calls it a broader blast radius); distinct identity needs publishing and manual RBAC.,[from-official-docs] https://learn.microsoft.com/en-us/azure/foundry/agents/concepts/agent-identity,rater1 | |
| azure-ai-foundry-agent-service,C,GA (docs page updated 2026-10-15); voice agents and agent guardrails in preview,D11,2,,Tier C: built-in end-to-end tracing and Application Insights integration. Admin-change audit enforceable by policy not documented.,[from-official-docs] https://learn.microsoft.com/en-us/azure/foundry/agents/overview,rater1 | |
| azure-ai-foundry-agent-service,C,GA (docs page updated 2026-10-15); voice agents and agent guardrails in preview,D12,1,,"Customer-owned storage for conversation state is supported, but retention and training terms are not documented in the evidence.",[from-official-docs] https://learn.microsoft.com/en-us/azure/foundry/agents/overview,rater1 | |
| azure-ai-foundry-agent-service,C,GA (docs page updated 2026-10-15); voice agents and agent guardrails in preview,D13,N/A,N/A (not documented: no distribution facts for the managed service or its SDK).,N/A (not documented: no distribution facts for the managed service or its SDK).,,rater1 | |
| azure-ai-foundry-agent-service,C,GA (docs page updated 2026-10-15); voice agents and agent guardrails in preview,D14,2,,"Tier C: Entra Agent ID Conditional Access, network access and governance plus RBAC; coverage of tools/MCP and telemetry not documented.",[from-official-docs] https://learn.microsoft.com/en-us/azure/foundry/agents/concepts/agent-identity,rater1 | |
| azure-ai-foundry-agent-service,C,GA (docs page updated 2026-10-15); voice agents and agent guardrails in preview,D15,N/A,N/A (only a third-party report; no documented policy or advisory facts).,N/A (only a third-party report; no documented policy or advisory facts).,[third-party-report] https://onlyfm252.starlabs.sg/patch-tuesday/2026-05/cve-2026-35435/,rater1 | |
| bedrock-agentcore,C,"AgentCore harness GA (docs 2026-10-15); SDKs bedrock-agentcore 1.18.1+, starter toolkit 0.3.14+",D01,1,,Tier C: all tools including shell are allowed unless allowedTools is set; HITL via inline tools and Cedar Gateway policy exist but are not on by default.,[from-official-docs] https://docs.aws.amazon.com/bedrock-agentcore/latest/devguide/harness-tools.html; [from-official-docs] https://docs.aws.amazon.com/bedrock-agentcore/latest/devguide/policy.html,rater1 | |
| bedrock-agentcore,C,"AgentCore harness GA (docs 2026-10-15); SDKs bedrock-agentcore 1.18.1+, starter toolkit 0.3.14+",D02,3,,Tier C: each session runs in its own Firecracker microVM; VPC mode provides egress control. The DNS covert channel (3P) concerns network egress and is scored under D03.,[from-official-docs] https://docs.aws.amazon.com/bedrock-agentcore/latest/devguide/harness-security.html,rater1 | |
| bedrock-agentcore,C,"AgentCore harness GA (docs 2026-10-15); SDKs bedrock-agentcore 1.18.1+, starter toolkit 0.3.14+",D03,2,,"Tier C: VPC/egress policy exists (3) but sessions use the public network by default and a third-party report documents DNS egress from the Code Interpreter sandbox mode, lowering to 2.",[from-official-docs] https://docs.aws.amazon.com/bedrock-agentcore/latest/devguide/harness-security.html; [third-party-report] https://labs.cloudsecurityalliance.org/research/csa-research-note-ai-sandbox-dns-exfiltration-bedrock-langsm/; [third-party-report] https://thehackernews.com/2026/03/ai-flaws-in-amazon-bedrock-langsmith.html,rater1 | |
| bedrock-agentcore,C,"AgentCore harness GA (docs 2026-10-15); SDKs bedrock-agentcore 1.18.1+, starter toolkit 0.3.14+",D04,N/A,N/A (cloud service; no local workspace configuration).,N/A (cloud service; no local workspace configuration).,,rater1 | |
| bedrock-agentcore,C,"AgentCore harness GA (docs 2026-10-15); SDKs bedrock-agentcore 1.18.1+, starter toolkit 0.3.14+",D05,2,,"Tier C: Gateway inbound/outbound auth and Registry with governed approve workflow; plaintext headers allowed in remote_mcp and tools overridable per invocation, so not 3.",[from-official-docs] https://docs.aws.amazon.com/bedrock-agentcore/latest/devguide/harness-tools.html; [from-official-docs] https://docs.aws.amazon.com/bedrock-agentcore/latest/devguide/what-is-bedrock-agentcore.html,rater1 | |
| bedrock-agentcore,C,"AgentCore harness GA (docs 2026-10-15); SDKs bedrock-agentcore 1.18.1+, starter toolkit 0.3.14+",D06,1,,Harness treats skill content as trusted input with no validation; skills can be overridden per call.,[from-official-docs] https://docs.aws.amazon.com/bedrock-agentcore/latest/devguide/harness-security.html,rater1 | |
| bedrock-agentcore,C,"AgentCore harness GA (docs 2026-10-15); SDKs bedrock-agentcore 1.18.1+, starter toolkit 0.3.14+",D07,2,,Tier C: Token Vault and IAM execution roles; per-user scoping lost with SigV4. Patched OTel-span disclosure CVE-2026-15737 noted.,[from-official-docs] https://docs.aws.amazon.com/bedrock-agentcore/latest/devguide/harness-security.html; [from-official-docs] https://github.com/aws/bedrock-agentcore-sdk-python/security/advisories/GHSA-hqf8-7w95-9r33,rater1 | |
| bedrock-agentcore,C,"AgentCore harness GA (docs 2026-10-15); SDKs bedrock-agentcore 1.18.1+, starter toolkit 0.3.14+",D08,N/A,N/A (not documented: no facts on unattended-run safeguards in this dimension).,N/A (not documented: no facts on unattended-run safeguards in this dimension).,,rater1 | |
| bedrock-agentcore,C,"AgentCore harness GA (docs 2026-10-15); SDKs bedrock-agentcore 1.18.1+, starter toolkit 0.3.14+",D09,2,,Memory scoped by actorId+sessionId with expiry; can be disabled (CLI default disabled). Not 3: service API provisions memory by default; no scanning.,[from-official-docs] https://docs.aws.amazon.com/bedrock-agentcore/latest/devguide/harness-memory.html,rater1 | |
| bedrock-agentcore,C,"AgentCore harness GA (docs 2026-10-15); SDKs bedrock-agentcore 1.18.1+, starter toolkit 0.3.14+",D10,1,,Any principal past the IAM/JWT gate reaches the full session and tools; model parameters and skills overrides are not validated.,[from-official-docs] https://docs.aws.amazon.com/bedrock-agentcore/latest/devguide/harness-security.html,rater1 | |
| bedrock-agentcore,C,"AgentCore harness GA (docs 2026-10-15); SDKs bedrock-agentcore 1.18.1+, starter toolkit 0.3.14+",D11,2,,Tier C: every action traced automatically to CloudWatch/X-Ray; policy decisions logged.,[from-official-docs] https://docs.aws.amazon.com/bedrock-agentcore/latest/devguide/harness.html,rater1 | |
| bedrock-agentcore,C,"AgentCore harness GA (docs 2026-10-15); SDKs bedrock-agentcore 1.18.1+, starter toolkit 0.3.14+",D12,1,,Content may be used to improve the customer's service; retention not documented.,[from-official-docs] https://docs.aws.amazon.com/bedrock-agentcore/latest/devguide/what-is-bedrock-agentcore.html,rater1 | |
| bedrock-agentcore,C,"AgentCore harness GA (docs 2026-10-15); SDKs bedrock-agentcore 1.18.1+, starter toolkit 0.3.14+",D13,1,,"Managed image pulled from ECR; no signature, provenance or SBOM documented.",[from-official-docs] https://docs.aws.amazon.com/bedrock-agentcore/latest/devguide/harness-security.html,rater1 | |
| bedrock-agentcore,C,"AgentCore harness GA (docs 2026-10-15); SDKs bedrock-agentcore 1.18.1+, starter toolkit 0.3.14+",D14,2,,"Tier C: IAM on harness/runtime ARNs counts as managed policy; organisation-wide coverage of tools, egress and telemetry not documented.",[from-official-docs] https://docs.aws.amazon.com/bedrock-agentcore/latest/devguide/harness-security.html,rater1 | |
| bedrock-agentcore,C,"AgentCore harness GA (docs 2026-10-15); SDKs bedrock-agentcore 1.18.1+, starter toolkit 0.3.14+",D15,1,,"Six GHSA/CVE advisories show a working disclosure channel, but no security policy or bounty is documented in the evidence, so score 2 ('policy plus advisories') is not met.",[from-official-docs] https://github.com/aws/bedrock-agentcore-sdk-python/security/advisories,rater1 | |
| claude-agent-sdk,C,"claude-agent-sdk 0.2.167 (PyPI, bundles Claude Code CLI 2.1.300); @anthropic-ai/claude-agent-sdk 0.3.300 (npm, GitHub release tag v0.3.300)",D01,1,,"Tier C: can_use_tool callback and deny-in-every-mode exist, but with permission_mode unset the start mode comes from CLI settings or a built-in default 'which can be auto mode', so approval is not on by default in the reference loop (same issue as claude-code D01); allowed_tools and bypass shadow the callback.",[from-official-docs] https://github.com/anthropics/claude-agent-sdk-python/blob/main/src/claude_agent_sdk/types.py; [from-official-docs] https://code.claude.com/docs/en/agent-sdk/permissions; [from-official-docs] https://github.com/anthropics/claude-agent-sdk-python/blob/main/src/claude_agent_sdk/_internal/query.py,rater1 | |
| claude-agent-sdk,C,"claude-agent-sdk 0.2.167 (PyPI, bundles Claude Code CLI 2.1.300); @anthropic-ai/claude-agent-sdk 0.3.300 (npm, GitHub release tag v0.3.300)",D02,1,,"Tier C: optional bubblewrap/Seatbelt sandbox (enabled=false by default, unsandboxed commands allowed); agent runs in the caller's isolation boundary.",[from-official-docs] https://github.com/anthropics/claude-agent-sdk-python/blob/main/src/claude_agent_sdk/types.py; [from-official-docs] https://code.claude.com/docs/en/agent-sdk/secure-deployment,rater1 | |
| claude-agent-sdk,C,"claude-agent-sdk 0.2.167 (PyPI, bundles Claude Code CLI 2.1.300); @anthropic-ai/claude-agent-sdk 0.3.300 (npm, GitHub release tag v0.3.300)",D03,1,,"Tier C: domain allowlisting only via the optional sandbox proxy (no TLS inspection, domain fronting documented).",[from-official-docs] https://code.claude.com/docs/en/agent-sdk/secure-deployment,rater1 | |
| claude-agent-sdk,C,"claude-agent-sdk 0.2.167 (PyPI, bundles Claude Code CLI 2.1.300); @anthropic-ai/claude-agent-sdk 0.3.300 (npm, GitHub release tag v0.3.300)",D04,1,,"Runtime auto-loads project .claude/settings.json (hooks, allow rules) by default; isolation (setting_sources=[]) is opt-in, i.e. the gate is off by default.",[from-official-docs] https://github.com/anthropics/claude-agent-sdk-python/blob/main/src/claude_agent_sdk/types.py; [from-official-docs] https://code.claude.com/docs/en/agent-sdk/permissions,rater1 | |
| claude-agent-sdk,C,"claude-agent-sdk 0.2.167 (PyPI, bundles Claude Code CLI 2.1.300); @anthropic-ai/claude-agent-sdk 0.3.300 (npm, GitHub release tag v0.3.300)",D05,1,,Tier C: approval primitive via callback and anchored mcp__ allow rules; default-on approval for MCP tools not documented.,[from-official-docs] https://code.claude.com/docs/en/agent-sdk/permissions; [from-official-docs] https://github.com/anthropics/claude-agent-sdk-python/blob/main/src/claude_agent_sdk/types.py,rater1 | |
| claude-agent-sdk,C,"claude-agent-sdk 0.2.167 (PyPI, bundles Claude Code CLI 2.1.300); @anthropic-ai/claude-agent-sdk 0.3.300 (npm, GitHub release tag v0.3.300)",D06,N/A,N/A (tier C: skills are application-supplied local files; no built-in marketplace or registry in the SDK layer).,N/A (tier C: skills are application-supplied local files; no built-in marketplace or registry in the SDK layer).,[from-official-docs] https://github.com/anthropics/claude-agent-sdk-python/blob/main/src/claude_agent_sdk/types.py; [from-official-docs] https://github.com/anthropics/claude-agent-sdk-python/blob/main/CHANGELOG.md,rater1 | |
| claude-agent-sdk,C,"claude-agent-sdk 0.2.167 (PyPI, bundles Claude Code CLI 2.1.300); @anthropic-ai/claude-agent-sdk 0.3.300 (npm, GitHub release tag v0.3.300)",D07,1,,"Tier C: keys come from the caller's environment, which the CLI subprocess inherits in full; credential proxy only recommended.",[from-official-docs] https://github.com/anthropics/claude-agent-sdk-python/blob/main/src/claude_agent_sdk/_internal/transport/subprocess_cli.py; [from-official-docs] https://code.claude.com/docs/en/agent-sdk/secure-deployment,rater1 | |
| claude-agent-sdk,C,"claude-agent-sdk 0.2.167 (PyPI, bundles Claude Code CLI 2.1.300); @anthropic-ai/claude-agent-sdk 0.3.300 (npm, GitHub release tag v0.3.300)",D08,2,,Tier C: unattended runs can be locked down with dontAsk (deny anything not pre-allowed) and deny rules that apply in every mode. Not 3: not the default and no platform enforcement.,[from-official-docs] https://code.claude.com/docs/en/agent-sdk/permissions,rater1 | |
| claude-agent-sdk,C,"claude-agent-sdk 0.2.167 (PyPI, bundles Claude Code CLI 2.1.300); @anthropic-ai/claude-agent-sdk 0.3.300 (npm, GitHub release tag v0.3.300)",D09,1,,"Auto-memory and CLAUDE.md come from the bundled CLI; SDK can strip the auto-memory section, but scoping, visibility and write controls are not documented here.",[from-official-docs] https://github.com/anthropics/claude-agent-sdk-python/blob/main/src/claude_agent_sdk/types.py,rater1 | |
| claude-agent-sdk,C,"claude-agent-sdk 0.2.167 (PyPI, bundles Claude Code CLI 2.1.300); @anthropic-ai/claude-agent-sdk 0.3.300 (npm, GitHub release tag v0.3.300)",D10,1,,"Subagents run in the parent's permission mode, so a parent bypass propagates to them (score-2 clause unmet); no worktree/container isolation in the SDK.",[from-official-docs] https://code.claude.com/docs/en/agent-sdk/permissions,rater1 | |
| claude-agent-sdk,C,"claude-agent-sdk 0.2.167 (PyPI, bundles Claude Code CLI 2.1.300); @anthropic-ai/claude-agent-sdk 0.3.300 (npm, GitHub release tag v0.3.300)",D11,2,,Tier C: session-store mirroring and OpenTelemetry trace propagation one setting away.,[from-official-docs] https://github.com/anthropics/claude-agent-sdk-python/blob/main/src/claude_agent_sdk/_internal/transport/subprocess_cli.py,rater1 | |
| claude-agent-sdk,C,"claude-agent-sdk 0.2.167 (PyPI, bundles Claude Code CLI 2.1.300); @anthropic-ai/claude-agent-sdk 0.3.300 (npm, GitHub release tag v0.3.300)",D12,N/A,N/A (tier C library layer: telemetry and retention belong to the bundled Claude Code CLI and the API plan; scored under claude-code).,N/A (tier C library layer: telemetry and retention belong to the bundled Claude Code CLI and the API plan; scored under claude-code).,[from-official-docs] https://github.com/anthropics/claude-agent-sdk-python/blob/main/README.md,rater1 | |
| claude-agent-sdk,C,"claude-agent-sdk 0.2.167 (PyPI, bundles Claude Code CLI 2.1.300); @anthropic-ai/claude-agent-sdk 0.3.300 (npm, GitHub release tag v0.3.300)",D13,1,,Tier C: PyPI/npm publication with a pinned bundled CLI; trusted publishing or provenance not documented.,[from-official-docs] https://github.com/anthropics/claude-agent-sdk-python/blob/main/src/claude_agent_sdk/_cli_version.py; [from-official-docs] https://github.com/anthropics/claude-agent-sdk-python/blob/main/CHANGELOG.md,rater1 | |
| claude-agent-sdk,C,"claude-agent-sdk 0.2.167 (PyPI, bundles Claude Code CLI 2.1.300); @anthropic-ai/claude-agent-sdk 0.3.300 (npm, GitHub release tag v0.3.300)",D14,2,,Organisation policies for Claude Code permissions apply to SDK sessions (org 'ask' survives bypass); full coverage and precedence for SDK sessions not documented or tested.,[from-official-docs] https://code.claude.com/docs/en/agent-sdk/secure-deployment; [from-official-docs] https://code.claude.com/docs/en/agent-sdk/permissions,rater1 | |
| claude-agent-sdk,C,"claude-agent-sdk 0.2.167 (PyPI, bundles Claude Code CLI 2.1.300); @anthropic-ai/claude-agent-sdk 0.3.300 (npm, GitHub release tag v0.3.300)",D15,1,,"A GHSA/CVE was published (HackerOne-credited) but the repo has no SECURITY.md and no policy is documented, so score 2 is not met.",[from-official-docs] https://github.com/anthropics/claude-agent-sdk-python/security/advisories/GHSA-h4mw-j7qp-8mwm; [from-official-docs] https://github.com/anthropics/claude-agent-sdk-python,rater1 | |
| claude-code,B,2.1.300,D01,1,,"Built-in starting mode on 2.1.300 is the classifier-based auto mode (interactive >=2.1.283; -p without feature flags >=2.1.285), which runs shell and edits without asking the user, so the score-1 clause 'default mode runs shell or writes without asking' applies and the score-2 clause 'default mode asks before shell commands' is not met. A classifier decision is not a user prompt. Score-3 controls otherwise exist (deny in every mode, bypass/auto not settable from project files, managed disableBypassPermissionsMode/disableAutoMode); with Manual as default this would be 3. See e1_notes.md.",[from-official-docs] https://code.claude.com/docs/en/permissions; [from-official-docs] https://code.claude.com/docs/en/settings-reference; [from-official-docs] https://code.claude.com/docs/en/permission-modes,rater1 | |
| claude-code,B,2.1.300,D02,1,,Seatbelt/bubblewrap sandbox exists but sandbox.enabled=false by default and falls open unless failIfUnavailable=true.,[from-official-docs] https://code.claude.com/docs/en/sandboxing,rater1 | |
| claude-code,B,2.1.300,D03,1,,"Proxy-enforced allowedDomains (default empty, managed lock) applies only inside the opt-in sandbox; WebFetch has a vendor hostname blocklist, not a restriction to approved hosts, and under the default auto mode fetch is not documented to ask.",[from-official-docs] https://code.claude.com/docs/en/sandboxing; [from-official-docs] https://code.claude.com/docs/en/data-usage#webfetch-domain-safety-check,rater1 | |
| claude-code,B,2.1.300,D04,2,,"Interactive trust dialog gates settings hooks, MCP approvals and project settings; -p/SDK is documented to skip it (allowed at 2). Not 3: headless needs no flag and env/hooks/.mcp.json are used in untrusted -p runs. Patched pre-trust bypasses CVE-2025-59536 and CVE-2026-21852 noted.",[from-official-docs] https://code.claude.com/docs/en/hooks#workspace-trust; [from-official-docs] https://code.claude.com/docs/en/permissions#what-runs-before-you-trust-a-folder; [from-official-docs] https://github.com/anthropics/claude-code/security/advisories/GHSA-4fgq-fpq9-mr3g,rater1 | |
| claude-code,B,2.1.300,D05,1,,"Per-server approval for .mcp.json and managed allow/deny lists exist, but per-tool MCP permissions and definition-change detection are not documented in the evidence, and -p connects project servers without asking.",[from-official-docs] https://code.claude.com/docs/en/mcp#project-server-approvals-and-workspace-trust; [from-official-docs] https://code.claude.com/docs/en/settings-reference; [from-official-docs] https://code.claude.com/docs/en/permissions#what-runs-before-you-trust-a-folder,rater1 | |
| claude-code,B,2.1.300,D06,1,,"Managed marketplace allowlist (strictKnownMarketplaces) and SHA pinning exist, but no vetting, scanning or signing is documented (score-2 clause unmet). Patched SHA-pin bypass CVE-2026-86063 noted.",[from-official-docs] https://code.claude.com/docs/en/plugin-marketplaces; [from-official-docs] https://github.com/anthropics/claude-code/security/advisories/GHSA-pq7j-f95f-qfcv; [from-official-docs] https://code.claude.com/docs/en/settings-reference,rater1 | |
| claude-code,B,2.1.300,D07,2,,"Keychain on macOS, 0600 file on Linux (verified locally); 30-day transcript retention documented; redaction only for uploaded transcripts. Not 3: local transcripts plaintext, env scrub off by default, no default secret-file read deny.",[from-official-docs] https://code.claude.com/docs/en/authentication; [verified-by-test] local:~/.claude/.credentials.json; [from-official-docs] https://code.claude.com/docs/en/data-usage#data-retention; [from-official-docs] https://code.claude.com/docs/en/env-vars#what-the-subprocess-environment-scrub-removes,rater1 | |
| claude-code,B,2.1.300,D08,1,,"-p defaults are permissive: folder treated as trusted, repo hooks/env and .mcp.json servers used, and auto mode is the starting mode when flags are not fetched. Action checks actor write access, but minimal-token permissions are not documented in the evidence.",[from-official-docs] https://code.claude.com/docs/en/cli-reference; [from-official-docs] https://code.claude.com/docs/en/github-actions; [from-official-docs] https://code.claude.com/docs/en/hooks#workspace-trust; [from-official-docs] https://code.claude.com/docs/en/permissions#what-runs-before-you-trust-a-folder,rater1 | |
| claude-code,B,2.1.300,D09,2,,"Auto memory is per-project plain files under ~/.claude/projects, visible and editable, with autoMemoryEnabled=false off switch. Not 3: on by default, writes not confirmed, no scanning documented.",[from-official-docs] https://code.claude.com/docs/en/memory,rater1 | |
| claude-code,B,2.1.300,D10,1,,"Subagents inherit parent permission rules; per-agent tools and worktree isolation exist, but whether a parent bypass mode propagates is not documented (only agent-declared bypass is ignored under managed disable), so the score-2 clause is not met.",[from-official-docs] https://code.claude.com/docs/en/sub-agents,rater1 | |
| claude-code,B,2.1.300,D11,2,,Structured JSONL transcripts with tool calls plus opt-in OpenTelemetry export. Not 3: policy enforcement of export and an admin-change audit log are not documented.,[from-official-docs] https://code.claude.com/docs/en/monitoring-usage; [from-official-docs] https://code.claude.com/docs/en/hooks,rater1 | |
| claude-code,B,2.1.300,D12,2,,"Opt-out of telemetry/error reporting documented; retention documented by plan; commercial data not trained on by default; ZDR available. Not 3: usage metrics on by default for the Claude API, admin enforcement not documented.",[from-official-docs] https://code.claude.com/docs/en/data-usage; [from-official-docs] https://code.claude.com/docs/en/data-usage#telemetry-services; [from-official-docs] https://code.claude.com/docs/en/env-vars,rater1 | |
| claude-code,B,2.1.300,D13,2,,"GPG-signed release manifest, signed package repos, macOS/Windows code signing; update channel pinnable/disableable. Not 3: no npm provenance attestation, no SBOM, Linux binaries unsigned.",[from-official-docs] https://code.claude.com/docs/en/setup; [from-official-docs] https://code.claude.com/docs/en/setup#binary-integrity-and-code-signing; [from-official-docs] https://registry.npmjs.org/@anthropic-ai/claude-code,rater1 | |
| claude-code,B,2.1.300,D14,2,,"Managed settings at top precedence cover permissions, hooks, MCP, sandbox domains and marketplaces. Not 3: managed control of telemetry not documented in the evidence and the precedence is not verified by test.",[from-official-docs] https://code.claude.com/docs/en/settings; [from-official-docs] https://code.claude.com/docs/en/managed-settings; [from-official-docs] https://code.claude.com/docs/en/settings-reference; [from-official-docs] https://code.claude.com/docs/en/sandboxing; [from-official-docs] https://code.claude.com/docs/en/plugin-marketplaces,rater1 | |
| claude-code,B,2.1.300,D15,3,,"SECURITY.md, HackerOne bounty, 35 GitHub security advisories mostly with CVE IDs (GitHub advisory feed).",[from-official-docs] https://github.com/anthropics/claude-code/blob/main/SECURITY.md; [from-official-docs] https://github.com/anthropics/claude-code/security/advisories,rater1 | |
| claude-managed-agents,C,"beta, header managed-agents-2026-04-01 (memory stores: agent-memory-2026-07-22); toolset agent_toolset_20260401",D01,1,,"Tier C: per-tool permission policies exist, but the built-in agent toolset defaults to always_allow (bash unattended); only MCP toolsets default to always_ask.",[from-official-docs] https://platform.claude.com/docs/en/managed-agents/permission-policies,rater1 | |
| claude-managed-agents,C,"beta, header managed-agents-2026-04-01 (memory stores: agent-memory-2026-07-22); toolset agent_toolset_20260401",D02,3,,Tier C: fresh provider-managed Linux container per session with limited-networking egress control.,[from-official-docs] https://platform.claude.com/docs/en/managed-agents/environments; [from-official-docs] https://platform.claude.com/docs/en/managed-agents/self-hosted-sandboxes,rater1 | |
| claude-managed-agents,C,"beta, header managed-agents-2026-04-01 (memory stores: agent-memory-2026-07-22); toolset agent_toolset_20260401",D03,2,,"Tier C: managed allowed_hosts egress policy exists (3 on the tier-C ladder), but the vendor documents that API-created environments default to unrestricted and that access is per host, not per operation, lowering to 2.",[from-official-docs] https://platform.claude.com/docs/en/managed-agents/environments,rater1 | |
| claude-managed-agents,C,"beta, header managed-agents-2026-04-01 (memory stores: agent-memory-2026-07-22); toolset agent_toolset_20260401",D04,N/A,N/A (no local workspace or repository-level config; agents and environments are API objects).,N/A (no local workspace or repository-level config; agents and environments are API objects).,,rater1 | |
| claude-managed-agents,C,"beta, header managed-agents-2026-04-01 (memory stores: agent-memory-2026-07-22); toolset agent_toolset_20260401",D05,2,,"Tier C: MCP tools default to always_ask, MCP hosts must be in allowed_hosts, vault-injected auth documented. Not 3: no managed tool allowlist; unauthenticated fallback when no vault match.",[from-official-docs] https://platform.claude.com/docs/en/managed-agents/environments; [from-official-docs] https://platform.claude.com/docs/en/managed-agents/vaults,rater1 | |
| claude-managed-agents,C,"beta, header managed-agents-2026-04-01 (memory stores: agent-memory-2026-07-22); toolset agent_toolset_20260401",D06,N/A,N/A (tier C: skills are part of the caller's agent definition; no built-in marketplace documented).,N/A (tier C: skills are part of the caller's agent definition; no built-in marketplace documented).,[from-official-docs] https://platform.claude.com/docs/en/managed-agents/overview,rater1 | |
| claude-managed-agents,C,"beta, header managed-agents-2026-04-01 (memory stores: agent-memory-2026-07-22); toolset agent_toolset_20260401",D07,2,,Tier C: write-only vault with egress-time substitution (agent never sees the value). Not 3: vaults are workspace-scoped and tokens obtained by exchange arrive unredacted (documented gaps).,[from-official-docs] https://platform.claude.com/docs/en/managed-agents/vaults,rater1 | |
| claude-managed-agents,C,"beta, header managed-agents-2026-04-01 (memory stores: agent-memory-2026-07-22); toolset agent_toolset_20260401",D08,2,,Tier C: per-tool permission policies are server-enforced for unattended and scheduled sessions; agent toolset default always_allow keeps it below 3.,[from-official-docs] https://platform.claude.com/docs/en/managed-agents/overview; [from-official-docs] https://platform.claude.com/docs/en/managed-agents/permission-policies,rater1 | |
| claude-managed-agents,C,"beta, header managed-agents-2026-04-01 (memory stores: agent-memory-2026-07-22); toolset agent_toolset_20260401",D09,2,,"Memory stores are workspace-scoped, attached explicitly, versioned and redactable, with read_only mode. Not 3: read_write by default, no central disable documented.",[from-official-docs] https://platform.claude.com/docs/en/managed-agents/memory,rater1 | |
| claude-managed-agents,C,"beta, header managed-agents-2026-04-01 (memory stores: agent-memory-2026-07-22); toolset agent_toolset_20260401",D10,1,,Threads in a multiagent session share all vault credentials; per-subagent scoping not documented.,[from-official-docs] https://platform.claude.com/docs/en/managed-agents/vaults,rater1 | |
| claude-managed-agents,C,"beta, header managed-agents-2026-04-01 (memory stores: agent-memory-2026-07-22); toolset agent_toolset_20260401",D11,2,,"Tier C: full server-side event history with tool_use/tool_result events fetchable via API, plus webhooks.",[from-official-docs] https://platform.claude.com/docs/en/managed-agents/overview; [from-official-docs] https://platform.claude.com/docs/en/managed-agents/vaults,rater1 | |
| claude-managed-agents,C,"beta, header managed-agents-2026-04-01 (memory stores: agent-memory-2026-07-22); toolset agent_toolset_20260401",D12,1,,"Stateful by design, not ZDR eligible; deletion possible but retention period and training terms not documented here.",[from-official-docs] https://platform.claude.com/docs/en/managed-agents/overview,rater1 | |
| claude-managed-agents,C,"beta, header managed-agents-2026-04-01 (memory stores: agent-memory-2026-07-22); toolset agent_toolset_20260401",D13,1,,ant CLI installed by curl of a release tarball with no checksum step shown.,[from-official-docs] https://platform.claude.com/docs/en/managed-agents/self-hosted-sandboxes,rater1 | |
| claude-managed-agents,C,"beta, header managed-agents-2026-04-01 (memory stores: agent-memory-2026-07-22); toolset agent_toolset_20260401",D14,1,,Controls are per workspace/API key; no organisation-enforced policy above agent authors documented.,[from-official-docs] https://platform.claude.com/docs/en/managed-agents/vaults,rater1 | |
| claude-managed-agents,C,"beta, header managed-agents-2026-04-01 (memory stores: agent-memory-2026-07-22); toolset agent_toolset_20260401",D15,N/A,N/A (not documented in this evidence file; vendor programme page not collected).,N/A (not documented in this evidence file; vendor programme page not collected).,,rater1 | |
| cline,A,VS Code 4.1.24 / CLI 3.0.71,D01,1,,"CLI starts with tool auto-approve on by default; command safety is the model's requires_approval flag, not a rule list; VS Code defaults auto-approve edits, MCP and all commands (legacy fields). 3P: 'ping' treated as safe enabled DNS exfiltration.",[from-official-docs] https://github.com/cline/cline/blob/main/apps/cli/src/main.ts; [from-official-docs] https://github.com/cline/cline/blob/main/docs/cli/cli-reference.mdx; [from-official-docs] https://github.com/cline/cline/blob/main/docs/features/auto-approve.mdx; [from-official-docs] https://github.com/cline/cline/blob/main/apps/vscode/src/shared/AutoApprovalSettings.ts; [third-party-report] https://mindgard.ai/disclosures/cline-bot-ai-coding-agent-data-exfiltration-via-prompt-injection-and-dns,rater1 | |
| cline,A,VS Code 4.1.24 / CLI 3.0.71,D02,0,,No OS sandbox for agent tool execution (the 'sandbox' option is for plugin code only).,[from-official-docs] https://github.com/cline/cline/blob/main/docs/sdk/plugins.mdx,rater1 | |
| cline,A,VS Code 4.1.24 / CLI 3.0.71,D03,0,,No domain allowlist or egress control.,[from-official-docs] https://github.com/cline/cline/blob/main/docs/features/auto-approve.mdx,rater1 | |
| cline,A,VS Code 4.1.24 / CLI 3.0.71,D04,0,,Workspace hooks and JS/TS plugins are discovered from the cwd; no trust gate is documented.,[from-official-docs] https://github.com/cline/cline/blob/main/sdk/packages/shared/src/storage/paths.ts; [from-official-docs] https://github.com/cline/cline/blob/main/docs/customization/plugins.mdx,rater1 | |
| cline,A,VS Code 4.1.24 / CLI 3.0.71,D05,1,,"MCP is user-level with per-server autoApprove tool lists, but VS Code defaults useMcp auto-approval to true (blanket) and org MCP controls exist only in VS Code/JetBrains; no change detection. Patched Hub WebSocket MCP-injection GHSA noted.",[from-official-docs] https://github.com/cline/cline/blob/main/docs/mcp/mcp-overview.mdx; [from-official-docs] https://github.com/cline/cline/blob/main/docs/enterprise-solutions/configuration/remote-configuration/overview.mdx; [from-official-docs] https://github.com/cline/cline/blob/main/apps/vscode/src/shared/AutoApprovalSettings.ts,rater1 | |
| cline,A,VS Code 4.1.24 / CLI 3.0.71,D06,1,,Plugins install from npm/git/URL and run as JS/TS; no vetting or signing; sandbox opt-in.,[from-official-docs] https://github.com/cline/cline/blob/main/docs/customization/plugins.mdx; [from-official-docs] https://github.com/cline/cline/blob/main/docs/sdk/plugins.mdx,rater1 | |
| cline,A,VS Code 4.1.24 / CLI 3.0.71,D07,1,,"providers.json is plaintext 0600 in a 0700 dir (source), but no redaction or log retention is documented and .clineignore is explicitly not a read boundary.",[from-official-docs] https://github.com/cline/cline/blob/main/sdk/packages/core/src/services/storage/provider-settings-manager.ts; [from-official-docs] https://github.com/cline/cline/blob/main/docs/customization/clineignore.mdx,rater1 | |
| cline,A,VS Code 4.1.24 / CLI 3.0.71,D08,1,,Documented CI usage is cline --yolo and the CLI defaults to auto-approve; no official CI action hardening. 3P Clinejection noted.,[from-official-docs] https://github.com/cline/cline/blob/main/apps/cli/README.md; [from-official-docs] https://github.com/cline/cline/blob/main/apps/cli/src/main.ts; [third-party-report] https://adnanthekhan.com/posts/clinejection/,rater1 | |
| cline,A,VS Code 4.1.24 / CLI 3.0.71,D09,N/A,N/A (feature absent: no agent-writable memory store; rules/skills are user-authored files).,N/A (feature absent: no agent-writable memory store; rules/skills are user-authored files).,,rater1 | |
| cline,A,VS Code 4.1.24 / CLI 3.0.71,D10,1,,"Subagents are off by default in VS Code and disabled under --yolo, but permission inheritance and per-agent restrictions are not documented.",[from-official-docs] https://github.com/cline/cline/blob/main/apps/cli/src/main.ts; [from-official-docs] https://github.com/cline/cline/blob/main/apps/vscode/src/shared/storage/state-keys.ts,rater1 | |
| cline,A,VS Code 4.1.24 / CLI 3.0.71,D11,2,,Local session store with history export and OpenTelemetry export for enterprise monitoring. Policy-enforced central audit not documented.,[from-official-docs] https://github.com/cline/cline/blob/main/docs/cli/cli-reference.mdx; [from-official-docs] https://github.com/cline/cline/blob/main/docs/enterprise-solutions/monitoring/overview.mdx,rater1 | |
| cline,A,VS Code 4.1.24 / CLI 3.0.71,D12,1,,"Telemetry on by default with documented opt-out, content-free except for Cline providers; retention not documented (score-1 clause).",[from-official-docs] https://github.com/cline/cline/blob/main/docs/enterprise-solutions/monitoring/telemetry.mdx,rater1 | |
| cline,A,VS Code 4.1.24 / CLI 3.0.71,D13,2,,npm releases published with OIDC provenance since the Feb 2026 token compromise; version-pinnable registry package. Not 3: no signed binaries/SBOM.,[from-official-docs] https://github.com/cline/cline/security/advisories/GHSA-9ppg-jx86-fqw7; [from-official-docs] https://github.com/cline/cline/blob/main/.github/workflows/cli-publish.yml,rater1 | |
| cline,A,VS Code 4.1.24 / CLI 3.0.71,D14,2,,"Hosted remote config can force rules, providers, MCP and disable YOLO, but coverage is per product and the CLI/Desktop are largely excluded.",[from-official-docs] https://github.com/cline/cline/blob/main/docs/enterprise-solutions/configuration/remote-configuration/overview.mdx; [from-official-docs] https://github.com/cline/cline/blob/main/docs/enterprise-solutions/configuration/infrastructure-configuration/control-other-cline-features/yolo-mode.mdx,rater1 | |
| cline,A,VS Code 4.1.24 / CLI 3.0.71,D15,3,,SECURITY.md with Bugcrowd VDP and contact; GHSAs with CVEs including a post-incident advisory.,[from-official-docs] https://github.com/cline/cline/blob/main/SECURITY.md; [from-official-docs] https://github.com/cline/cline/security/advisories,rater1 | |
| codex-cli,A,0.162.1,D01,1,,"Default on-request inside the sandbox asks before escaping it and requirements.toml can constrain approval policy centrally, but the evidence does not show that repo files cannot set approval_policy: trusted project .codex/config.toml layers load and D04 lists the keys project config cannot override, approval_policy not among them. Deny-in-every-mode not documented.",[from-official-docs] https://github.com/openai/codex/blob/main/codex-rs/protocol/src/protocol.rs; [from-official-docs] https://developers.openai.com/codex/agent-approvals-security; [from-official-docs] https://github.com/openai/codex/blob/main/codex-rs/utils/cli/src/shared_options.rs; [from-official-docs] https://developers.openai.com/codex/enterprise/managed-configuration; [from-official-docs] https://developers.openai.com/codex/config-advanced,rater1 | |
| codex-cli,A,0.162.1,D02,2,,Seatbelt/bubblewrap sandbox on by default with writes limited to the workspace and network off. Not 3: the evidence does not show repo config cannot change sandbox_mode in a trusted project. Patched CVE-2025-59532 (model cwd as writable root) noted.,[from-official-docs] https://developers.openai.com/codex/agent-approvals-security; [from-official-docs] https://developers.openai.com/codex/sandboxing,rater1 | |
| codex-cli,A,0.162.1,D03,2,,"No client fetch tool; sandboxed shell has network off by default and an optional proxy enforces domain allow/deny. Documented gaps: web search, MCP and model traffic bypass the proxy; DNS-rebinding check best-effort. Central management of the domain list not documented.",[from-official-docs] https://developers.openai.com/codex/agent-approvals-security,rater1 | |
| codex-cli,A,0.162.1,D04,2,,"Project .codex layers (config, hooks, rules) are ignored until the project is trusted, and hooks need hash review. Not 3: env/git-helper coverage, headless flag and central folder pre-approval not documented. Patched CVE-2025-61260 noted.",[from-official-docs] https://developers.openai.com/codex/config-advanced; [from-official-docs] https://developers.openai.com/codex/hooks; [from-official-docs] https://github.com/advisories/GHSA-xrxf-jgv3-qmrm,rater1 | |
| codex-cli,A,0.162.1,D05,2,,Project MCP servers load only from trusted project config; per-tool enabled/disabled_tools; admin allowlist in requirements.toml. Not 3: no definition-change detection or pinning documented.,[from-official-docs] https://github.com/openai/codex/blob/main/codex-rs/core/config.schema.json; [from-official-docs] https://developers.openai.com/codex/agent-approvals-security; [from-official-docs] https://developers.openai.com/codex/enterprise/managed-configuration; [from-official-docs] https://developers.openai.com/codex/config-advanced,rater1 | |
| codex-cli,A,0.162.1,D06,1,,"Managed requirements can restrict marketplace sources, but no vetting, signing or scanning is documented (score-2 clause unmet).",[from-official-docs] https://developers.openai.com/codex/enterprise/managed-configuration,rater1 | |
| codex-cli,A,0.162.1,D07,1,,"Keyring store is available, but the default store and file mode are not documented, and no log retention, redaction or secret-file read deny is documented.",[from-official-docs] https://developers.openai.com/codex/config-reference,rater1 | |
| codex-cli,A,0.162.1,D08,2,,"codex exec defaults to a read-only sandbox; codex-action drops sudo by default and has allow-users/allow-bots=false. Not 3: minimal token, prompt-injection guidance and PR-head config blocking not documented.",[from-official-docs] https://developers.openai.com/codex/noninteractive; [from-official-docs] https://github.com/openai/codex-action,rater1 | |
| codex-cli,A,0.162.1,D09,2,,"Memories are off by default (feature flag), stored as local files the user can inspect. Not 3: per-project scoping, scanning and central disable not documented.",[from-official-docs] https://developers.openai.com/codex/memories,rater1 | |
| codex-cli,A,0.162.1,D10,N/A,"N/A (not documented: the only fact concerns the approvals-reviewer subagent, not delegation of tool authority).","N/A (not documented: the only fact concerns the approvals-reviewer subagent, not delegation of tool authority).",[from-official-docs] https://github.com/openai/codex/blob/main/codex-rs/core/config.schema.json,rater1 | |
| codex-cli,A,0.162.1,D11,2,,Structured rollout/history JSONL plus opt-in OpenTelemetry export.,[from-official-docs] https://developers.openai.com/codex/config-advanced,rater1 | |
| codex-cli,A,0.162.1,D12,1,,Anonymous usage metrics on by default with documented opt-out; retention and training terms not documented here (score-1 clause).,[from-official-docs] https://developers.openai.com/codex/config-advanced,rater1 | |
| codex-cli,A,0.162.1,D13,2,,Linux artifacts cosigned (sigstore); npm/Homebrew installs are version-pinnable. Not 3: SBOM and hardened publishing not documented.,[from-official-docs] https://github.com/openai/codex/blob/main/README.md,rater1 | |
| codex-cli,A,0.162.1,D14,2,,"requirements.toml (system file, MDM, cloud) constrains approval, sandbox, web search, hooks, MCP and marketplaces; telemetry not covered and not verified.",[from-official-docs] https://developers.openai.com/codex/config-basic; [from-official-docs] https://developers.openai.com/codex/enterprise/managed-configuration,rater1 | |
| codex-cli,A,0.162.1,D15,3,,SECURITY.md with Bugcrowd bounty and multiple GHSAs with CVEs.,[from-official-docs] https://github.com/openai/codex/blob/main/SECURITY.md; [from-official-docs] https://github.com/openai/codex/security/advisories/GHSA-w5fx-fh39-j5rw; [from-official-docs] https://github.com/advisories/GHSA-xrxf-jgv3-qmrm,rater1 | |
| continue,A,cn 1.5.47 / final 2.0.0,D01,2,,cn asks before Edit/Write/Bash by default; user-level allow/ask/exclude rules in ~/.continue/permissions.yaml; --auto is a CLI flag and repo files are not in the permission precedence. Not 3: --auto overrides everything (no deny in every mode) and no managed disable.,[from-official-docs] https://github.com/continuedev/continue/blob/main/docs/cli/tool-permissions.mdx,rater1 | |
| continue,A,cn 1.5.47 / final 2.0.0,D02,0,,No sandbox in CLI or IDE extensions.,[from-official-docs] https://github.com/continuedev/continue/blob/main/README.md,rater1 | |
| continue,A,cn 1.5.47 / final 2.0.0,D03,0,,Fetch tool allowed by default; no egress control.,[from-official-docs] https://github.com/continuedev/continue/blob/main/docs/cli/tool-permissions.mdx,rater1 | |
| continue,A,cn 1.5.47 / final 2.0.0,D04,0,,Hooks from .continue and .claude settings in the project and <workspace>/.continue/.env load with no trust prompt.,[from-official-docs] https://github.com/continuedev/continue/blob/main/extensions/cli/src/hooks/hookConfig.ts; [from-official-docs] https://github.com/continuedev/continue/blob/main/extensions/cli/src/CLIPlatformClient.ts,rater1 | |
| continue,A,cn 1.5.47 / final 2.0.0,D05,0,,Workspace .continue/mcpServers files are picked up automatically with no approval or allow/deny controls.,[from-official-docs] https://github.com/continuedev/continue/blob/main/docs/customize/deep-dives/mcp.mdx,rater1 | |
| continue,A,cn 1.5.47 / final 2.0.0,D06,N/A,N/A (not documented: no fact about extension vetting; the hub was removed in 2.0.0).,N/A (not documented: no fact about extension vetting; the hub was removed in 2.0.0).,[from-official-docs] https://github.com/continuedev/continue/blob/main/README.md,rater1 | |
| continue,A,cn 1.5.47 / final 2.0.0,D07,1,,Secrets resolved from env and plaintext .env files including the workspace; no redaction or read protection.,[from-official-docs] https://github.com/continuedev/continue/blob/main/extensions/cli/spec/mcp.md,rater1 | |
| continue,A,cn 1.5.47 / final 2.0.0,D08,1,,"Headless cn -p excludes ask-class tools unless --allow is given (safe), but the official general-review action's token permissions and actor checks are not documented (score-2 clause unmet).",[from-official-docs] https://github.com/continuedev/continue/blob/main/docs/cli/headless-mode.mdx; [from-official-docs] https://github.com/continuedev/continue/blob/main/actions/README.md,rater1 | |
| continue,A,cn 1.5.47 / final 2.0.0,D09,N/A,N/A (feature absent: no persistent agent memory).,N/A (feature absent: no persistent agent memory).,,rater1 | |
| continue,A,cn 1.5.47 / final 2.0.0,D10,1,,Subagents exist (hook events) with undocumented permission inheritance.,[from-official-docs] https://github.com/continuedev/continue/blob/main/extensions/cli/src/hooks/types.ts,rater1 | |
| continue,A,cn 1.5.47 / final 2.0.0,D11,1,,Local session JSON files; an export path is not documented in the eligible facts.,[from-official-docs] https://github.com/continuedev/continue/blob/main/extensions/cli/src/session.ts,rater1 | |
| continue,A,cn 1.5.47 / final 2.0.0,D12,N/A,N/A (no vendor service: final 2.0.0 removed anonymous telemetry and hub authentication).,N/A (no vendor service: final 2.0.0 removed anonymous telemetry and hub authentication).,[from-official-docs] https://github.com/continuedev/continue/blob/main/README.md,rater1 | |
| continue,A,cn 1.5.47 / final 2.0.0,D13,1,,npm registry package without documented provenance or signatures; project read-only.,[from-official-docs] https://github.com/continuedev/continue/blob/main/README.md; [from-official-docs] https://registry.npmjs.org/@continuedev/cli,rater1 | |
| continue,A,cn 1.5.47 / final 2.0.0,D14,N/A,N/A (not documented: no managed-settings facts collected; hub org config removed).,N/A (not documented: no managed-settings facts collected; hub org config removed).,,rater1 | |
| continue,A,cn 1.5.47 / final 2.0.0,D15,1,,SECURITY.md exists; no published advisories.,[from-official-docs] https://github.com/continuedev/continue/blob/main/SECURITY.md,rater1 | |
| copilot-studio,C,None,D01,1,,"Tier C: centrally enforced data policies can block connectors/HTTP in real time and tools use end-user credentials, but no per-action approval primitive is documented.",[from-official-docs] https://learn.microsoft.com/en-us/microsoft-copilot-studio/security-scan; [from-official-docs] https://learn.microsoft.com/en-us/microsoft-copilot-studio/admin-data-loss-prevention,rater1 | |
| copilot-studio,C,None,D02,2,,"Tier C: each harness task runs in a platform sandbox by default; mechanism and egress coupling undocumented, so not 3.",[from-official-docs] https://learn.microsoft.com/en-us/microsoft-copilot-studio/harnesses-overview,rater1 | |
| copilot-studio,C,None,D03,3,,Tier C: managed endpoint filtering for HTTP/knowledge connectors and Entra network egress controls.,[from-official-docs] https://learn.microsoft.com/en-us/microsoft-copilot-studio/security-and-governance,rater1 | |
| copilot-studio,C,None,D04,N/A,N/A (SaaS; no local workspace or repository-level config).,N/A (SaaS; no local workspace or repository-level config).,,rater1 | |
| copilot-studio,C,None,D05,2,,"Tier C: data policies act as a platform allowlist for MCP tools, but tool definitions change dynamically without re-approval (documented gap).",[from-official-docs] https://learn.microsoft.com/en-us/microsoft-copilot-studio/admin-data-loss-prevention; [from-official-docs] https://learn.microsoft.com/en-us/microsoft-copilot-studio/agent-extend-action-mcp; [from-official-docs] https://www.microsoft.com/microsoft-copilot/blog/copilot-studio/new-and-improved-github-copilot-harness-agent-skills-and-richer-context/,rater1 | |
| copilot-studio,C,None,D06,1,,Makers can upload skills with scripts shared across agents; no vetting or signing documented.,[from-official-docs] https://learn.microsoft.com/en-us/microsoft-copilot-studio/agents-experience/skills-overview,rater1 | |
| copilot-studio,C,None,D07,2,,"Tier C: end-user credentials by default, CMK and Lockbox; 3P CoPhish shows maker-configured token forwarding (admin-mitigable).",[from-official-docs] https://learn.microsoft.com/en-us/microsoft-copilot-studio/security-and-governance; [third-party-report] https://securitylabs.datadoghq.com/articles/cophish-using-microsoft-copilot-studio-as-a-wrapper/,rater1 | |
| copilot-studio,C,None,D08,N/A,N/A (no facts on unattended-run enforcement; SaaS with no headless CLI).,N/A (no facts on unattended-run enforcement; SaaS with no headless CLI).,,rater1 | |
| copilot-studio,C,None,D09,1,,"Memory exists (preview); scope, visibility and off switch not documented.",[from-official-docs] https://www.microsoft.com/microsoft-copilot/blog/copilot-studio/new-and-improved-github-copilot-harness-agent-skills-and-richer-context/,rater1 | |
| copilot-studio,C,None,D10,1,,"Connected agents (preview) each get an Entra Agent ID, but scoping between agents is not documented.",[from-official-docs] https://www.microsoft.com/microsoft-copilot/blog/copilot-studio/new-and-improved-github-copilot-harness-agent-skills-and-richer-context/,rater1 | |
| copilot-studio,C,None,D11,2,,"Purview audit of admin and user events that cannot be disabled (3 on the criteria), lowered to 2 by a third-party report of unlogged admin events and inconsistent events after the fix.",[from-official-docs] https://learn.microsoft.com/en-us/microsoft-copilot-studio/admin-logging-copilot-studio; [third-party-report] https://securitylabs.datadoghq.com/articles/copilot-studio-logging-gaps/,rater1 | |
| copilot-studio,C,None,D12,1,,"Zero-day retention and residency options documented, but the training-use default is not documented (score-2 clause unmet).",[from-official-docs] https://learn.microsoft.com/en-us/microsoft-copilot-studio/security-and-governance,rater1 | |
| copilot-studio,C,None,D13,N/A,N/A (SaaS; no distributable client in scope).,N/A (SaaS; no distributable client in scope).,,rater1 | |
| copilot-studio,C,None,D14,2,,"Tenant data policies, environment routing and publishing controls that makers cannot override; telemetry/sandbox coverage not documented and not verified.",[from-official-docs] https://learn.microsoft.com/en-us/microsoft-copilot-studio/security-and-governance,rater1 | |
| copilot-studio,C,None,D15,2,,MSRC handles disclosure and issues CVEs for the service; bounty applicability not documented.,[from-official-docs] https://cveawg.mitre.org/api/cve/CVE-2026-21520,rater1 | |
| crewai,C,1.15.28,D01,N/A,N/A (not documented: the absence of a per-tool approval primitive is only an unverified finding).,N/A (not documented: the absence of a per-tool approval primitive is only an unverified finding).,[from-official-docs] https://github.com/crewAIInc/crewAI/blob/main/lib/crewai/src/crewai/agent/core.py,rater1 | |
| crewai,C,1.15.28,D02,0,,Tier C: no built-in sandbox after CodeInterpreterTool removal; external services are application responsibility. Weak-fallback CVE-2026-2275 noted.,[from-official-docs] https://github.com/crewAIInc/crewAI/blob/main/lib/crewai/src/crewai/agent/core.py; [from-official-docs] https://kb.cert.org/vuls/id/221883,rater1 | |
| crewai,C,1.15.28,D03,0,,No egress controls in core; SSRF CVE-2026-2286 from insecure defaults.,[from-official-docs] https://kb.cert.org/vuls/id/221883,rater1 | |
| crewai,C,1.15.28,D04,N/A,N/A (library; only an unverified fact).,N/A (library; only an unverified fact).,,rater1 | |
| crewai,C,1.15.28,D05,1,,"Tier C: static tool_filter only; no approval hook, allowlist or change detection; docs rely on trusting servers.",[from-official-docs] https://github.com/crewAIInc/crewAI/blob/main/docs/edge/en/mcp/security.mdx,rater1 | |
| crewai,C,1.15.28,D06,N/A,N/A (not documented: only unverified facts).,N/A (not documented: only unverified facts).,,rater1 | |
| crewai,C,1.15.28,D07,1,,Tier C: keys from environment/.env.,[from-official-docs] https://github.com/crewAIInc/crewAI/blob/main/docs/edge/en/telemetry.mdx,rater1 | |
| crewai,C,1.15.28,D08,N/A,N/A (not documented: only unverified facts).,N/A (not documented: only unverified facts).,,rater1 | |
| crewai,C,1.15.28,D09,1,,"Agent-written LanceDB memory persisted per project name in user data dir; no confirmation, review or off switch documented.",[from-official-docs] https://github.com/crewAIInc/crewAI/blob/main/lib/crewai-core/src/crewai_core/paths.py,rater1 | |
| crewai,C,1.15.28,D10,N/A,N/A (not documented: delegation facts are unverified).,N/A (not documented: delegation facts are unverified).,,rater1 | |
| crewai,C,1.15.28,D11,2,,"Tier C: event bus and tracing one setting away (CREWAI_TRACING_ENABLED, OTel).",[from-official-docs] https://github.com/crewAIInc/crewAI/blob/main/lib/crewai/src/crewai/events/listeners/tracing/utils.py,rater1 | |
| crewai,C,1.15.28,D12,1,,Telemetry on by default (metadata incl. agent roles and tool names) with documented opt-out; retention not documented.,[from-official-docs] https://github.com/crewAIInc/crewAI/blob/main/lib/crewai/src/crewai/telemetry/constants.py; [from-official-docs] https://github.com/crewAIInc/crewAI/blob/main/docs/edge/en/telemetry.mdx,rater1 | |
| crewai,C,1.15.28,D13,N/A,N/A (not documented: only unverified facts).,N/A (not documented: only unverified facts).,,rater1 | |
| crewai,C,1.15.28,D14,N/A,N/A (not documented: only unverified facts; enterprise AMP out of scope).,N/A (not documented: only unverified facts; enterprise AMP out of scope).,,rater1 | |
| crewai,C,1.15.28,D15,0,,"No SECURITY.md, no GitHub advisories; CVEs came through CERT/CC with partial remediation; no vendor contact or policy found.",[from-official-docs] https://kb.cert.org/vuls/id/221883,rater1 | |
| crush,A,0.98.1,D01,1,,"Prompts by default with allowed_tools, but project-level crush.json PreToolUse hooks can auto-approve (repo-controlled approval), safe-command prefixes skip prompts and 'crush run' approves everything; no deny rules.",[from-official-docs] https://github.com/charmbracelet/crush/blob/main/internal/agent/tools/safe.go; [from-official-docs] https://github.com/charmbracelet/crush/blob/main/internal/app/app.go; [from-official-docs] https://github.com/charmbracelet/crush/blob/main/docs/hooks/README.md,rater1 | |
| crush,A,0.98.1,D02,N/A,N/A (not documented: the absence of a sandbox is only an unverified finding).,N/A (not documented: the absence of a sandbox is only an unverified finding).,,rater1 | |
| crush,A,0.98.1,D03,N/A,N/A (not documented: no egress facts collected).,N/A (not documented: no egress facts collected).,,rater1 | |
| crush,A,0.98.1,D04,0,,Project crushrc (Bash) and $(...) in crush.json execute at config load with no trust gate (verified); README calls them trusted code.,[from-official-docs] https://github.com/charmbracelet/crush/blob/main/internal/shellconfig/load.go; [from-official-docs] https://github.com/charmbracelet/crush/blob/main/internal/config/load.go,rater1 | |
| crush,A,0.98.1,D05,1,,"MCP servers have per-tool enabled/disabled lists, but project-defined servers start with no approval step and there is no managed policy.",[from-official-docs] https://github.com/charmbracelet/crush/blob/main/schema.json,rater1 | |
| crush,A,0.98.1,D06,1,,Skills auto-discovered from project dirs including other harnesses' folders; no vetting.,[from-official-docs] https://github.com/charmbracelet/crush/blob/main/internal/config/config.go,rater1 | |
| crush,A,0.98.1,D07,1,,Keys in config/env; state and logs in project .crush; no redaction or secret-file protection documented.,[from-official-docs] https://github.com/charmbracelet/crush/blob/main/README.md,rater1 | |
| crush,A,0.98.1,D08,0,,Non-interactive 'crush run' auto-approves every permission request (verified).,[from-official-docs] https://github.com/charmbracelet/crush/blob/main/internal/app/app.go,rater1 | |
| crush,A,0.98.1,D09,N/A,N/A (feature absent: no agent-written memory beyond user-authored context files).,N/A (feature absent: no agent-written memory beyond user-authored context files).,[from-official-docs] https://github.com/charmbracelet/crush/blob/main/internal/config/config.go,rater1 | |
| crush,A,0.98.1,D10,N/A,N/A (not documented: no sub-agent facts collected).,N/A (not documented: no sub-agent facts collected).,,rater1 | |
| crush,A,0.98.1,D11,1,,Local log file and SQLite session DB; no tool-call export.,[from-official-docs] https://github.com/charmbracelet/crush/blob/main/README.md,rater1 | |
| crush,A,0.98.1,D12,1,,Pseudonymous metrics on by default with documented opt-out and no prompt collection; retention not documented.,[from-official-docs] https://github.com/charmbracelet/crush/blob/main/README.md,rater1 | |
| crush,A,0.98.1,D13,2,,"cosign signatures, checksums and SBOMs published (verified in release config). Not 3: provenance attestation, update-channel control and hardened publishing not documented.",[from-official-docs] https://github.com/charmbracelet/crush/blob/main/.goreleaser.yml,rater1 | |
| crush,A,0.98.1,D14,0,,"/etc/crush/crush.json has the lowest precedence, so nothing is enforceable (verified).",[from-official-docs] https://github.com/charmbracelet/crush/blob/main/internal/config/load.go,rater1 | |
| crush,A,0.98.1,D15,1,,Org-level SECURITY.md email; no advisories or bounty.,[from-official-docs] https://github.com/charmbracelet/.github/blob/main/SECURITY.md,rater1 | |
| cursor,B,Cursor 3.23 docs (2026-10-01); CLI 2026.08.11-e8db854 (third-party),D01,1,,"Default IDE mode is classifier-based Auto-review (no user prompt) and 'Ask Every Time' is deprecated, so the default runs shell without asking the user (score-1 clause), as for claude-code. Deny beats allow in the CLI. Patched allowlist-bypass CVEs noted.",[from-official-docs] https://cursor.com/docs/agent/security/run-modes; [from-official-docs] https://cursor.com/docs/cli/reference/configuration; [from-official-docs] https://github.com/cursor/cursor/security/advisories,rater1 | |
| cursor,B,Cursor 3.23 docs (2026-10-01); CLI 2026.08.11-e8db854 (third-party),D02,1,,"Seatbelt/Landlock sandbox is used for shell in the default Auto-review mode, but the workspace-only write scope and Windows coverage are not documented (score-2 clause unmet). Patched escape CVEs noted.",[from-official-docs] https://cursor.com/docs/agent/security/run-modes; [from-official-docs] https://github.com/cursor/cursor/security/advisories,rater1 | |
| cursor,B,Cursor 3.23 docs (2026-10-01); CLI 2026.08.11-e8db854 (third-party),D03,1,,Sandbox network allowlist is on by default but its default includes ~100 package domains (upload-capable registries); fetch-tool default not documented.,[from-official-docs] https://cursor.com/docs/agent/security/run-modes,rater1 | |
| cursor,B,Cursor 3.23 docs (2026-10-01); CLI 2026.08.11-e8db854 (third-party),D04,1,,"Project hooks need a trusted workspace, but VS Code Workspace Trust ships disabled (3P: tasks.json runOn folderOpen runs); patched CVE-2026-48124 hook bypass.",[from-official-docs] https://cursor.com/docs/agent/hooks; [third-party-report] https://devops.com/oasis-security-identifies-security-weakness-in-cursor-ai-coding-tool/; [from-official-docs] https://github.com/cursor/cursor/security/advisories/GHSA-pc9j-3qc2-95wv,rater1 | |
| cursor,B,Cursor 3.23 docs (2026-10-01); CLI 2026.08.11-e8db854 (third-party),D05,1,,"Tools prompt and an admin MCP allowlist exists, but there is no separate approval for project-level servers; patched MCPoison re-approval CVE.",[from-official-docs] https://cursor.com/docs/context/mcp; [from-official-docs] https://github.com/cursor/cursor/security/advisories,rater1 | |
| cursor,B,Cursor 3.23 docs (2026-10-01); CLI 2026.08.11-e8db854 (third-party),D06,1,,"Team marketplace and MDM extension allowlisting exist, but no vetting or signing is documented.",[from-official-docs] https://cursor.com/docs/enterprise,rater1 | |
| cursor,B,Cursor 3.23 docs (2026-10-01); CLI 2026.08.11-e8db854 (third-party),D07,1,,Credential storage location not documented; .cursorignore protection had several patched bypass CVEs.,[from-official-docs] https://github.com/cursor/cursor/security/advisories,rater1 | |
| cursor,B,Cursor 3.23 docs (2026-10-01); CLI 2026.08.11-e8db854 (third-party),D08,1,,"Print mode needs --force to write and allow/deny apply, but there is no official CI integration or untrusted-input guidance.",[from-official-docs] https://cursor.com/docs/cli/headless; [from-official-docs] https://github.com/cursor/cursor/security/advisories/GHSA-v64q-396f-7m79,rater1 | |
| cursor,B,Cursor 3.23 docs (2026-10-01); CLI 2026.08.11-e8db854 (third-party),D09,N/A,N/A (not documented: memories not covered in the pages read; rules are user-authored).,N/A (not documented: memories not covered in the pages read; rules are user-authored).,[from-official-docs] https://cursor.com/docs/context/rules,rater1 | |
| cursor,B,Cursor 3.23 docs (2026-10-01); CLI 2026.08.11-e8db854 (third-party),D10,1,,Subagent hook events exist; isolation and inheritance undocumented.,[from-official-docs] https://cursor.com/docs/agent/hooks,rater1 | |
| cursor,B,Cursor 3.23 docs (2026-10-01); CLI 2026.08.11-e8db854 (third-party),D11,2,,"Enterprise audit logs with SIEM streaming and hooks for tool-call logging; local transcripts and policy enforcement not documented, so not 3.",[from-official-docs] https://cursor.com/docs/enterprise,rater1 | |
| cursor,B,Cursor 3.23 docs (2026-10-01); CLI 2026.08.11-e8db854 (third-party),D12,1,,"Privacy Mode (no training, ZDR) is enforceable org-wide, but the default for new users and the retention period are not documented.",[from-official-docs] https://cursor.com/data-use; [from-official-docs] https://cursor.com/docs/enterprise,rater1 | |
| cursor,B,Cursor 3.23 docs (2026-10-01); CLI 2026.08.11-e8db854 (third-party),D13,1,,curl|bash installer with no documented checksum or signature.,[from-official-docs] https://cursor.com/docs/cli/headless,rater1 | |
| cursor,B,Cursor 3.23 docs (2026-10-01); CLI 2026.08.11-e8db854 (third-party),D14,2,,"Team/Enterprise settings override local and cover sandbox, auto-run, network, MCP allowlist, extensions and privacy mode. Not 3: 'documented and verified' unmet for a docs-only harness.",[from-official-docs] https://cursor.com/docs/enterprise; [from-official-docs] https://cursor.com/docs/agent/hooks; [from-official-docs] https://cursor.com/docs/context/mcp,rater1 | |
| cursor,B,Cursor 3.23 docs (2026-10-01); CLI 2026.08.11-e8db854 (third-party),D15,2,,Security contact and 29 CVE-numbered GitHub advisories; no bug bounty or safe harbour.,[from-official-docs] https://cursor.com/security; [from-official-docs] https://github.com/cursor/cursor/security/advisories,rater1 | |
| deepagents,C,deepagents 0.7.25 / deepagents-code 0.1.85,D01,1,,Tier C: interrupt_on HITL and filesystem permission rules exist but default to none; dcode headless runs non-shell tools unattended. Patched always-allow bypass GHSA noted.,[from-official-docs] https://github.com/langchain-ai/deepagents/blob/main/libs/deepagents/deepagents/graph.py; [from-official-docs] https://github.com/langchain-ai/deepagents/blob/main/libs/code/THREAT_MODEL.md; [from-official-docs] https://github.com/langchain-ai/deepagents/security/advisories/GHSA-459r-276m-w72x,rater1 | |
| deepagents,C,deepagents 0.7.25 / deepagents-code 0.1.85,D02,1,,Tier C: isolation only through opt-in remote sandbox providers; local shell backend unisolated.,[from-official-docs] https://github.com/langchain-ai/deepagents/blob/main/libs/deepagents/deepagents/backends/filesystem.py,rater1 | |
| deepagents,C,deepagents 0.7.25 / deepagents-code 0.1.85,D03,N/A,N/A (not documented: no egress facts collected).,N/A (not documented: no egress facts collected).,,rater1 | |
| deepagents,C,deepagents 0.7.25 / deepagents-code 0.1.85,D04,2,,"dcode gates project hooks behind workspace trust and project MCP behind a trust list; instruction files are read before approval by design. Not 3: env/plugins, headless flag and central folder policy not documented.",[from-official-docs] https://github.com/langchain-ai/deepagents/blob/main/libs/code/README.md,rater1 | |
| deepagents,C,deepagents 0.7.25 / deepagents-code 0.1.85,D05,1,,"Tier C: project .mcp.json needs approval (trust list), but per-tool permissions and auth handling are not documented, and an env var enables all project servers.",[from-official-docs] https://github.com/langchain-ai/deepagents/blob/main/libs/code/THREAT_MODEL.md,rater1 | |
| deepagents,C,deepagents 0.7.25 / deepagents-code 0.1.85,D06,N/A,N/A (not documented: no extension facts collected).,N/A (not documented: no extension facts collected).,,rater1 | |
| deepagents,C,deepagents 0.7.25 / deepagents-code 0.1.85,D07,1,,Tier C: API keys from the process environment (self-reported never written to disk).,[from-official-docs] https://github.com/langchain-ai/deepagents/blob/main/libs/code/THREAT_MODEL.md,rater1 | |
| deepagents,C,deepagents 0.7.25 / deepagents-code 0.1.85,D08,1,,Headless -n runs non-shell tools (including file mutation) without approval; shell is off unless allow-listed; the GitHub Action does not forward --auto-approve.,[from-official-docs] https://github.com/langchain-ai/deepagents/blob/main/libs/code/THREAT_MODEL.md,rater1 | |
| deepagents,C,deepagents 0.7.25 / deepagents-code 0.1.85,D09,1,,"Persistent dcode memory and SDK memory/store exist; scoping, review and off switch not documented.",[from-official-docs] https://github.com/langchain-ai/deepagents/blob/main/libs/deepagents/deepagents/graph.py,rater1 | |
| deepagents,C,deepagents 0.7.25 / deepagents-code 0.1.85,D10,N/A,N/A (not documented: no sub-agent isolation facts collected).,N/A (not documented: no sub-agent isolation facts collected).,,rater1 | |
| deepagents,C,deepagents 0.7.25 / deepagents-code 0.1.85,D11,N/A,N/A (not documented: no audit-logging facts collected).,N/A (not documented: no audit-logging facts collected).,,rater1 | |
| deepagents,C,deepagents 0.7.25 / deepagents-code 0.1.85,D12,N/A,N/A (not documented: no telemetry facts collected).,N/A (not documented: no telemetry facts collected).,,rater1 | |
| deepagents,C,deepagents 0.7.25 / deepagents-code 0.1.85,D13,N/A,N/A (not documented: no distribution facts collected).,N/A (not documented: no distribution facts collected).,,rater1 | |
| deepagents,C,deepagents 0.7.25 / deepagents-code 0.1.85,D14,2,,"managed_config.toml enforces keys such as startup.mode, shell.allow_list and allowed models; MCP/plugin/telemetry coverage not documented.",[from-official-docs] https://github.com/langchain-ai/deepagents/blob/main/libs/code/THREAT_MODEL.md,rater1 | |
| deepagents,C,deepagents 0.7.25 / deepagents-code 0.1.85,D15,1,,"A GHSA was published with a fix, but no SECURITY.md or policy is documented (3P: root SECURITY.md 404), so score 2 is not met.",[from-official-docs] https://github.com/langchain-ai/deepagents/security/advisories/GHSA-459r-276m-w72x; [third-party-report] https://github.com/langchain-ai/deepagents/blob/main/SECURITY.md,rater1 | |
| deepseek-harness,A,0.2.1-alpha.3,D01,1,,"Default workspace-write sandbox with 'ask' approvals that fail closed, but approval is coarse (preset on/off) with no per-command allow/deny rules (score-1 clause).",[from-official-docs] https://github.com/deepseek-ai/deepseek-harness/blob/master/packages/bundle/base/cordis.patch.yml; [from-official-docs] https://github.com/deepseek-ai/deepseek-harness/blob/master/packages/interaction/user-approval/README.md,rater1 | |
| deepseek-harness,A,0.2.1-alpha.3,D02,2,,bwrap/Landlock/Seatbelt sandbox on by default in workspace-write and fails closed. Not 3: network not confined (file effects only); no central enforcement.,[from-official-docs] https://github.com/deepseek-ai/deepseek-harness/blob/master/packages/sandbox/sandbox-local/README.md; [from-official-docs] https://github.com/deepseek-ai/deepseek-harness/blob/master/SAFETY.md,rater1 | |
| deepseek-harness,A,0.2.1-alpha.3,D03,0,,Sandbox policy has no network vocabulary and web_fetch/web_search have no allowlist.,[from-official-docs] https://github.com/deepseek-ai/deepseek-harness/blob/master/packages/sandbox/sandbox-policy/README.md; [from-official-docs] https://github.com/deepseek-ai/deepseek-harness/blob/master/docs/tool-catalog.md,rater1 | |
| deepseek-harness,A,0.2.1-alpha.3,D04,2,,"Executable config (plugins, MCP, policies) is read only from $DSH_HOME, never from the repo, which meets the intent of a gate for hooks/MCP/settings; repo AGENTS.md/skills and project .env (credential precedence) load silently, so not 3.",[from-official-docs] https://github.com/deepseek-ai/deepseek-harness/blob/master/apps/cli/README.md; [from-official-docs] https://github.com/deepseek-ai/deepseek-harness/blob/master/packages/context/agent-instructions/README.md; [from-official-docs] https://github.com/deepseek-ai/deepseek-harness/blob/master/packages/credentials/credentials-local/README.md,rater1 | |
| deepseek-harness,A,0.2.1-alpha.3,D05,1,,"No servers by default and only user-profile servers exist (a repo cannot add one), but there are no allow/deny lists, per-tool permissions or change detection.",[from-official-docs] https://github.com/deepseek-ai/deepseek-harness/blob/master/packages/mcp/mcp-client/README.md,rater1 | |
| deepseek-harness,A,0.2.1-alpha.3,D06,1,,Plugins are npm packages run in-process via pnpm with peer-range checks; no signing or security vetting; SAFETY.md asks users to review.,[from-official-docs] https://github.com/deepseek-ai/deepseek-harness/blob/master/apps/cli/README.md; [from-official-docs] https://github.com/deepseek-ai/deepseek-harness/blob/master/SAFETY.md; [third-party-report] https://github.com/imsai-sh/awesome-deepseek-harness-plugins,rater1 | |
| deepseek-harness,A,0.2.1-alpha.3,D07,1,,Plaintext $DSH_HOME/.credentials.yaml that the docs say cannot be isolated from the agent; web UI redacts keys.,[from-official-docs] https://github.com/deepseek-ai/deepseek-harness/blob/master/packages/credentials/credentials-local/README.md; [from-official-docs] https://github.com/deepseek-ai/deepseek-harness/blob/master/docs/user/guide/providers.md,rater1 | |
| deepseek-harness,A,0.2.1-alpha.3,D08,1,,"Headless with 'ask' and no answerer fails closed and 'never' is documented for CI, but there is no official CI integration (score-2 clause unmet).",[from-official-docs] https://github.com/deepseek-ai/deepseek-harness/blob/master/apps/cli/reference/README.md; [from-official-docs] https://github.com/deepseek-ai/deepseek-harness/blob/master/packages/interaction/user-approval/README.md,rater1 | |
| deepseek-harness,A,0.2.1-alpha.3,D09,N/A,N/A (feature absent: memory only via default-off third-party MCP overlays).,N/A (feature absent: memory only via default-off third-party MCP overlays).,[from-official-docs] https://github.com/deepseek-ai/deepseek-harness/blob/master/docs/user/guide/mcp-memory.md,rater1 | |
| deepseek-harness,A,0.2.1-alpha.3,D10,2,,"Delegated children pin approval policy 'never', so approval-gated actions cannot escalate through delegation. Not 3: no per-subagent sandbox/worktree or delegated-action audit documented.",[from-official-docs] https://github.com/deepseek-ai/deepseek-harness/blob/master/packages/subagent; [from-official-docs] https://github.com/deepseek-ai/deepseek-harness/blob/master/packages/interaction/permission-presets/README.md,rater1 | |
| deepseek-harness,A,0.2.1-alpha.3,D11,2,,"Approval requests/outcomes in a session audit log, JSONL sessions, OTel export controllable by env.",[from-official-docs] https://github.com/deepseek-ai/deepseek-harness/blob/master/packages/interaction/user-approval/README.md; [from-official-docs] https://github.com/deepseek-ai/deepseek-harness/blob/master/packages/bundle/base/cordis.patch.yml; [from-official-docs] https://github.com/deepseek-ai/deepseek-harness/blob/master/apps/cli/reference/README.md,rater1 | |
| deepseek-harness,A,0.2.1-alpha.3,D12,1,,Session-log contributor sends unaccepted session content with DeepSeek requests by default (opt-out available); retention/training terms not documented.,[from-official-docs] https://github.com/deepseek-ai/deepseek-harness/blob/master/apps/cli/reference/README.md,rater1 | |
| deepseek-harness,A,0.2.1-alpha.3,D13,1,,"npm registry distribution with no documented provenance, signature or checksum.",[from-official-docs] https://registry.npmjs.org/@deepseek-ai/dsh,rater1 | |
| deepseek-harness,A,0.2.1-alpha.3,D14,0,,"Config layers are bundle, profile, home and --patch only; no administrator-enforced layer (open-source dev preview).",[from-official-docs] https://github.com/deepseek-ai/deepseek-harness/blob/master/apps/cli/README.md,rater1 | |
| deepseek-harness,A,0.2.1-alpha.3,D15,0,,No SECURITY.md or disclosure channel; 0 advisories.,[from-official-docs] https://github.com/deepseek-ai/deepseek-harness/blob/master/SAFETY.md,rater1 | |
| devin,B,Devin CLI v3000.11.3 (2026-09-22); docs-level,D01,2,,"Default normal mode with deny>ask>allow; project config is limited to permissions, read_config_from and hooks, so modes (bypass/yolo) cannot be set from the repo; org deny/ask cannot be overridden by any mode. Not 3: a central switch to disable bypass mode and user-level deny in bypass are not documented. Patched deny-rule bypasses noted.",[from-official-docs] https://docs.devin.ai/cli/reference/permissions.md; [from-official-docs] https://docs.devin.ai/cli/changelog/stable.md; [from-official-docs] https://docs.devin.ai/cli/reference/configuration/config-file.md,rater1 | |
| devin,B,Devin CLI v3000.11.3 (2026-09-22); docs-level,D02,1,,Seatbelt/bwrap sandbox is off by default locally (admin can require it); research preview; Windows unsupported.,[from-official-docs] https://docs.devin.ai/cli/sandbox.md,rater1 | |
| devin,B,Devin CLI v3000.11.3 (2026-09-22); docs-level,D03,1,,Domain allow/deny exists only inside the opt-in sandbox; network full by default.,[from-official-docs] https://docs.devin.ai/cli/sandbox.md,rater1 | |
| devin,B,Devin CLI v3000.11.3 (2026-09-22); docs-level,D04,2,,Workspace trust is on by default and -p fails in untrusted dirs (headless needs an explicit flag); imported Cursor/Claude/Windsurf MCP and hooks follow trust. Not 3: env/plugins/git helpers and central folder approval not documented.,[from-official-docs] https://docs.devin.ai/cli/reference/commands.md; [from-official-docs] https://docs.devin.ai/cli/reference/configuration/read-config-from.md; [from-official-docs] https://docs.devin.ai/cli/reference/configuration/config-file.md,rater1 | |
| devin,B,Devin CLI v3000.11.3 (2026-09-22); docs-level,D05,1,,"Tool calls prompt with mcp__ rules and team MCP settings, but project servers load with no approval step.",[from-official-docs] https://docs.devin.ai/cli/extensibility/mcp/configuration.md,rater1 | |
| devin,B,Devin CLI v3000.11.3 (2026-09-22); docs-level,D06,1,,Plugins and skills with an org switch to disable plugins; no vetting or signing documented.,[from-official-docs] https://docs.devin.ai/cli/changelog/stable.md,rater1 | |
| devin,B,Devin CLI v3000.11.3 (2026-09-22); docs-level,D07,1,,OAuth tokens stored locally with backend and mode undocumented; no redaction documented.,[from-official-docs] https://docs.devin.ai/cli/extensibility/mcp/configuration.md,rater1 | |
| devin,B,Devin CLI v3000.11.3 (2026-09-22); docs-level,D08,1,,"Headless fails closed in untrusted dirs, but there is no official CI integration and CI must disable workspace trust.",[from-official-docs] https://docs.devin.ai/cli/reference/commands.md,rater1 | |
| devin,B,Devin CLI v3000.11.3 (2026-09-22); docs-level,D09,N/A,N/A (feature absent: Devin Local persists no memories).,N/A (feature absent: Devin Local persists no memories).,[from-official-docs] https://docs.devin.ai/desktop/devin-local.md,rater1 | |
| devin,B,Devin CLI v3000.11.3 (2026-09-22); docs-level,D10,1,,Session approvals cover subagents (inherited); per-agent restriction not documented.,[from-official-docs] https://docs.devin.ai/desktop/devin-local.md,rater1 | |
| devin,B,Devin CLI v3000.11.3 (2026-09-22); docs-level,D11,2,,Structured ATIF transcript export plus lifecycle hooks and admin analytics; policy enforcement not documented.,[from-official-docs] https://docs.devin.ai/cli/reference/commands.md,rater1 | |
| devin,B,Devin CLI v3000.11.3 (2026-09-22); docs-level,D12,1,,Training on by default for non-enterprise (paid users can opt out); retention 'for the duration of the customer relationship' is not bounded.,[from-official-docs] https://docs.devin.ai/admin/security.md,rater1 | |
| devin,B,Devin CLI v3000.11.3 (2026-09-22); docs-level,D13,1,,curl|bash and brew cask with no documented checksum or signature.,[from-official-docs] https://docs.devin.ai/work-with-devin/devin-cli.md,rater1 | |
| devin,B,Devin CLI v3000.11.3 (2026-09-22); docs-level,D14,2,,"Root-owned system.json and Team Settings enforce permissions, sandbox and plugin switch; MCP/telemetry coverage not documented and not verified.",[from-official-docs] https://docs.devin.ai/cli/enterprise/system-config.md; [from-official-docs] https://docs.devin.ai/cli/reference/permissions.md; [from-official-docs] https://docs.devin.ai/cli/changelog/stable.md,rater1 | |
| devin,B,Devin CLI v3000.11.3 (2026-09-22); docs-level,D15,1,,"Security contact only; no advisories, bounty or feed.",[from-official-docs] https://docs.devin.ai/admin/security.md; [third-party-report] https://vibegraveyard.ai/story/devin-ai-prompt-injection-no-protection/,rater1 | |
| factory-droid,B,0.242.0,D01,1,,"Block rules hold in every mode (even --skip-permissions-unsafe), droid exec is read-only by default and org maxAutonomyLevel caps autonomy, but the default interactive autonomy level is not stated in the facts, so 'default mode asks before shell' cannot be confirmed (would be 3 if Off is the default).",[from-official-docs] https://docs.factory.com/autonomy-and-safety/auto-run; [from-official-docs] https://docs.factory.com/droid-exec/overview; [from-official-docs] https://docs.factory.com/enterprise/hierarchical-settings-and-org-control,rater1 | |
| factory-droid,B,0.242.0,D02,1,,"Seatbelt/bubblewrap sandbox with filtering proxy that fails closed, but enabled only by sandbox.enabled=true (opt-in).",[from-official-docs] https://docs.factory.com/autonomy-and-safety/sandbox,rater1 | |
| factory-droid,B,0.242.0,D03,1,,Proxy domain allowlist applies only when the opt-in sandbox is enabled; org can disable network tools; fetch default not documented.,[from-official-docs] https://docs.factory.com/enterprise/hierarchical-settings-and-org-control,rater1 | |
| factory-droid,B,0.242.0,D04,1,,"Project permission rules need workspace trust and apiKeyHelper is stripped from repo settings, but trust gating of project hooks and .factory/mcp.json is not documented (unverified only).",[from-official-docs] https://docs.factory.com/autonomy-and-safety/permission-rules; [from-official-docs] https://docs.factory.com/model-independence/byok,rater1 | |
| factory-droid,B,0.242.0,D05,2,,Per-tool approvals bound to a fingerprint of the server transport (re-pointed server re-prompts) and org allowlist-only mcpPolicy. Not 3: OAuth scopes display not documented (tokens in keyring).,[from-official-docs] https://docs.factory.com/harness/mcp; [from-official-docs] https://docs.factory.com/enterprise/hierarchical-settings-and-org-control,rater1 | |
| factory-droid,B,0.242.0,D06,1,,"Org strictEnabledPlugins/strictKnownMarketplaces and internal marketplaces exist, but no vetting, signing or scanning is documented; plugins ship hooks and MCP servers.",[from-official-docs] https://docs.factory.com/enterprise/hierarchical-settings-and-org-control,rater1 | |
| factory-droid,B,0.242.0,D07,1,,"MCP OAuth tokens in keyring (file fallback) and Droid Shield commit scanning, but log/transcript retention is not documented (score-2 clause unmet).",[from-official-docs] https://docs.factory.com/harness/mcp; [from-official-docs] https://docs.factory.com/droid-cli/settings; [from-official-docs] https://docs.factory.com/enterprise/hierarchical-settings-and-org-control,rater1 | |
| factory-droid,B,0.242.0,D08,1,,"droid exec is read-only by default and fails without partial changes, but there is no official CI integration documenting token permissions or actor checks.",[from-official-docs] https://docs.factory.com/droid-exec/overview,rater1 | |
| factory-droid,B,0.242.0,D09,N/A,N/A (feature absent: no agent-written memory documented; AGENTS.md is user-authored).,N/A (feature absent: no agent-written memory documented; AGENTS.md is user-authored).,[from-official-docs] https://docs.factory.com/harness/agents-md,rater1 | |
| factory-droid,B,0.242.0,D10,2,,"Subagents inherit autonomy or use a separate subagent autonomy setting, always clamped to the org maximum. Not 3: no per-subagent sandbox/worktree.",[from-official-docs] https://docs.factory.com/autonomy-and-safety/auto-run,rater1 | |
| factory-droid,B,0.242.0,D11,3,,"Structured OTel spans to a customer collector via an org-pinned sink users cannot redirect, plus an organisation audit log of who did what.",[from-official-docs] https://docs.factory.com/enterprise/telemetry/privacy; [from-official-docs] https://docs.factory.com/enterprise/compliance-audit-and-monitoring,rater1 | |
| factory-droid,B,0.242.0,D12,1,,Per-user telemetry to Factory by default (aggregate optional); retention and training terms not documented.,[from-official-docs] https://docs.factory.com/enterprise/telemetry/privacy; [from-official-docs] https://docs.factory.com/droid-cli/settings; [from-official-docs] https://docs.factory.com/enterprise/privacy-and-data-flows,rater1 | |
| factory-droid,B,0.242.0,D13,1,,"curl|sh, brew cask and npm with no documented checksum or signature.",[from-official-docs] https://docs.factory.com/droid-cli/quickstart; [from-official-docs] https://registry.npmjs.org/droid,rater1 | |
| factory-droid,B,0.242.0,D14,2,,"Authoritative system-managed settings cover autonomy cap, managed hooks, MCP policy, plugins and telemetry sink. Not 3: sandbox enforcement not documented and nothing verified for a docs-only harness.",[from-official-docs] https://docs.factory.com/enterprise/hierarchical-settings-and-org-control; [from-official-docs] https://docs.factory.com/enterprise/telemetry/privacy,rater1 | |
| factory-droid,B,0.242.0,D15,0,,"Only compliance claims (SOC 2, ISO); no security contact, policy or advisories documented (absence of a VDP is unverified).",[from-official-docs] https://docs.factory.com/enterprise/compliance-audit-and-monitoring,rater1 | |
| gemini-cli,A,0.64.0,D01,2,,"Default approval mode asks (verified); user and admin TOML policy rules; workspace-tier policies disabled; YOLO is a flag and admins can disable it (disableYoloMode, secureMode). Not 3: deny rules applying in every mode are not documented. Patched CVE-2026-12537 (YOLO ignored allowlists) noted.",[from-official-docs] https://github.com/google-gemini/gemini-cli/blob/main/packages/cli/src/config/config.ts; [from-official-docs] https://github.com/google-gemini/gemini-cli/blob/main/docs/reference/policy-engine.md; [from-official-docs] https://github.com/google-gemini/gemini-cli/security/advisories/GHSA-wpqr-6v78-jr5g,rater1 | |
| gemini-cli,A,0.64.0,D02,1,,Seatbelt/Docker/Podman sandboxes are opt-in; default Seatbelt profile allows network.,[from-official-docs] https://github.com/google-gemini/gemini-cli/blob/main/docs/cli/sandbox.md,rater1 | |
| gemini-cli,A,0.64.0,D03,1,,Network is off only inside the opt-in sandbox; no domain allowlist.,[from-official-docs] https://github.com/google-gemini/gemini-cli/blob/main/docs/cli/sandbox.md,rater1 | |
| gemini-cli,A,0.64.0,D04,2,,"Folder trust on by default (code-verified): untrusted folders ignore project settings, .env and extensions; hooks fingerprinted; headless needs --skip-trust. Not 3: git-helper coverage and central folder pre-approval not documented. Patched CVE-2026-12537 noted.",[from-official-docs] https://github.com/google-gemini/gemini-cli/blob/main/packages/cli/src/config/trustedFolders.ts; [from-official-docs] https://github.com/google-gemini/gemini-cli/blob/main/packages/core/src/utils/trust.ts; [from-official-docs] https://github.com/google-gemini/gemini-cli/blob/main/docs/hooks/index.md; [from-official-docs] https://github.com/google-gemini/gemini-cli/security/advisories/GHSA-wpqr-6v78-jr5g,rater1 | |
| gemini-cli,A,0.64.0,D05,2,,Project servers come from trust-gated settings; includeTools/excludeTools per tool; mcp.allowed/excluded and admin-required servers; keychain OAuth. Not 3: no definition-change detection or pinning.,[from-official-docs] https://github.com/google-gemini/gemini-cli/blob/main/schemas/settings.schema.json; [from-official-docs] https://github.com/google-gemini/gemini-cli/blob/main/packages/cli/src/config/trustedFolders.ts,rater1 | |
| gemini-cli,A,0.64.0,D06,2,,"Extension sources restricted by allowedExtensions/blockGitExtensions, integrity checks, extension management disabled in untrusted folders. Not 3: no signed publishing.",[from-official-docs] https://github.com/google-gemini/gemini-cli/blob/main/schemas/settings.schema.json; [from-official-docs] https://github.com/google-gemini/gemini-cli/blob/main/packages/cli/src/config/trustedFolders.ts,rater1 | |
| gemini-cli,A,0.64.0,D07,1,,OAuth cache in ~/.gemini/oauth_creds.json (mode not documented); only MCP tokens use keychain; no redaction or log retention in the facts.,[from-official-docs] https://github.com/google-gemini/gemini-cli/blob/main/packages/core/src/config/storage.ts,rater1 | |
| gemini-cli,A,0.64.0,D08,1,,"Headless requires explicit trust, but the official run-gemini-cli action's token permissions and actor checks are not documented in the facts.",[from-official-docs] https://github.com/google-gemini/gemini-cli/blob/main/docs/cli/headless.md; [from-official-docs] https://github.com/google-gemini/gemini-cli/blob/main/packages/core/src/utils/trust.ts; [from-official-docs] https://github.com/google-gemini/gemini-cli/security/advisories/GHSA-wpqr-6v78-jr5g,rater1 | |
| gemini-cli,A,0.64.0,D09,1,,save_memory writes to the global ~/.gemini/GEMINI.md loaded every session; no confirmation or off switch documented.,[from-official-docs] https://github.com/google-gemini/gemini-cli/blob/main/docs/tools/memory.md,rater1 | |
| gemini-cli,A,0.64.0,D10,N/A,N/A (not documented: no sub-agent facts collected).,N/A (not documented: no sub-agent facts collected).,,rater1 | |
| gemini-cli,A,0.64.0,D11,2,,Local chat session files plus opt-in OpenTelemetry.,[from-official-docs] https://github.com/google-gemini/gemini-cli/blob/main/packages/core/src/config/storage.ts; [from-official-docs] https://github.com/google-gemini/gemini-cli/blob/main/schemas/settings.schema.json,rater1 | |
| gemini-cli,A,0.64.0,D12,1,,Usage statistics on by default with opt-out; prompt logging opt-in; retention not documented.,[from-official-docs] https://github.com/google-gemini/gemini-cli/blob/main/schemas/settings.schema.json; [from-official-docs] https://developers.googleblog.com/an-important-update-transitioning-gemini-cli-to-antigravity-cli/,rater1 | |
| gemini-cli,A,0.64.0,D13,1,,npm/brew distribution; macOS zips explicitly unsigned; no provenance documented.,[from-official-docs] https://github.com/google-gemini/gemini-cli/releases/tag/v0.64.0,rater1 | |
| gemini-cli,A,0.64.0,D14,2,,System settings file and admin policy tier with secureMode/disableYoloMode; MCP/telemetry coverage and verification missing.,[from-official-docs] https://github.com/google-gemini/gemini-cli/blob/main/docs/cli/enterprise.md; [from-official-docs] https://github.com/google-gemini/gemini-cli/blob/main/docs/reference/policy-engine.md; [from-official-docs] https://github.com/google-gemini/gemini-cli/blob/main/packages/cli/src/config/config.ts,rater1 | |
| gemini-cli,A,0.64.0,D15,3,,"SECURITY.md routes to Google VRP (bounty), coordination via GitHub Security Advisories, published critical GHSA/CVE.",[from-official-docs] https://github.com/google-gemini/gemini-cli/blob/main/SECURITY.md; [from-official-docs] https://github.com/google-gemini/gemini-cli/security/advisories/GHSA-wpqr-6v78-jr5g,rater1 | |
| github-copilot,B,copilot-cli v1.0.98 (release 2026-10-14); docs-level,D01,2,,"CLI prompts for tools, paths and URLs by default; --deny-tool beats allow-all; yolo is a flag/slash command that enterprises can block. Not 3: that approve-everything cannot be switched on from agent-writable files is not documented (CVE-2025-53773 did exactly that, patched). Patched classifier bypass CVE-2026-29783 noted.",[from-official-docs] https://docs.github.com/en/copilot/how-tos/copilot-cli/set-up-copilot-cli/configure-copilot-cli; [from-official-docs] https://code.visualstudio.com/docs/agents/reference/ai-settings; [from-official-docs] https://github.com/advisories/GHSA-3m2x-p87c-pwv6; [from-official-docs] https://github.com/github/copilot-cli/security/advisories/GHSA-g8r9-g2v8-jv6f; [from-official-docs] https://code.visualstudio.com/docs/copilot/security,rater1 | |
| github-copilot,B,copilot-cli v1.0.98 (release 2026-10-14); docs-level,D02,1,,Local CLI/VS Code sandbox exists but is off by default (VS Code allows unsandboxed commands by default); cloud agent is isolated by design.,[from-official-docs] https://docs.github.com/en/copilot/reference/copilot-cli-reference/cli-config-dir-reference; [from-official-docs] https://docs.github.com/en/copilot/concepts/agents/cloud-agent/about-cloud-agent,rater1 | |
| github-copilot,B,copilot-cli v1.0.98 (release 2026-10-14); docs-level,D03,2,,CLI asks for every URL by default with allow/deny URL lists and a sandbox network policy (opt-in). Not 3: no default-deny sandbox network locally; VS Code network filter off by default; 3P: cloud firewall excludes MCP and setup steps.,[from-official-docs] https://docs.github.com/en/copilot/how-tos/copilot-cli/set-up-copilot-cli/configure-copilot-cli; [from-official-docs] https://code.visualstudio.com/docs/agents/reference/ai-settings; [third-party-report] https://github.blog/changelog/2026-04-03-organization-firewall-settings-for-copilot-cloud-agent/,rater1 | |
| github-copilot,B,copilot-cli v1.0.98 (release 2026-10-14); docs-level,D04,1,,"CLI trust prompt skips project MCP in untrusted dirs and VS Code Restricted Mode disables agents/MCP, but pre-trust handling of hooks is not documented. Patched core.fsmonitor git-helper RCE CVE-2026-45033 noted.",[from-official-docs] https://docs.github.com/en/copilot/how-tos/copilot-cli/set-up-copilot-cli/configure-copilot-cli; [from-official-docs] https://github.com/github/copilot-cli/security/advisories/GHSA-9ccr-r5hg-74gf,rater1 | |
| github-copilot,B,copilot-cli v1.0.98 (release 2026-10-14); docs-level,D05,2,,"Project servers are workspace-trust gated, tool calls prompt, and the ChatMCP registryOnly policy is a managed allowlist. Not 3: no per-server prompt once trusted, no change detection; cloud agent uses MCP tools without approval.",[from-official-docs] https://code.visualstudio.com/docs/copilot/security; [from-official-docs] https://docs.github.com/en/copilot/how-tos/use-copilot-agents/cloud-agent/extend-cloud-agent-with-mcp; [from-official-docs] https://docs.github.com/en/copilot/how-tos/copilot-cli/set-up-copilot-cli/configure-copilot-cli,rater1 | |
| github-copilot,B,copilot-cli v1.0.98 (release 2026-10-14); docs-level,D06,1,,Plugins with marketplaces and a VS Code policy for extension tools; no vetting or signing documented.,[from-official-docs] https://docs.github.com/en/copilot/reference/copilot-cli-reference/cli-config-dir-reference,rater1 | |
| github-copilot,B,copilot-cli v1.0.98 (release 2026-10-14); docs-level,D07,1,,MCP OAuth/secrets in keychain with plaintext fallbacks; log retention and redaction not documented (score-2 clauses unmet).,[from-official-docs] https://docs.github.com/en/copilot/reference/copilot-cli-reference/cli-config-dir-reference; [from-official-docs] https://docs.github.com/en/copilot/how-tos/use-copilot-agents/cloud-agent/extend-cloud-agent-with-mcp,rater1 | |
| github-copilot,B,copilot-cli v1.0.98 (release 2026-10-14); docs-level,D08,2,,"copilot -p needs --allow-tool for unprompted tools; docs recommend least-privilege GITHUB_TOKEN; cloud agent only triggered by write-access users, workflows need approval. Not 3: default read-only token and PR-head config blocking not documented.",[from-official-docs] https://docs.github.com/en/copilot/how-tos/copilot-cli/automate-with-actions; [from-official-docs] https://docs.github.com/en/copilot/concepts/agents/coding-agent/risks-and-mitigations,rater1 | |
| github-copilot,B,copilot-cli v1.0.98 (release 2026-10-14); docs-level,D09,1,,"Cross-session session-store index and personal instructions persist; review, confirmation and off switch not documented.",[from-official-docs] https://docs.github.com/en/copilot/reference/copilot-cli-reference/cli-config-dir-reference,rater1 | |
| github-copilot,B,copilot-cli v1.0.98 (release 2026-10-14); docs-level,D10,1,,Custom agents exist; permission inheritance and isolation undocumented.,[from-official-docs] https://docs.github.com/en/copilot/reference/copilot-cli-reference/cli-config-dir-reference,rater1 | |
| github-copilot,B,copilot-cli v1.0.98 (release 2026-10-14); docs-level,D11,1,,"Structured local events.jsonl and logs, but no telemetry/central export documented in the facts (score-2 clause unmet).",[from-official-docs] https://docs.github.com/en/copilot/reference/copilot-cli-reference/cli-config-dir-reference,rater1 | |
| github-copilot,B,copilot-cli v1.0.98 (release 2026-10-14); docs-level,D12,N/A,N/A (only a third-party report; official retention/training page not collected).,N/A (only a third-party report; official retention/training page not collected).,[third-party-report] https://dev.classmethod.jp/en/articles/github-copilot-interaction-data-ai-training-opt-out-2026/,rater1 | |
| github-copilot,B,copilot-cli v1.0.98 (release 2026-10-14); docs-level,D13,1,,"curl script, npm, brew and winget with no documented signature or checksum; auto-update channel configurable.",[from-official-docs] https://github.com/github/copilot-cli; [from-official-docs] https://docs.github.com/en/copilot/reference/copilot-cli-reference/cli-config-dir-reference,rater1 | |
| github-copilot,B,copilot-cli v1.0.98 (release 2026-10-14); docs-level,D14,2,,"Enterprise blocks /yolo and bypass, VS Code policies for auto-approve, agent mode, MCP and extension tools, org firewall, CLI disable. Not 3: sandbox/telemetry coverage not documented and nothing verified.",[from-official-docs] https://code.visualstudio.com/docs/copilot/security; [from-official-docs] https://docs.github.com/en/copilot/reference/copilot-cli-reference/cli-config-dir-reference,rater1 | |
| github-copilot,B,copilot-cli v1.0.98 (release 2026-10-14); docs-level,D15,3,,GitHub Bug Bounty via HackerOne and GHSA advisories with CVEs on github/copilot-cli.,[from-official-docs] https://bounty.github.com/; [from-official-docs] https://github.com/github/copilot-cli/security/advisories,rater1 | |
| github-copilot-sdk,C,1.0.22,D01,3,,"Tier C: onPermissionRequest runs before every tool kind and requests stay pending (denied) if no handler is set, i.e. approval on by default; managed settings (managedApprovalRequired, approveAll throws under enableManagedSettings) let a platform policy enforce it. Per-tool skipPermission remains a documented footgun.",[from-official-docs] https://github.com/github/copilot-sdk/blob/main/nodejs/README.md,rater1 | |
| github-copilot-sdk,C,1.0.22,D02,0,,Tier C: no sandbox in the SDK; isolation is application responsibility (container recommended).,[from-official-docs] https://github.com/github/copilot-sdk/blob/main/docs/setup/backend-services.md,rater1 | |
| github-copilot-sdk,C,1.0.22,D03,1,,Tier C: managed network policy hook for client-owned browsers and a 'url' permission kind; built-in SSRF protection not documented.,[from-official-docs] https://github.com/github/copilot-sdk/blob/main/nodejs/README.md,rater1 | |
| github-copilot-sdk,C,1.0.22,D04,2,,Runtime auto-loads repository customization (.github/copilot/settings.json) only under folder trust covering the checkout root; SDK inputs outside the checkout are not covered.,[from-official-docs] https://github.com/github/copilot-sdk/blob/main/docs/features/customization-locations.md,rater1 | |
| github-copilot-sdk,C,1.0.22,D05,1,,"Tier C: MCP calls go through the default-pending permission gate with a per-session tools allowlist, but MCP auth/header handling is not documented (rung 2 needs both).",[from-official-docs] https://github.com/github/copilot-sdk/blob/main/docs/features/mcp.md; [from-official-docs] https://github.com/github/copilot-sdk/blob/main/nodejs/README.md,rater1 | |
| github-copilot-sdk,C,1.0.22,D06,N/A,N/A (tier C: no built-in marketplace or registry; the skill install review is an application-hosted flow).,N/A (tier C: no built-in marketplace or registry; the skill install review is an application-hosted flow).,[from-official-docs] https://github.com/github/copilot-sdk/blob/main/README.md,rater1 | |
| github-copilot-sdk,C,1.0.22,D07,2,,Tier C: CLI login credentials in the system keychain; env tokens and BYOK keys as alternatives.,[from-official-docs] https://github.com/github/copilot-sdk/blob/main/docs/setup/local-cli.md,rater1 | |
| github-copilot-sdk,C,1.0.22,D08,2,,Tier C: every tool call needs a permission decision and managed approval can be required for unattended runs; headless server binds loopback with a connection token by default.,[from-official-docs] https://github.com/github/copilot-sdk/blob/main/nodejs/README.md,rater1 | |
| github-copilot-sdk,C,1.0.22,D09,N/A,N/A (tier C: persisted session state is not documented as replayed across sessions; only a 'memory' permission kind is named).,N/A (tier C: persisted session state is not documented as replayed across sessions; only a 'memory' permission kind is named).,[from-official-docs] https://github.com/github/copilot-sdk/blob/main/docs/features/session-persistence.md,rater1 | |
| github-copilot-sdk,C,1.0.22,D10,N/A,N/A (not documented: sub-agent facts are unverified).,N/A (not documented: sub-agent facts are unverified).,,rater1 | |
| github-copilot-sdk,C,1.0.22,D11,2,,"Tier C: built-in OTel (OTLP or file JSONL) one setting away, content capture opt-in.",[from-official-docs] https://github.com/github/copilot-sdk/blob/main/docs/observability/opentelemetry.md,rater1 | |
| github-copilot-sdk,C,1.0.22,D12,N/A,N/A (not documented: only an unverified fact).,N/A (not documented: only an unverified fact).,,rater1 | |
| github-copilot-sdk,C,1.0.22,D13,N/A,N/A (not documented: provenance not checked; only an unverified fact).,N/A (not documented: provenance not checked; only an unverified fact).,,rater1 | |
| github-copilot-sdk,C,1.0.22,D14,2,,Managed settings with account/device policy resolution and per-request managedApprovalRequired; coverage of MCP/telemetry not documented.,[from-official-docs] https://github.com/github/copilot-sdk/blob/main/nodejs/README.md,rater1 | |
| github-copilot-sdk,C,1.0.22,D15,1,,SECURITY.md contact; OSS repos out of bounty scope; no repo advisories.,[from-official-docs] https://github.com/github/copilot-sdk/blob/main/SECURITY.md,rater1 | |
| github-copilot-sdk,C,1.0.22,D16,1,,Headless CLI server binds 127.0.0.1 with a random connection token by default; Origin/Host validation not documented.,[from-official-docs] https://github.com/github/copilot-sdk/blob/main/nodejs/README.md,rater1 | |
| google-adk,C,2.12.0,D01,1,,Tier C: require_confirmation exists but defaults False; BashTool allows every prefix by default.,[from-official-docs] https://github.com/google/adk-python/blob/main/src/google/adk/tools/function_tool.py; [from-official-docs] https://github.com/google/adk-python/blob/main/src/google/adk/tools/bash_tool.py,rater1 | |
| google-adk,C,2.12.0,D02,1,,Tier C: container/GKE/Vertex executors are optional; the local executor is explicitly unsafe.,[from-official-docs] https://github.com/google/adk-python/blob/main/src/google/adk/code_executors/unsafe_local_code_executor.py,rater1 | |
| google-adk,C,2.12.0,D03,N/A,N/A (not documented: no egress facts collected).,N/A (not documented: no egress facts collected).,,rater1 | |
| google-adk,C,2.12.0,D04,N/A,"N/A (tier C: the runtime loads no repository-level configuration; the facts concern the local server, a D16 topic).","N/A (tier C: the runtime loads no repository-level configuration; the facts concern the local server, a D16 topic).",[from-official-docs] https://github.com/google/adk-python/blob/main/src/google/adk/cli/cli_tools_click.py,rater1 | |
| google-adk,C,2.12.0,D05,1,,"Tier C: tool_filter, require_confirmation and auth/header providers exist, but confirmation is off by default.",[from-official-docs] https://github.com/google/adk-python/blob/main/src/google/adk/tools/mcp_tool/mcp_toolset.py,rater1 | |
| google-adk,C,2.12.0,D06,N/A,N/A (not documented: no extension facts collected).,N/A (not documented: no extension facts collected).,,rater1 | |
| google-adk,C,2.12.0,D07,N/A,N/A (not documented: no credential facts collected).,N/A (not documented: no credential facts collected).,,rater1 | |
| google-adk,C,2.12.0,D08,N/A,N/A (not documented: no unattended-run facts collected).,N/A (not documented: no unattended-run facts collected).,,rater1 | |
| google-adk,C,2.12.0,D09,1,,"Memory service for cross-session recall exists; scoping, review and off switch not documented.",[from-official-docs] https://github.com/google/adk-python/blob/main/src/google/adk/sessions/database_session_service.py,rater1 | |
| google-adk,C,2.12.0,D10,N/A,N/A (not documented: no multi-agent scoping facts collected).,N/A (not documented: no multi-agent scoping facts collected).,,rater1 | |
| google-adk,C,2.12.0,D11,2,,Tier C: built-in OpenTelemetry tracing/metrics and a SQLite span exporter.,[from-official-docs] https://github.com/google/adk-python/blob/main/src/google/adk/telemetry/setup.py,rater1 | |
| google-adk,C,2.12.0,D12,N/A,N/A (not documented: only an unverified fact).,N/A (not documented: only an unverified fact).,,rater1 | |
| google-adk,C,2.12.0,D13,N/A,N/A (not documented: no distribution facts collected).,N/A (not documented: no distribution facts collected).,,rater1 | |
| google-adk,C,2.12.0,D14,N/A,N/A (not documented: no managed-policy facts collected).,N/A (not documented: no managed-policy facts collected).,,rater1 | |
| google-adk,C,2.12.0,D15,1,,"A critical GHSA/CVE was issued, but no security policy is documented (SECURITY.md absence and VRP applicability are unverified), so score 2 is not met.",[from-official-docs] https://github.com/advisories/GHSA-rg7c-g689-fr3x,rater1 | |
| google-adk,C,2.12.0,D16,0,,adk web/api_server binds loopback without documented authentication; patched unauthenticated RCE CVE-2026-4810.,[from-official-docs] https://github.com/google/adk-python/blob/main/src/google/adk/cli/cli_tools_click.py,rater1 | |
| goose,A,1.54.0,D01,1,,"Per-tool always_allow/ask_before/never_allow and approve modes exist, but the shipped default is autonomous mode (no approvals).",[from-official-docs] https://github.com/aaif-goose/goose/blob/main/crates/goose-provider-types/src/goose_mode.rs; [from-official-docs] https://github.com/aaif-goose/goose/blob/main/documentation/docs/guides/managing-tools/goose-permissions.md; [from-official-docs] https://github.com/aaif-goose/goose/blob/main/crates/goose/src/config/permission.rs,rater1 | |
| goose,A,1.54.0,D02,0,,No sandbox; README recommends running in a VM or container.,[from-official-docs] https://github.com/aaif-goose/goose/blob/main/README.md,rater1 | |
| goose,A,1.54.0,D03,0,,No egress control or domain allowlist.,[from-official-docs] https://github.com/aaif-goose/goose/blob/main/documentation/docs/guides/config-files.md,rater1 | |
| goose,A,1.54.0,D04,0,,"No trust concept documented; goose review honoured attacker .git/config (core.fsmonitor) before any model call (CVE-2026-72718, patched 1.44.0).",[from-official-docs] https://github.com/aaif-goose/goose/security/advisories/GHSA-r5pp-p5r8-466r,rater1 | |
| goose,A,1.54.0,D05,1,,"GOOSE_ALLOWLIST restricts installable extension commands and an OSV malware check runs before launch, but the default autonomous mode auto-approves every tool (blanket) and there is no change detection.",[from-official-docs] https://github.com/aaif-goose/goose/blob/main/documentation/docs/guides/allowlist.md; [from-official-docs] https://github.com/aaif-goose/goose/blob/main/crates/goose/src/agents/extension_malware_check.rs; [from-official-docs] https://github.com/aaif-goose/goose/blob/main/documentation/docs/guides/managing-tools/goose-permissions.md,rater1 | |
| goose,A,1.54.0,D06,1,,Recipes carry extension blocks shareable by deeplink; no signing or vetting.,[from-official-docs] https://github.com/aaif-goose/goose/blob/main/documentation/docs/guides/recipes/recipe-reference.md,rater1 | |
| goose,A,1.54.0,D07,1,,"Keyring by default with plaintext secrets.yaml fallback; log retention, redaction and secret-file protection not documented (score-2 clauses unmet).",[from-official-docs] https://github.com/aaif-goose/goose/blob/main/documentation/docs/guides/config-files.md,rater1 | |
| goose,A,1.54.0,D08,1,,goose run inherits the autonomous default; prompt-injection detector off by default; README guidance to use a VM/container.,[from-official-docs] https://github.com/aaif-goose/goose/blob/main/documentation/docs/guides/config-files.md; [from-official-docs] https://github.com/aaif-goose/goose/blob/main/crates/goose-provider-types/src/goose_mode.rs,rater1 | |
| goose,A,1.54.0,D09,1,,"Built-in memory extension exists; scope, confirmation and off switch not documented.",[from-official-docs] https://github.com/aaif-goose/goose/blob/main/documentation/docs/guides/config-files.md,rater1 | |
| goose,A,1.54.0,D10,1,,Summon extension delegates to other agents; scoping not documented.,[from-official-docs] https://github.com/aaif-goose/goose/blob/main/documentation/docs/guides/recipes/recipe-reference.md,rater1 | |
| goose,A,1.54.0,D11,1,,Local SQLite session store and logs; no export documented.,[from-official-docs] https://github.com/aaif-goose/goose/blob/main/documentation/docs/guides/logs.md,rater1 | |
| goose,A,1.54.0,D12,1,,"Telemetry opt-in (default false, asked on first use), but retention is not documented (score-1 clause).",[from-official-docs] https://github.com/aaif-goose/goose/blob/main/documentation/docs/guides/usage-data.md,rater1 | |
| goose,A,1.54.0,D13,1,,curl|bash installer without documented checksum or signature.,[from-official-docs] https://github.com/aaif-goose/goose/blob/main/README.md,rater1 | |
| goose,A,1.54.0,D14,1,,GOOSE_ALLOWLIST and custom distributions cover a small subset; no managed settings file.,[from-official-docs] https://github.com/aaif-goose/goose/blob/main/documentation/docs/guides/custom-distributions.md; [from-official-docs] https://github.com/aaif-goose/goose/blob/main/documentation/docs/guides/allowlist.md,rater1 | |
| goose,A,1.54.0,D15,2,,Private reporting via GitHub Security tab with AAIF escalation contact; one GHSA with CVE; no bounty.,[from-official-docs] https://github.com/aaif-goose/goose/blob/main/README.md; [from-official-docs] https://github.com/aaif-goose/goose/security/advisories,rater1 | |
| grok-build,A,1.0.52,D01,3,,"Ask is the default; deny rules and PreToolUse hooks still apply in always-approve; the mode is set only in user config, never project .grok/config.toml; root-owned requirements.toml disable_bypass_permissions_mode disables --yolo and all toggles.",[from-official-docs] https://docs.x.ai/build/features/permissions; [from-official-docs] https://docs.x.ai/build/enterprise,rater1 | |
| grok-build,A,1.0.52,D02,1,,Landlock/Seatbelt profiles exist but the default profile is off; child-network blocking Linux only.,[from-official-docs] https://docs.x.ai/build/features/sandbox,rater1 | |
| grok-build,A,1.0.52,D03,1,,"web_fetch is off by default, but there is no domain allowlist; network control is only the on/off restrict_network of opt-in sandbox profiles.",[from-official-docs] https://docs.x.ai/build/settings/reference; [from-official-docs] https://docs.x.ai/build/features/sandbox,rater1 | |
| grok-build,A,1.0.52,D04,2,,"Folder trust gates .mcp.json, project permission/MCP/plugins config, LSP, instructions, skills and hooks; --trust is an explicit flag. Not 3: env and git-helper coverage and central folder pre-approval not documented.",[from-official-docs] https://docs.x.ai/build/features/hooks; [from-official-docs] https://github.com/xai-org/grok-build/blob/main/crates/codegen/xai-grok-shell/src/agent/folder_trust.rs; [from-official-docs] https://docs.x.ai/build/settings/reference,rater1 | |
| grok-build,A,1.0.52,D05,2,,"Project-scope servers behind folder trust, tool permission rules, managed MCP allowlist via Claude managed-settings compatibility. Not 3: no change detection; OAuth tokens in a JSON file.",[from-official-docs] https://docs.x.ai/build/features/mcp-servers; [from-official-docs] https://docs.x.ai/build/enterprise; [from-official-docs] https://github.com/xai-org/grok-build/blob/main/crates/codegen/xai-grok-shell/src/agent/folder_trust.rs,rater1 | |
| grok-build,A,1.0.52,D06,1,,"Per-plugin trust store and managed marketplace restrictions, but no vetting, signing or scanning documented.",[from-official-docs] https://github.com/xai-org/grok-build/blob/main/crates/codegen/xai-grok-shell/src/agent/folder_trust.rs; [from-official-docs] https://docs.x.ai/build/enterprise,rater1 | |
| grok-build,A,1.0.52,D07,1,,Session and MCP OAuth tokens in files under ~/.grok; keychain and file modes not documented.,[from-official-docs] https://docs.x.ai/build/settings/reference; [from-official-docs] https://docs.x.ai/build/enterprise,rater1 | |
| grok-build,A,1.0.52,D08,1,,"Official CI example uses dontAsk, explicit --allow and --sandbox strict, but actor/trigger checks and token permissions are not documented, and --always-approve is offered for headless.",[from-official-docs] https://docs.x.ai/build/cli/headless-scripting; [from-official-docs] https://docs.x.ai/build/enterprise,rater1 | |
| grok-build,A,1.0.52,D09,2,,"Cross-session memory is off by default (GROK_MEMORY=0). Not 3: scoping, scanning and central disable not documented.",[from-official-docs] https://docs.x.ai/build/settings/reference,rater1 | |
| grok-build,A,1.0.52,D10,1,,Subagents disabled by default (GROK_SUBAGENTS=0); permission inheritance not documented; worktree isolation only third-party reported (cannot raise).,[from-official-docs] https://docs.x.ai/build/settings/reference; [third-party-report] https://mer.vin/2026/05/grok-build-cli-xai-terminal-coding-agent-with-plan-mode-subagents-and-headless-ci/,rater1 | |
| grok-build,A,1.0.52,D11,1,,Local session and log files; no export documented.,[from-official-docs] https://docs.x.ai/build/cli/headless-scripting,rater1 | |
| grok-build,A,1.0.52,D12,1,,"Telemetry (Mixpanel, Sentry) exists with scrubbing; ZDR at team level; default and opt-out not documented. 3P: v0.2.93 uploaded whole workspaces regardless of the toggle (disabled server-side 2026-07-13).",[from-official-docs] https://github.com/xai-org/grok-build/tree/main/crates/codegen; [from-official-docs] https://docs.x.ai/build/enterprise; [third-party-report] https://devops.com/xai-open-sources-grok-build-coding-agent-after-cloud-upload-exposes-ssh-keys-repos/,rater1 | |
| grok-build,A,1.0.52,D13,1,,"npm and curl|bash with auto-update on by default (disableable); no signature, checksum or provenance documented.",[from-official-docs] https://docs.x.ai/build/overview; [from-official-docs] https://registry.npmjs.org/@xai-official/grok; [from-official-docs] https://docs.x.ai/build/settings/reference,rater1 | |
| grok-build,A,1.0.52,D14,2,,"Root-owned requirements.toml pins cannot be overridden (bypass, sandbox, auth) and a Claude managed-settings subset covers MCP/marketplaces/telemetry flags. Not 3: not verified for a docs-only harness.",[from-official-docs] https://docs.x.ai/build/enterprise,rater1 | |
| grok-build,A,1.0.52,D15,1,,"SECURITY.md routes to the HackerOne bounty, but no advisories are published (the 2026 upload incident had none).",[from-official-docs] https://x.ai/security.txt; [from-official-docs] https://github.com/xai-org/grok-build/blob/main/SECURITY.md; [third-party-report] https://devops.com/xai-open-sources-grok-build-coding-agent-after-cloud-upload-exposes-ssh-keys-repos/,rater1 | |
| hermes-agent,A,0.22.1,D01,1,,"Default approvals.mode is 'smart', where an auxiliary LLM auto-approves commands without asking the user (score-1 clause, as for classifier defaults elsewhere); deny globs hold before --yolo and a hardline blocklist exists; SECURITY.md calls the gate a heuristic, not a boundary.",[from-official-docs] https://github.com/NousResearch/hermes-agent/blob/main/website/docs/user-guide/security.md; [from-official-docs] https://github.com/NousResearch/hermes-agent/blob/main/SECURITY.md,rater1 | |
| hermes-agent,A,0.22.1,D02,1,,Default terminal backend runs on the host; Docker/SSH/cloud backends are opt-in and confine only shell/file tools.,[from-official-docs] https://github.com/NousResearch/hermes-agent/blob/main/SECURITY.md; [third-party-report] https://github.com/advisories/GHSA-wm96-9gfh-vvgq,rater1 | |
| hermes-agent,A,0.22.1,D03,1,,"Upstream allowlist via iron-proxy exists only for the opt-in Docker backend, whose default network_mode is host.",[from-official-docs] https://github.com/NousResearch/hermes-agent/blob/main/website/docs/user-guide/egress/iron-proxy.md; [from-official-docs] https://github.com/NousResearch/hermes-agent/blob/main/website/docs/user-guide/egress/network-isolation.md,rater1 | |
| hermes-agent,A,0.22.1,D04,2,,"Project plugins disabled by default, shell hooks need first-use consent, MCP has a trust surface; context files load unprompted but scanned. Not 3: hook consent keys on the command string (script edits silently trusted); headless and central policy not documented. Patched scanner/plugin CVEs noted.",[from-official-docs] https://github.com/NousResearch/hermes-agent/blob/main/website/docs/user-guide/features/plugins.md; [from-official-docs] https://github.com/NousResearch/hermes-agent/blob/main/website/docs/user-guide/features/hooks.md; [third-party-report] https://github.com/advisories/GHSA-pgp4-xr4j-h5cg,rater1 | |
| hermes-agent,A,0.22.1,D05,1,,"MCP servers only in user/managed config with an mcp-trust approval surface and env scrubbing, but per-tool permissions and change detection are not documented in the facts.",[from-official-docs] https://github.com/NousResearch/hermes-agent/blob/main/SECURITY.md; [from-official-docs] https://github.com/NousResearch/hermes-agent/blob/main/website/docs/user-guide/features/hooks.md,rater1 | |
| hermes-agent,A,0.22.1,D06,1,,Skills run arbitrary Python at import; Skills Guard scanning is only a review aid (3P bypass CVE); source scoping or pinning not documented.,[from-official-docs] https://github.com/NousResearch/hermes-agent/blob/main/SECURITY.md; [third-party-report] https://github.com/advisories/GHSA-238w-f66p-w349,rater1 | |
| hermes-agent,A,0.22.1,D07,1,,"Credentials stripped from child environments and logs redacted, but credential storage (keychain or verified 0600) is not documented in the facts; 3P: world-readable state files fixed in 0.16.0.",[from-official-docs] https://github.com/NousResearch/hermes-agent/blob/main/SECURITY.md; [third-party-report] https://github.com/advisories/GHSA-99f9-j8r3-p853; [from-official-docs] https://github.com/NousResearch/hermes-agent/blob/main/website/docs/user-guide/configuration.md,rater1 | |
| hermes-agent,A,0.22.1,D08,1,,"single_query, cron and unattended modes default to deny and the gateway is default-deny, but there is no official CI integration (score-2 clause unmet).",[from-official-docs] https://github.com/NousResearch/hermes-agent/blob/main/website/docs/user-guide/security.md,rater1 | |
| hermes-agent,A,0.22.1,D09,1,,Agent-written global MEMORY.md/USER.md injected every session; write_approval exists but is off by default; scanner had a CVE (3P).,[from-official-docs] https://github.com/NousResearch/hermes-agent/blob/main/website/docs/user-guide/features/memory.md; [third-party-report] https://github.com/advisories/GHSA-33qv-c5qm-799v,rater1 | |
| hermes-agent,A,0.22.1,D10,1,,delegate_task children inherit the parent's tool access (the model cannot widen it) with isolated context and terminals; per-child restriction and bypass propagation not documented.,[from-official-docs] https://github.com/NousResearch/hermes-agent/blob/main/website/docs/user-guide/features/delegation.md,rater1 | |
| hermes-agent,A,0.22.1,D11,1,,Local logs with secret redaction and pre-approval hooks; structured tool-call log export not documented.,[from-official-docs] https://github.com/NousResearch/hermes-agent/blob/main/website/docs/user-guide/configuration.md,rater1 | |
| hermes-agent,A,0.22.1,D12,N/A,N/A (not documented: only an unverified fact).,N/A (not documented: only an unverified fact).,,rater1 | |
| hermes-agent,A,0.22.1,D13,1,,Main curl|bash installer without checksum/signature; only the Termux APT repo is signed.,[from-official-docs] https://github.com/NousResearch/hermes-agent/blob/main/README.md; [from-official-docs] https://github.com/NousResearch/hermes-agent/blob/main/website/docs/user-guide/configuration.md,rater1 | |
| hermes-agent,A,0.22.1,D14,2,,Root-owned /etc/hermes managed scope pins selected keys above user config and shell env; coverage of MCP/plugins/sandbox/telemetry not documented.,[from-official-docs] https://github.com/NousResearch/hermes-agent/blob/main/website/docs/user-guide/managed-scope.md,rater1 | |
| hermes-agent,A,0.22.1,D15,2,,SECURITY.md with GHSA/email route and 10 CVE-bearing advisories; no bug bounty.,[from-official-docs] https://github.com/NousResearch/hermes-agent/blob/main/SECURITY.md; [from-official-docs] https://api.github.com/advisories?affects=hermes-agent,rater1 | |
| jules,B,unverified (service changelog newest 2026-03-09),D01,1,,"Approval is coarse plan review, and plans can be auto-approved by a 'Planning Critic'; no per-action approval.",[from-official-docs] https://jules.google/docs/changelog/,rater1 | |
| jules,B,unverified (service changelog newest 2026-03-09),D02,2,,Each task runs in a short-lived cloud VM (isolation on by default). Not 3: VM has internet access and nothing is centrally configurable.,[from-official-docs] https://jules.google/docs/environment/,rater1 | |
| jules,B,unverified (service changelog newest 2026-03-09),D03,0,,VM has internet access with no egress controls documented.,[from-official-docs] https://jules.google/docs/faq/,rater1 | |
| jules,B,unverified (service changelog newest 2026-03-09),D04,0,,Repository AGENTS.md and setup script are consumed with no trust concept (execution is inside the ephemeral VM).,[from-official-docs] https://jules.google/docs/faq/,rater1 | |
| jules,B,unverified (service changelog newest 2026-03-09),D05,1,,"Only vendor-curated MCP integrations configured in Settings; no per-tool permissions, managed policy or change detection documented.",[from-official-docs] https://jules.google/docs/changelog/,rater1 | |
| jules,B,unverified (service changelog newest 2026-03-09),D06,N/A,N/A (feature absent: no plugin ecosystem found).,N/A (feature absent: no plugin ecosystem found).,,rater1 | |
| jules,B,unverified (service changelog newest 2026-03-09),D07,N/A,N/A (not documented: secret handling in the VM is undocumented).,N/A (not documented: secret handling in the VM is undocumented).,[from-official-docs] https://jules.google/docs/faq/,rater1 | |
| jules,B,unverified (service changelog newest 2026-03-09),D08,1,,"API/CLI runs produce PRs for human review, but there is no documented CI hardening or untrusted-input guidance.",[from-official-docs] https://jules.google/docs/cli/reference,rater1 | |
| jules,B,unverified (service changelog newest 2026-03-09),D09,N/A,"N/A (not documented: only a per-repo environment snapshot is described, not agent memory).","N/A (not documented: only a per-repo environment snapshot is described, not agent memory).",[from-official-docs] https://jules.google/docs/environment/,rater1 | |
| jules,B,unverified (service changelog newest 2026-03-09),D10,N/A,N/A (not documented: no sub-agent facts).,N/A (not documented: no sub-agent facts).,,rater1 | |
| jules,B,unverified (service changelog newest 2026-03-09),D11,N/A,N/A (not documented: no audit-logging facts).,N/A (not documented: no audit-logging facts).,,rater1 | |
| jules,B,unverified (service changelog newest 2026-03-09),D12,1,,"No training on private repositories, but retention is undocumented (score-1 clause).",[from-official-docs] https://jules.google/docs/faq/,rater1 | |
| jules,B,unverified (service changelog newest 2026-03-09),D13,1,,npm registry CLI package with no documented signature or provenance.,[from-official-docs] https://registry.npmjs.org/@google/jules,rater1 | |
| jules,B,unverified (service changelog newest 2026-03-09),D14,0,,Workspace and enterprise accounts not supported; no administrator-enforced settings.,[from-official-docs] https://jules.google/docs/usage-limits/,rater1 | |
| jules,B,unverified (service changelog newest 2026-03-09),D15,N/A,N/A (not documented: no Jules-specific disclosure facts collected).,N/A (not documented: no Jules-specific disclosure facts collected).,,rater1 | |
| junie,B,3800.1,D01,1,,"Most commands, code execution and MCP calls need approval by default with a user allowlist, but no deny rules are documented and brave mode can be set via config.json whose scope (user vs project) is not stated. Patched CVE-2025-59458 noted.",[from-official-docs] https://junie.jetbrains.com/docs/action-allowlist-junie-cli.html; [from-official-docs] https://junie.jetbrains.com/docs/junie-cli.html; [from-official-docs] https://nvd.nist.gov/vuln/detail/CVE-2025-59458,rater1 | |
| junie,B,3800.1,D02,1,,"Optional srt sandbox, off by default, not covering MCP servers or hooks.",[from-official-docs] https://junie.jetbrains.com/docs/junie-cli.html,rater1 | |
| junie,B,3800.1,D03,1,,allowedDomains only inside the opt-in sandbox; unrestricted by default.,[from-official-docs] https://junie.jetbrains.com/docs/junie-cli.html,rater1 | |
| junie,B,3800.1,D04,2,,"Interactive trust prompt; untrusted projects load no project config, MCP or hooks; non-interactive runs are documented as trusted by design (allowed at 2). Patched CVE-2026-41153 noted.",[from-official-docs] https://junie.jetbrains.com/docs/junie-headless.html; [from-official-docs] https://nvd.nist.gov/vuln/detail/CVE-2026-41153,rater1 | |
| junie,B,3800.1,D05,1,,"Tool calls need approval with an mcpTools allowlist category, but servers are enabled by default once configured, with no managed list or change detection.",[from-official-docs] https://junie.jetbrains.com/docs/junie-cli-mcp-configuration.html; [from-official-docs] https://junie.jetbrains.com/docs/junie-cli.html; [from-official-docs] https://junie.jetbrains.com/docs/junie-on-github.html,rater1 | |
| junie,B,3800.1,D06,1,,Extensions and agent skills exist; no vetting documented (only unverified).,[from-official-docs] https://junie.jetbrains.com/docs/junie-cli.html,rater1 | |
| junie,B,3800.1,D07,1,,"readSecretFile category and .aiignore restrict secret reads (brave mode overrides), but credential storage is not documented.",[from-official-docs] https://junie.jetbrains.com/docs/junie-ide-plugin.html,rater1 | |
| junie,B,3800.1,D08,1,,"Official GitHub Action requests write scopes, checks only write permission and gives no untrusted-input guidance; headless auto-trusts project config.",[from-official-docs] https://junie.jetbrains.com/docs/junie-on-github.html; [from-official-docs] https://junie.jetbrains.com/docs/junie-headless.html,rater1 | |
| junie,B,3800.1,D09,N/A,N/A (feature absent: guideline files are user-authored and no autonomous memory writes are documented).,N/A (feature absent: guideline files are user-authored and no autonomous memory writes are documented).,[from-official-docs] https://junie.jetbrains.com/docs/guidelines-and-memory.html,rater1 | |
| junie,B,3800.1,D10,N/A,N/A (not documented: sub-agent facts are unverified).,N/A (not documented: sub-agent facts are unverified).,,rater1 | |
| junie,B,3800.1,D11,1,,"Structured per-session events.jsonl and transcript, but no export documented (score-2 clause unmet).",[from-official-docs] https://junie.jetbrains.com/docs/junie-cli.html,rater1 | |
| junie,B,3800.1,D12,1,,Anonymous statistics toggle exists; default and retention not documented.,[from-official-docs] https://junie.jetbrains.com/docs/environment-variables.html,rater1 | |
| junie,B,3800.1,D13,2,,install.sh verifies SHA-256; auto-update and update channels are configurable. Not 3: no signatures/SBOM.,[from-official-docs] https://github.com/JetBrains/junie,rater1 | |
| junie,B,3800.1,D14,1,,Config relocation knobs and enterprise sign-in only; no admin-enforced policy file.,[from-official-docs] https://junie.jetbrains.com/docs/junie-cli-configuration.html; [from-official-docs] https://junie.jetbrains.com/docs/junie-cli.html,rater1 | |
| junie,B,3800.1,D15,2,,"Documented reporting route (YouTrack 'Security Problem', 24h ack) and NVD CVEs; no bounty.",[from-official-docs] https://www.jetbrains.com/legal/docs/terms/coordinated-disclosure/; [from-official-docs] https://nvd.nist.gov/vuln/detail/CVE-2026-41153,rater1 | |
| kilo-code,A,7.8.10,D01,1,,"Base ruleset is '*': allow, so edits and other non-bash tools run without asking by default; bash asks but its benign allowlist includes cp/mv/touch; SECURITY.md calls permissions a UX feature.",[from-official-docs] https://github.com/Kilo-Org/kilocode/blob/main/packages/opencode/src/agent/agent.ts; [from-official-docs] https://github.com/Kilo-Org/kilocode/blob/main/packages/opencode/src/kilocode/agent/index.ts; [from-official-docs] https://github.com/Kilo-Org/kilocode/blob/main/packages/opencode/src/cli/cmd/run.ts; [from-official-docs] https://github.com/Kilo-Org/kilocode/blob/main/SECURITY.md,rater1 | |
| kilo-code,A,7.8.10,D02,1,,"bubblewrap/Seatbelt sandbox fails closed and project config cannot weaken it, but sandbox.enabled defaults to false.",[from-official-docs] https://github.com/Kilo-Org/kilocode/blob/main/packages/kilo-docs/pages/getting-started/settings/sandboxing.md; [from-official-docs] https://github.com/Kilo-Org/kilocode/blob/main/packages/kilo-sandbox/src/backend.ts,rater1 | |
| kilo-code,A,7.8.10,D03,1,,Host:port allowlist with default-deny network exists only when the opt-in sandbox is enabled.,[from-official-docs] https://github.com/Kilo-Org/kilocode/blob/main/packages/kilo-docs/pages/getting-started/settings/sandboxing.md,rater1 | |
| kilo-code,A,7.8.10,D04,1,,"No trust prompt, but project config is parsed as untrusted ({env:} rejected, {file:} confined, variable MCP headers dropped) and can be disabled by flag; project MCP entries still load without approval.",[from-official-docs] https://github.com/Kilo-Org/kilocode/blob/main/packages/opencode/src/config/config.ts; [from-official-docs] https://github.com/Kilo-Org/kilocode/blob/main/packages/core/src/flag/flag.ts; [from-official-docs] https://github.com/Kilo-Org/kilocode/blob/main/packages/kilo-docs/pages/automate/mcp/using-in-cli.md,rater1 | |
| kilo-code,A,7.8.10,D05,1,,"MCP tools fall under the permission globs (base allow), project MCP entries load without approval, and the org MCP allowlist is only a proposal.",[from-official-docs] https://github.com/Kilo-Org/kilocode/blob/main/packages/kilo-docs/pages/automate/mcp/using-in-cli.md; [from-official-docs] https://github.com/Kilo-Org/kilocode/blob/main/packages/kilo-docs/pages/contributing/features/enterprise-mcp-controls.md; [from-official-docs] https://github.com/Kilo-Org/kilocode/blob/main/packages/opencode/src/agent/agent.ts,rater1 | |
| kilo-code,A,7.8.10,D06,1,,npm/local plugins run as trusted host code; KILO_PURE disables them; no vetting or signing.,[from-official-docs] https://github.com/Kilo-Org/kilocode/blob/main/packages/kilo-docs/pages/automate/extending/plugins.md,rater1 | |
| kilo-code,A,7.8.10,D07,1,,"auth.json written 0600 and .env reads always prompt, but log retention is not documented (score-2 clause unmet).",[from-official-docs] https://github.com/Kilo-Org/kilocode/blob/main/packages/opencode/src/auth/index.ts; [from-official-docs] https://github.com/Kilo-Org/kilocode/blob/main/packages/kilo-docs/pages/customize/agent-permissions.md,rater1 | |
| kilo-code,A,7.8.10,D08,1,,"Non-interactive runs auto-reject permission requests unless --auto (safe), but the review integrations' token permissions and actor checks are not documented.",[from-official-docs] https://github.com/Kilo-Org/kilocode/blob/main/packages/kilo-docs/pages/code-with-ai/platforms/cli.md,rater1 | |
| kilo-code,A,7.8.10,D09,2,,"Memory save/recall tools default to 'ask', so each write is confirmed and can be denied by rule. Not 3: storage scope, scanning and central disable not documented.",[from-official-docs] https://github.com/Kilo-Org/kilocode/blob/main/packages/opencode/src/kilocode/agent/index.ts,rater1 | |
| kilo-code,A,7.8.10,D10,1,,"permission.task gates subagents and built-in read-only agents have a bash deny ceiling, but propagation of --auto/bypass to subagents is not documented.",[from-official-docs] https://github.com/Kilo-Org/kilocode/blob/main/packages/kilo-docs/pages/customize/agent-permissions.md,rater1 | |
| kilo-code,A,7.8.10,D11,2,,OTLP export of structured telemetry via OTEL_EXPORTER_OTLP_ENDPOINT alongside the local session store.,[from-official-docs] https://github.com/Kilo-Org/kilocode/blob/main/packages/kilo-docs/pages/code-with-ai/platforms/cli.md,rater1 | |
| kilo-code,A,7.8.10,D12,1,,Telemetry on by default with documented opt-out; privacy policy claims no code/prompt storage; retention not documented.,[from-official-docs] https://github.com/Kilo-Org/kilocode/blob/main/packages/kilo-docs/pages/code-with-ai/platforms/cli.md; [from-official-docs] https://github.com/Kilo-Org/kilocode/blob/main/PRIVACY.md,rater1 | |
| kilo-code,A,7.8.10,D13,1,,"npm, curl installer, brew, AUR with no documented signature or provenance.",[from-official-docs] https://github.com/Kilo-Org/kilocode/blob/main/README.md,rater1 | |
| kilo-code,A,7.8.10,D14,1,,"Managed config dir and MDM plist exist, but precedence over user/project config and covered controls are not documented.",[from-official-docs] https://github.com/Kilo-Org/kilocode/blob/main/packages/opencode/src/config/managed.ts,rater1 | |
| kilo-code,A,7.8.10,D15,1,,"SECURITY.md exists, but the two CVEs were filed by third parties via NVD with vendor non-response; no vendor advisories.",[from-official-docs] https://nvd.nist.gov/vuln/detail/CVE-2026-8765; [from-official-docs] https://nvd.nist.gov/vuln/detail/CVE-2025-11445,rater1 | |
| kimi-cli,A,2.2.0,D01,2,,Default manual mode asks for edits and commands; allow/deny/ask rules; yolo/auto are user flags and there is no project-level config file; static deny rules remain in auto. Not 3: no administrator control.,[from-official-docs] https://github.com/MoonshotAI/kimi-code/blob/main/docs/en/configuration/config-files.md; [from-official-docs] https://github.com/MoonshotAI/kimi-code/blob/main/docs/en/configuration/overrides.md; [from-official-docs] https://github.com/MoonshotAI/kimi-code/blob/main/docs/en/reference/kimi-command.md,rater1 | |
| kimi-cli,A,2.2.0,D02,N/A,N/A (not documented: absence of a sandbox is only unverified).,N/A (not documented: absence of a sandbox is only unverified).,,rater1 | |
| kimi-cli,A,2.2.0,D03,N/A,N/A (not documented: absence of egress control is only unverified).,N/A (not documented: absence of egress control is only unverified).,,rater1 | |
| kimi-cli,A,2.2.0,D04,2,,The only repo-level executable config (project MCP) is shown in a trust prompt and kept disabled in headless runs unless trusted; hooks and settings are user-level only. Not 3: no central folder pre-approval.,[from-official-docs] https://github.com/MoonshotAI/kimi-code/blob/main/docs/en/customization/mcp.md; [from-official-docs] https://github.com/MoonshotAI/kimi-code/blob/main/docs/en/configuration/overrides.md; [from-official-docs] https://github.com/MoonshotAI/kimi-code/blob/main/docs/en/customization/hooks.md,rater1 | |
| kimi-cli,A,2.2.0,D05,1,,"Project MCP is trust-gated and OAuth tokens are 0600, but 'Ask When Needed' auto-approves MCP calls and there is no managed list or change detection.",[from-official-docs] https://github.com/MoonshotAI/kimi-code/blob/main/docs/en/customization/mcp.md; [from-official-docs] https://github.com/MoonshotAI/kimi-code/blob/main/docs/en/configuration/data-locations.md; [from-official-docs] https://github.com/MoonshotAI/kimi-code/blob/main/docs/en/configuration/config-files.md,rater1 | |
| kimi-cli,A,2.2.0,D06,1,,"Marketplace installs can pin tag/commit and need explicit action, but no vetting or signing is documented; plugins add MCP servers and prompt instructions.",[from-official-docs] https://github.com/MoonshotAI/kimi-code/blob/main/docs/en/customization/plugins.md,rater1 | |
| kimi-cli,A,2.2.0,D07,1,,"Credentials dir 0700/files 0600 and sensitive-file reads ask, but plaintext api_key in config is permitted and log retention is not documented.",[from-official-docs] https://github.com/MoonshotAI/kimi-code/blob/main/docs/en/configuration/data-locations.md; [from-official-docs] https://github.com/MoonshotAI/kimi-code/blob/main/docs/en/configuration/config-files.md; [from-official-docs] https://github.com/MoonshotAI/kimi-code/blob/main/docs/en/guides/interaction.md,rater1 | |
| kimi-cli,A,2.2.0,D08,1,,kimi -p runs under the auto policy with no approvals (static deny rules still apply); no official CI integration.,[from-official-docs] https://github.com/MoonshotAI/kimi-code/blob/main/docs/en/reference/kimi-command.md,rater1 | |
| kimi-cli,A,2.2.0,D09,N/A,N/A (not documented: absence of memory is only unverified).,N/A (not documented: absence of memory is only unverified).,,rater1 | |
| kimi-cli,A,2.2.0,D10,1,,Built-in subagents run in isolated contexts; permission inheritance not documented.,[from-official-docs] https://github.com/MoonshotAI/kimi-code/blob/main/README.md,rater1 | |
| kimi-cli,A,2.2.0,D11,1,,"Per-session data dirs, diagnostic log and input history; no structured tool-call export or OTel.",[from-official-docs] https://github.com/MoonshotAI/kimi-code/blob/main/docs/en/configuration/data-locations.md,rater1 | |
| kimi-cli,A,2.2.0,D12,1,,Anonymous telemetry on by default with documented opt-out; retention not documented.,[from-official-docs] https://github.com/MoonshotAI/kimi-code/blob/main/docs/en/configuration/env-vars.md,rater1 | |
| kimi-cli,A,2.2.0,D13,1,,curl|bash installer without checksum; auto-install updates on by default.,[from-official-docs] https://github.com/MoonshotAI/kimi-code/blob/main/docs/en/configuration/data-locations.md; [from-official-docs] https://github.com/MoonshotAI/kimi-code/blob/main/README.md,rater1 | |
| kimi-cli,A,2.2.0,D14,0,,A single user-level config file and no admin layer; no enterprise offering documented.,[from-official-docs] https://github.com/MoonshotAI/kimi-code/blob/main/docs/en/configuration/overrides.md,rater1 | |
| kimi-cli,A,2.2.0,D15,1,,SECURITY.md with GHSA/email route; 0 published advisories.,[from-official-docs] https://github.com/MoonshotAI/kimi-code/blob/main/SECURITY.md; [from-official-docs] https://api.github.com/repos/MoonshotAI/kimi-code/security-advisories,rater1 | |
| kiro,B,IDE 1.3.0 / CLI 2.31.0,D01,1,,"Every command needs approval by default with allow/ask/deny capability rules, but approve-everything spellings include mcpServers.autoApprove ['*'] in workspace mcp.json (repo-controlled once trusted) and prefix-matched trustedCommands '*'. Patched stdin-pipe bypass CVE-2026-9255 noted.",[from-official-docs] https://kiro.dev/docs/permissions; [from-official-docs] https://kiro.dev/docs/privacy-and-security/; [from-official-docs] https://nvd.nist.gov/vuln/detail/CVE-2026-9255; [from-official-docs] https://kiro.dev/docs/mcp/configuration/,rater1 | |
| kiro,B,IDE 1.3.0 / CLI 2.31.0,D02,1,,Seatbelt/bubblewrap command sandbox is off by default and does not cover built-in file tools.,[from-official-docs] https://kiro.dev/docs/sandbox,rater1 | |
| kiro,B,IDE 1.3.0 / CLI 2.31.0,D03,2,,web_fetch needs approval by default (only fs_read and read-only commands are pre-allowed) and is admin-deniable; a sandbox network allowlist (DEFAULT_BLOCKED when enabled) exists but the sandbox is off by default.,[from-official-docs] https://kiro.dev/docs/sandbox; [from-official-docs] https://kiro.dev/docs/enterprise/governance/permissions/; [from-official-docs] https://kiro.dev/docs/permissions,rater1 | |
| kiro,B,IDE 1.3.0 / CLI 2.31.0,D04,1,,"Untrusted workspaces do not load custom agents, steering, MCP config or skills and ask before every shell/MCP call, but hooks are not in the documented list; three patched trust-bypass CVEs in 2026.",[from-official-docs] https://kiro.dev/docs/permissions; [from-official-docs] https://aws.amazon.com/security/security-bulletins/2026-074-aws/; [from-official-docs] https://aws.amazon.com/security/security-bulletins/2026-037-aws/,rater1 | |
| kiro,B,IDE 1.3.0 / CLI 2.31.0,D05,2,,"Workspace MCP config is trust-gated, disabledTools per tool, and an admin registry allowlist fails closed. Not 3: no change detection; admin controls client-side and not applied to Builder ID users. 3P mcp.json rewrite fixed in 0.11.x.",[from-official-docs] https://kiro.dev/docs/mcp/configuration/; [from-official-docs] https://kiro.dev/docs/enterprise/governance/mcp/; [third-party-report] https://thehackernews.com/2026/07/aws-kiro-flaw-let-poisoned-web-page.html; [from-official-docs] https://kiro.dev/docs/permissions,rater1 | |
| kiro,B,IDE 1.3.0 / CLI 2.31.0,D06,1,,"Powers and skills install with prompts on writes, but no vetting or signing is documented (only unverified).",[from-official-docs] https://kiro.dev/docs/permissions,rater1 | |
| kiro,B,IDE 1.3.0 / CLI 2.31.0,D07,1,,The agent can reach env vars and AWS credentials by default; secret-file deny rules are only examples; token storage undocumented.,[from-official-docs] https://kiro.dev/docs/privacy-and-security/; [from-official-docs] https://kiro.dev/docs/permissions,rater1 | |
| kiro,B,IDE 1.3.0 / CLI 2.31.0,D08,1,,"Headless treats ask as deny and --trust-tools takes exact ids, but there is only a sample workflow, no official CI integration with permissions/actor checks.",[from-official-docs] https://kiro.dev/docs/cli/headless/,rater1 | |
| kiro,B,IDE 1.3.0 / CLI 2.31.0,D09,1,,Global and workspace steering files plus an enabled CLI knowledge base; off switch and confirmation (outside untrusted workspaces) not documented.,[from-official-docs] https://kiro.dev/docs/permissions; [from-official-docs] https://kiro.dev/docs/reference/settings/,rater1 | |
| kiro,B,IDE 1.3.0 / CLI 2.31.0,D10,1,,Subagents with trustedAgents/availableAgents settings; isolation and inheritance not documented.,[from-official-docs] https://kiro.dev/docs/custom-agents/configuration-reference/,rater1 | |
| kiro,B,IDE 1.3.0 / CLI 2.31.0,D11,1,,"Enterprise prompt logging and usage reports to customer S3, but no structured tool-call log is documented (score-2 clause unmet).",[from-official-docs] https://kiro.dev/docs/privacy-and-security/data-protection/,rater1 | |
| kiro,B,IDE 1.3.0 / CLI 2.31.0,D12,1,,Free/individual tiers are opted in by default to telemetry and content collection (opt-out available); Enterprise opted out.,[from-official-docs] https://kiro.dev/docs/privacy-and-security/data-protection/,rater1 | |
| kiro,B,IDE 1.3.0 / CLI 2.31.0,D13,2,,Install script verifies SHA-256 from a manifest; auto-update can be disabled. Not 3: no signatures/SBOM.,[from-official-docs] https://cli.kiro.dev/install; [from-official-docs] https://kiro.dev/docs/reference/settings/,rater1 | |
| kiro,B,IDE 1.3.0 / CLI 2.31.0,D14,2,,"MDM-deployed managed-settings.json (deny/ask) read by IDE and CLI and an MCP registry toggle; client-side enforced and limited to deny/ask, not verified.",[from-official-docs] https://kiro.dev/docs/enterprise/governance/permissions/; [from-official-docs] https://kiro.dev/docs/enterprise/governance/mcp/,rater1 | |
| kiro,B,IDE 1.3.0 / CLI 2.31.0,D15,3,,"AWS security bulletins with CVEs and fixed versions, AWS VDP on HackerOne with safe harbour, security contact.",[from-official-docs] https://aws.amazon.com/security/security-bulletins/2026-074-aws/; [from-official-docs] https://aws.amazon.com/security/vulnerability-reporting/,rater1 | |
| langgraph,C,1.2.15,D01,N/A,N/A (not documented: the interrupt()/HITL fact is unverified).,N/A (not documented: the interrupt()/HITL fact is unverified).,,rater1 | |
| langgraph,C,1.2.15,D02,N/A,N/A (not documented: no sandbox facts collected).,N/A (not documented: no sandbox facts collected).,,rater1 | |
| langgraph,C,1.2.15,D03,N/A,N/A (not documented: no egress facts collected).,N/A (not documented: no egress facts collected).,,rater1 | |
| langgraph,C,1.2.15,D04,N/A,N/A (tier C library: loads no workspace configuration; server-auth advisories are a D16 topic).,N/A (tier C library: loads no workspace configuration; server-auth advisories are a D16 topic).,[from-official-docs] https://github.com/langchain-ai/langgraph/security/advisories,rater1 | |
| langgraph,C,1.2.15,D05,N/A,N/A (not documented: no MCP facts collected).,N/A (not documented: no MCP facts collected).,,rater1 | |
| langgraph,C,1.2.15,D06,N/A,N/A (not documented: no extension facts collected).,N/A (not documented: no extension facts collected).,,rater1 | |
| langgraph,C,1.2.15,D07,N/A,"N/A (not documented: the only fact concerns checkpoint deserialisation, not credential storage).","N/A (not documented: the only fact concerns checkpoint deserialisation, not credential storage).",[from-official-docs] https://github.com/langchain-ai/langgraph/blob/main/libs/checkpoint/langgraph/checkpoint/serde/jsonplus.py,rater1 | |
| langgraph,C,1.2.15,D08,N/A,N/A (not documented: no unattended-run facts collected).,N/A (not documented: no unattended-run facts collected).,,rater1 | |
| langgraph,C,1.2.15,D09,2,,"Checkpointer/Store persistence is opt-in and namespace-scoped. Not 3: no confirmation, scanning or central disable. Patched namespace CVE-2026-71433 noted.",[from-official-docs] https://github.com/langchain-ai/langgraph/security/advisories,rater1 | |
| langgraph,C,1.2.15,D10,N/A,N/A (not documented: no multi-agent scoping facts collected).,N/A (not documented: no multi-agent scoping facts collected).,,rater1 | |
| langgraph,C,1.2.15,D11,N/A,N/A (not documented: tracing fact is unverified).,N/A (not documented: tracing fact is unverified).,,rater1 | |
| langgraph,C,1.2.15,D12,N/A,N/A (not documented: no telemetry facts collected).,N/A (not documented: no telemetry facts collected).,,rater1 | |
| langgraph,C,1.2.15,D13,N/A,N/A (not documented: no distribution facts collected).,N/A (not documented: no distribution facts collected).,,rater1 | |
| langgraph,C,1.2.15,D14,N/A,N/A (not documented: no managed-policy facts collected).,N/A (not documented: no managed-policy facts collected).,,rater1 | |
| langgraph,C,1.2.15,D15,1,,"10+ GHSAs/CVEs with credited reporters, but no SECURITY.md (3P) or documented policy, so score 2 is not met.",[from-official-docs] https://github.com/langchain-ai/langgraph/security/advisories; [third-party-report] https://github.com/langchain-ai/langgraph/blob/main/SECURITY.md,rater1 | |
| langgraph-platform,C,langgraph 1.2.15 / langgraph-api 0.16.0,D01,N/A,"N/A (not documented: facts cover app-defined resource authorization, not tool-call approval).","N/A (not documented: facts cover app-defined resource authorization, not tool-call approval).",[from-official-docs] https://docs.langchain.com/langsmith/auth; [from-official-docs] https://github.com/advisories/GHSA-fvww-7h3r-vfhp,rater1 | |
| langgraph-platform,C,langgraph 1.2.15 / langgraph-api 0.16.0,D02,N/A,N/A (not documented: no sandbox facts collected).,N/A (not documented: no sandbox facts collected).,,rater1 | |
| langgraph-platform,C,langgraph 1.2.15 / langgraph-api 0.16.0,D03,1,,Tier C: loopback webhook delivery denied by default (URL validation); general egress is customer infrastructure.,[from-official-docs] https://github.com/advisories/GHSA-2c9q-c2q9-qgqv,rater1 | |
| langgraph-platform,C,langgraph 1.2.15 / langgraph-api 0.16.0,D04,N/A,N/A (server runtime; reads no repository-level configuration).,N/A (server runtime; reads no repository-level configuration).,,rater1 | |
| langgraph-platform,C,langgraph 1.2.15 / langgraph-api 0.16.0,D05,N/A,"N/A (not documented: MCP is application-level, no platform facts).","N/A (not documented: MCP is application-level, no platform facts).",,rater1 | |
| langgraph-platform,C,langgraph 1.2.15 / langgraph-api 0.16.0,D06,N/A,N/A (feature absent: no extension marketplace in the platform).,N/A (feature absent: no extension marketplace in the platform).,,rater1 | |
| langgraph-platform,C,langgraph 1.2.15 / langgraph-api 0.16.0,D07,1,,"Tier C: config encrypted at rest and client-side trace hiding/anonymiser, but no secret-manager or managed-identity integration documented.",[from-official-docs] https://github.com/advisories/GHSA-jfj5-wrj9-63x4; [from-official-docs] https://docs.langchain.com/langsmith/data-purging-compliance,rater1 | |
| langgraph-platform,C,langgraph 1.2.15 / langgraph-api 0.16.0,D08,N/A,N/A (not documented: no unattended-run enforcement facts).,N/A (not documented: no unattended-run enforcement facts).,,rater1 | |
| langgraph-platform,C,langgraph 1.2.15 / langgraph-api 0.16.0,D09,1,,Persistent threads and a per-user-scoped Store; visibility and off switch not documented. Patched checkpoint/store CVEs noted.,[from-official-docs] https://docs.langchain.com/langsmith/auth; [from-official-docs] https://github.com/advisories/GHSA-47pj-3jcm-6whg,rater1 | |
| langgraph-platform,C,langgraph 1.2.15 / langgraph-api 0.16.0,D10,N/A,N/A (not documented: multi-agent isolation is application-defined).,N/A (not documented: multi-agent isolation is application-defined).,,rater1 | |
| langgraph-platform,C,langgraph 1.2.15 / langgraph-api 0.16.0,D11,2,,Tier C: tracing to LangSmith with documented retention tiers.,[from-official-docs] https://docs.langchain.com/langsmith/data-purging-compliance,rater1 | |
| langgraph-platform,C,langgraph 1.2.15 / langgraph-api 0.16.0,D12,1,,"Regions and 14-day base retention documented, but the training-use default is not stated.",[from-official-docs] https://docs.langchain.com/langsmith/regions-faq; [from-official-docs] https://docs.langchain.com/langsmith/data-purging-compliance,rater1 | |
| langgraph-platform,C,langgraph 1.2.15 / langgraph-api 0.16.0,D13,1,,"PyPI/container distribution with no documented signing, provenance or SBOM.",[from-official-docs] https://docs.langchain.com/langsmith/deploy-to-self-hosted-overview.md,rater1 | |
| langgraph-platform,C,langgraph 1.2.15 / langgraph-api 0.16.0,D14,0,,No administrator-enforced settings documented; self-hosted deployments have no default authentication and auth must be custom-built.,[from-official-docs] https://docs.langchain.com/langsmith/auth,rater1 | |
| langgraph-platform,C,langgraph 1.2.15 / langgraph-api 0.16.0,D15,1,,"Regular GHSA/CVE issuance, but no SECURITY.md or policy and no bounty, so score 2 ('policy plus advisories') is not met.",[from-official-docs] https://github.com/langchain-ai/langgraph/security/advisories,rater1 | |
| langgraph-platform,C,langgraph 1.2.15 / langgraph-api 0.16.0,D16,0,,Self-hosted deployments have no default authentication; custom auth hooks required.,[from-official-docs] https://docs.langchain.com/langsmith/auth,rater1 | |
| manus,C,None,D01,N/A,N/A (only a third-party report; no documented approval model).,N/A (only a third-party report; no documented approval model).,[third-party-report] https://aurascape.ai/resources/auralabs-research/silentbridge-zero-click-agent-takeover-meta-manus/,rater1 | |
| manus,C,None,D02,2,,"Tier C: per-task isolated cloud VM by default (root inside); egress control not documented, so not 3. 3P: passwordless sudo and env secrets inside the VM (pre-mitigation).",[from-official-docs] https://manus.im/blog/manus-sandbox; [third-party-report] https://aurascape.ai/resources/auralabs-research/silentbridge-zero-click-agent-takeover-meta-manus/,rater1 | |
| manus,C,None,D03,N/A,N/A (only a third-party report; no documented egress facts).,N/A (only a third-party report; no documented egress facts).,[third-party-report] https://aurascape.ai/resources/auralabs-research/silentbridge-zero-click-agent-takeover-meta-manus/,rater1 | |
| manus,C,None,D04,N/A,N/A (SaaS; no local workspace configuration).,N/A (SaaS; no local workspace configuration).,,rater1 | |
| manus,C,None,D05,N/A,N/A (not documented: no MCP governance facts).,N/A (not documented: no MCP governance facts).,,rater1 | |
| manus,C,None,D06,N/A,N/A (not documented: no extension facts).,N/A (not documented: no extension facts).,,rater1 | |
| manus,C,None,D07,N/A,N/A (only a third-party report; no documented credential facts).,N/A (only a third-party report; no documented credential facts).,[third-party-report] https://aurascape.ai/resources/auralabs-research/silentbridge-zero-click-agent-takeover-meta-manus/,rater1 | |
| manus,C,None,D08,N/A,N/A (not documented: no unattended-run facts).,N/A (not documented: no unattended-run facts).,,rater1 | |
| manus,C,None,D09,N/A,N/A (not documented: no memory facts).,N/A (not documented: no memory facts).,,rater1 | |
| manus,C,None,D10,N/A,N/A (not documented: no sub-agent facts).,N/A (not documented: no sub-agent facts).,,rater1 | |
| manus,C,None,D11,N/A,N/A (not documented: no audit-log facts).,N/A (not documented: no audit-log facts).,,rater1 | |
| manus,C,None,D12,1,,Trust page claims certifications but is silent on retention and training (retention undocumented).,[from-official-docs] https://manus.im/security,rater1 | |
| manus,C,None,D13,N/A,N/A (SaaS; no distributable client in scope).,N/A (SaaS; no distributable client in scope).,,rater1 | |
| manus,C,None,D14,N/A,N/A (not documented: enterprise policy controls only in secondary sources).,N/A (not documented: enterprise policy controls only in secondary sources).,,rater1 | |
| manus,C,None,D15,N/A,N/A (only third-party reports; no documented policy or advisories).,N/A (only third-party reports; no documented policy or advisories).,[third-party-report] https://manus.im/security; [third-party-report] https://aurascape.ai/resources/auralabs-research/silentbridge-zero-click-agent-takeover-meta-manus/,rater1 | |
| microsoft-agent-framework,C,python 1.21.0 / dotnet 1.25.0,D01,1,,Tier C: plain tools default to never_require; the harness agent enables approval but with heuristic auto-approval by default (treated like classifier defaults: not a user prompt).,[from-official-docs] https://github.com/microsoft/agent-framework/blob/main/python/packages/core/agent_framework/_tools.py; [from-official-docs] https://github.com/microsoft/agent-framework/blob/main/python/samples/02-agents/harness/README.md,rater1 | |
| microsoft-agent-framework,C,python 1.21.0 / dotnet 1.25.0,D02,1,,Tier C: optional Hyperlight CodeAct sandbox; shell executor is builder-supplied.,[from-official-docs] https://github.com/microsoft/agent-framework/blob/main/python/packages/hyperlight/README.md,rater1 | |
| microsoft-agent-framework,C,python 1.21.0 / dotnet 1.25.0,D03,N/A,N/A (not documented: no egress facts collected).,N/A (not documented: no egress facts collected).,,rater1 | |
| microsoft-agent-framework,C,python 1.21.0 / dotnet 1.25.0,D04,N/A,N/A (tier C library: no workspace configuration facts).,N/A (tier C library: no workspace configuration facts).,,rater1 | |
| microsoft-agent-framework,C,python 1.21.0 / dotnet 1.25.0,D05,1,,Tier C: MCPTool approval_mode and allowed_tools exist; default-on approval not documented. Three MCP-path GHSAs patched Oct 2026 noted.,[from-official-docs] https://github.com/microsoft/agent-framework/security/advisories; [from-official-docs] https://github.com/microsoft/agent-framework/blob/main/python/packages/core/agent_framework/_mcp.py,rater1 | |
| microsoft-agent-framework,C,python 1.21.0 / dotnet 1.25.0,D06,N/A,N/A (not documented: no extension facts collected).,N/A (not documented: no extension facts collected).,,rater1 | |
| microsoft-agent-framework,C,python 1.21.0 / dotnet 1.25.0,D07,1,,"Tier C: sensitivity labelling exists, with a patched GHSA where declarative workflows forwarded Sensitive values to MCP servers; no secret-manager integration documented.",[from-official-docs] https://github.com/microsoft/agent-framework/security/advisories/GHSA-9f26-7mqc-9vc4,rater1 | |
| microsoft-agent-framework,C,python 1.21.0 / dotnet 1.25.0,D08,N/A,N/A (not documented: no unattended-run facts collected).,N/A (not documented: no unattended-run facts collected).,,rater1 | |
| microsoft-agent-framework,C,python 1.21.0 / dotnet 1.25.0,D09,2,,"File-based durable memory only when a memory_store is provided (off by default). Not 3: scoping, scanning and central disable not documented.",[from-official-docs] https://github.com/microsoft/agent-framework/blob/main/python/samples/02-agents/harness/README.md,rater1 | |
| microsoft-agent-framework,C,python 1.21.0 / dotnet 1.25.0,D10,N/A,N/A (not documented: background-agent facts are unverified).,N/A (not documented: background-agent facts are unverified).,,rater1 | |
| microsoft-agent-framework,C,python 1.21.0 / dotnet 1.25.0,D11,2,,Tier C: built-in OpenTelemetry module; sensitive content opt-in.,[from-official-docs] https://github.com/microsoft/agent-framework/blob/main/python/packages/core/agent_framework/observability.py,rater1 | |
| microsoft-agent-framework,C,python 1.21.0 / dotnet 1.25.0,D12,1,,User-agent feature telemetry on unless disabled by env; retention not documented.,[from-official-docs] https://github.com/microsoft/agent-framework/blob/main/python/packages/core/agent_framework/_telemetry.py,rater1 | |
| microsoft-agent-framework,C,python 1.21.0 / dotnet 1.25.0,D13,N/A,N/A (not documented: no distribution facts collected).,N/A (not documented: no distribution facts collected).,,rater1 | |
| microsoft-agent-framework,C,python 1.21.0 / dotnet 1.25.0,D14,N/A,N/A (not documented: no managed-policy facts collected).,N/A (not documented: no managed-policy facts collected).,,rater1 | |
| microsoft-agent-framework,C,python 1.21.0 / dotnet 1.25.0,D15,2,,MSRC SECURITY.md and GHSAs with credited reporters; bounty not documented in the facts.,[from-official-docs] https://github.com/microsoft/agent-framework/blob/main/SECURITY.md,rater1 | |
| mistral-vibe,A,2.26.1,D01,1,,The default agent accept-edits auto-approves file edits (score-1 clause: writes without asking); ask/plan agents and per-tool permissions exist; --auto-approve/--yolo are flags.,[from-official-docs] https://github.com/mistralai/mistral-vibe/blob/main/README.md; [from-official-docs] https://github.com/mistralai/mistral-vibe/blob/main/vibe/core/config/vibe_schema.py,rater1 | |
| mistral-vibe,A,2.26.1,D02,N/A,N/A (not documented: absence of a sandbox is only unverified).,N/A (not documented: absence of a sandbox is only unverified).,,rater1 | |
| mistral-vibe,A,2.26.1,D03,N/A,N/A (not documented: absence of egress control is only unverified).,N/A (not documented: absence of egress control is only unverified).,,rater1 | |
| mistral-vibe,A,2.26.1,D04,2,,"Folder trust gates project config, hooks, AGENTS.md, skills, tools and agents; automation needs an explicit --trust. Not 3: env/git-helper coverage and central folder policy not documented.",[from-official-docs] https://github.com/mistralai/mistral-vibe/blob/main/README.md; [from-official-docs] https://github.com/mistralai/mistral-vibe/blob/main/vibe/core/config/layers/project.py; [from-official-docs] https://github.com/mistralai/mistral-vibe/blob/main/vibe/cli/entrypoint.py,rater1 | |
| mistral-vibe,A,2.26.1,D05,1,,"Project MCP servers come from trust-gated config with disabled_tools globs, but there is no managed list or change detection.",[from-official-docs] https://github.com/mistralai/mistral-vibe/blob/main/README.md; [from-official-docs] https://github.com/mistralai/mistral-vibe/blob/main/vibe/core/config/layers/project.py,rater1 | |
| mistral-vibe,A,2.26.1,D06,1,,Custom tools/skills are Python run with agent privileges from trusted dirs; no vetting.,[from-official-docs] https://github.com/mistralai/mistral-vibe/blob/main/README.md,rater1 | |
| mistral-vibe,A,2.26.1,D07,1,,"Global .env created 0600 and OTel redaction on by default, but no keychain and no log retention documented (score-2 clause unmet).",[from-official-docs] https://github.com/mistralai/mistral-vibe/blob/main/vibe/setup/auth/api_key_persistence.py; [from-official-docs] https://github.com/mistralai/mistral-vibe/blob/main/README.md,rater1 | |
| mistral-vibe,A,2.26.1,D08,1,,"-p supports --enabled-tools narrowing and the official action runs without --trust, but the action's token permissions and actor checks are not documented and the -p default permission mode is not stated.",[from-official-docs] https://github.com/mistralai/mistral-vibe/blob/main/README.md; [from-official-docs] https://github.com/mistralai/mistral-vibe/blob/main/action.yml; [from-official-docs] https://github.com/mistralai/mistral-vibe/blob/main/vibe/cli/entrypoint.py,rater1 | |
| mistral-vibe,A,2.26.1,D09,N/A,N/A (not documented: absence of memory is only unverified).,N/A (not documented: absence of memory is only unverified).,,rater1 | |
| mistral-vibe,A,2.26.1,D10,1,,"Subagents inherit the parent's hooks and explore is read-only, but bypass propagation and general per-agent restrictions are not documented.",[from-official-docs] https://github.com/mistralai/mistral-vibe/blob/main/README.md,rater1 | |
| mistral-vibe,A,2.26.1,D11,2,,"Log file, session journal export via --output-dir and opt-in OTLP traces.",[from-official-docs] https://github.com/mistralai/mistral-vibe/blob/main/README.md,rater1 | |
| mistral-vibe,A,2.26.1,D12,1,,Telemetry on by default to api.mistral.ai with documented opt-out; retention not documented.,[from-official-docs] https://github.com/mistralai/mistral-vibe/blob/main/vibe/core/config/vibe_schema.py; [from-official-docs] https://github.com/mistralai/mistral-vibe/blob/main/README.md,rater1 | |
| mistral-vibe,A,2.26.1,D13,1,,curl|bash and PyPI with no checksum/signature; update checks prompt and can be disabled.,[from-official-docs] https://github.com/mistralai/mistral-vibe/blob/main/README.md,rater1 | |
| mistral-vibe,A,2.26.1,D14,N/A,N/A (not documented: absence of an admin layer is only unverified).,N/A (not documented: absence of an admin layer is only unverified).,,rater1 | |
| mistral-vibe,A,2.26.1,D15,1,,SECURITY.md with private email; 0 published advisories.,[from-official-docs] https://github.com/mistralai/mistral-vibe/blob/main/SECURITY.md; [from-official-docs] https://api.github.com/repos/mistralai/mistral-vibe/security-advisories,rater1 | |
| oh-my-pi,A,18.11.0,D01,1,,"always-ask and write modes and per-tool rules exist, but tools.approvalMode defaults (and falls back) to yolo, auto-approving read, write and exec; the eval tool bypasses bash deny patterns.",[from-official-docs] https://github.com/can1357/oh-my-pi/blob/main/docs/approval-mode.md; [from-official-docs] https://github.com/can1357/oh-my-pi/blob/main/packages/coding-agent/src/tools/approval.ts,rater1 | |
| oh-my-pi,A,18.11.0,D02,0,,No sandbox; bash patterns are documented as not containment.,[from-official-docs] https://github.com/can1357/oh-my-pi/blob/main/docs/approval-mode.md,rater1 | |
| oh-my-pi,A,18.11.0,D03,N/A,N/A (not documented: absence of egress control is only unverified).,N/A (not documented: absence of egress control is only unverified).,,rater1 | |
| oh-my-pi,A,18.11.0,D04,0,,"Source states OMP performs no project-trust gating; project settings, hooks, extensions and foreign-tool configs load unconditionally.",[from-official-docs] https://github.com/can1357/oh-my-pi/blob/main/packages/coding-agent/src/extensibility/extensions/types.ts; [from-official-docs] https://github.com/can1357/oh-my-pi/blob/main/docs/config-usage.md,rater1 | |
| oh-my-pi,A,18.11.0,D05,1,,MCP servers are auto-imported from many tools' project files without consent; user enabled/disabled server lists exist.,[from-official-docs] https://github.com/can1357/oh-my-pi/blob/main/docs/mcp-config.md,rater1 | |
| oh-my-pi,A,18.11.0,D06,1,,"Claude-compatible marketplace with in-process plugins (hooks, tools, MCP); exact --trusted-extension allowlist is opt-in; no signing or vetting.",[from-official-docs] https://github.com/can1357/oh-my-pi/blob/main/docs/marketplace.md; [from-official-docs] https://github.com/can1357/oh-my-pi/blob/main/docs/extension-loading.md,rater1 | |
| oh-my-pi,A,18.11.0,D07,1,,OAuth credentials in agent.db (mode not documented); secret obfuscation off by default.,[from-official-docs] https://github.com/can1357/oh-my-pi/blob/main/docs/secrets.md; [from-official-docs] https://github.com/can1357/oh-my-pi/blob/main/docs/environment-variables.md,rater1 | |
| oh-my-pi,A,18.11.0,D08,0,,omp -p runs under the yolo default with no prompts and no CI guidance.,[from-official-docs] https://github.com/can1357/oh-my-pi/blob/main/docs/cli-reference.md; [from-official-docs] https://github.com/can1357/oh-my-pi/blob/main/docs/approval-mode.md,rater1 | |
| oh-my-pi,A,18.11.0,D09,2,,"Memory backends are off by default and contents are visible via memory:// URLs. Not 3: agent-written without confirmation, no scanning or central disable.",[from-official-docs] https://github.com/can1357/oh-my-pi/blob/main/docs/memory.md,rater1 | |
| oh-my-pi,A,18.11.0,D10,1,,Task subagents exist; permission inheritance not documented.,[from-official-docs] https://github.com/can1357/oh-my-pi/blob/main/docs/settings.md,rater1 | |
| oh-my-pi,A,18.11.0,D11,2,,JSONL sessions and logs plus optional OTLP export.,[from-official-docs] https://github.com/can1357/oh-my-pi/blob/main/docs/session.md; [from-official-docs] https://github.com/can1357/oh-my-pi/blob/main/docs/environment-variables.md,rater1 | |
| oh-my-pi,A,18.11.0,D12,1,,"OTLP only when configured, but a persistent install-id and hostname are sent to provider/gateway endpoints with no documented opt-out or retention.",[from-official-docs] https://github.com/can1357/oh-my-pi/blob/main/docs/install-id.md; [from-official-docs] https://github.com/can1357/oh-my-pi/blob/main/docs/environment-variables.md,rater1 | |
| oh-my-pi,A,18.11.0,D13,2,,Installer checks SHA-256 of the prebuilt binary and --ref pins a release; macOS signing documented. Not 3: no provenance/SBOM.,[from-official-docs] https://github.com/can1357/oh-my-pi/blob/main/README.md; [from-official-docs] https://github.com/can1357/oh-my-pi/blob/main/docs/macos-signing-notarization.md,rater1 | |
| oh-my-pi,A,18.11.0,D14,N/A,N/A (not documented: absence of an admin layer is only unverified).,N/A (not documented: absence of an admin layer is only unverified).,,rater1 | |
| oh-my-pi,A,18.11.0,D15,1,,SECURITY.md with private email/GHSA route; 0 advisories.,[from-official-docs] https://github.com/can1357/oh-my-pi/blob/main/.github/SECURITY.md; [from-official-docs] https://api.github.com/repos/can1357/oh-my-pi/security-advisories,rater1 | |
| openai-agentkit,C,Agent Builder: scheduled shutdown 2026-11-30; ChatKit (chatkit-python/js Apache-2.0); openai-guardrails-python (MIT),D01,2,,"Tier C: approval is required by default before data goes to a connector or remote MCP server, with a human-approval node in Agent Builder. Not 3: platform-enforced approval policy not documented.",[from-official-docs] https://developers.openai.com/api/docs/guides/tools-connectors-mcp; [from-official-docs] https://developers.openai.com/api/docs/guides/agent-builder-safety,rater1 | |
| openai-agentkit,C,Agent Builder: scheduled shutdown 2026-11-30; ChatKit (chatkit-python/js Apache-2.0); openai-guardrails-python (MIT),D02,N/A,"N/A (hosted visual builder; code execution happens in Responses API tools or customer-run exported SDK code, not documented here).","N/A (hosted visual builder; code execution happens in Responses API tools or customer-run exported SDK code, not documented here).",[from-official-docs] https://developers.openai.com/api/docs/guides/agent-builder,rater1 | |
| openai-agentkit,C,Agent Builder: scheduled shutdown 2026-11-30; ChatKit (chatkit-python/js Apache-2.0); openai-guardrails-python (MIT),D03,N/A,N/A (not documented in the pages read).,N/A (not documented in the pages read).,,rater1 | |
| openai-agentkit,C,Agent Builder: scheduled shutdown 2026-11-30; ChatKit (chatkit-python/js Apache-2.0); openai-guardrails-python (MIT),D04,N/A,N/A (no local workspace configuration).,N/A (no local workspace configuration).,,rater1 | |
| openai-agentkit,C,Agent Builder: scheduled shutdown 2026-11-30; ChatKit (chatkit-python/js Apache-2.0); openai-guardrails-python (MIT),D05,2,,"Tier C: remote MCP calls need approval by default and OAuth token handling is documented; the vendor states it does not verify remote servers, and Connector Registry governance is not documented.",[from-official-docs] https://developers.openai.com/api/docs/guides/tools-connectors-mcp,rater1 | |
| openai-agentkit,C,Agent Builder: scheduled shutdown 2026-11-30; ChatKit (chatkit-python/js Apache-2.0); openai-guardrails-python (MIT),D06,N/A,N/A (not documented).,N/A (not documented).,,rater1 | |
| openai-agentkit,C,Agent Builder: scheduled shutdown 2026-11-30; ChatKit (chatkit-python/js Apache-2.0); openai-guardrails-python (MIT),D07,1,,Tier C: connectors take OAuth tokens passed per request by the customer; no vault-style isolation documented.,[from-official-docs] https://developers.openai.com/api/docs/guides/tools-connectors-mcp; [from-official-docs] https://openai.github.io/openai-guardrails-python/,rater1 | |
| openai-agentkit,C,Agent Builder: scheduled shutdown 2026-11-30; ChatKit (chatkit-python/js Apache-2.0); openai-guardrails-python (MIT),D08,N/A,N/A (hosted builder; no unattended-run facts).,N/A (hosted builder; no unattended-run facts).,,rater1 | |
| openai-agentkit,C,Agent Builder: scheduled shutdown 2026-11-30; ChatKit (chatkit-python/js Apache-2.0); openai-guardrails-python (MIT),D09,N/A,N/A (not documented).,N/A (not documented).,,rater1 | |
| openai-agentkit,C,Agent Builder: scheduled shutdown 2026-11-30; ChatKit (chatkit-python/js Apache-2.0); openai-guardrails-python (MIT),D10,N/A,N/A (not documented).,N/A (not documented).,,rater1 | |
| openai-agentkit,C,Agent Builder: scheduled shutdown 2026-11-30; ChatKit (chatkit-python/js Apache-2.0); openai-guardrails-python (MIT),D11,2,,Tier C: per-node execution traces and trace graders; API data logged 30 days with store=true.,[from-official-docs] https://developers.openai.com/api/docs/guides/agent-builder; [from-official-docs] https://developers.openai.com/api/docs/guides/tools-connectors-mcp,rater1 | |
| openai-agentkit,C,Agent Builder: scheduled shutdown 2026-11-30; ChatKit (chatkit-python/js Apache-2.0); openai-guardrails-python (MIT),D12,1,,"30-day retention with a ZDR option, but the training-use default is not stated in the facts.",[from-official-docs] https://developers.openai.com/api/docs/guides/tools-connectors-mcp,rater1 | |
| openai-agentkit,C,Agent Builder: scheduled shutdown 2026-11-30; ChatKit (chatkit-python/js Apache-2.0); openai-guardrails-python (MIT),D13,N/A,N/A (not documented: packaging provenance not checked).,N/A (not documented: packaging provenance not checked).,,rater1 | |
| openai-agentkit,C,Agent Builder: scheduled shutdown 2026-11-30; ChatKit (chatkit-python/js Apache-2.0); openai-guardrails-python (MIT),D14,N/A,N/A (not documented: Connector Registry governance page not retrievable).,N/A (not documented: Connector Registry governance page not retrievable).,,rater1 | |
| openai-agentkit,C,Agent Builder: scheduled shutdown 2026-11-30; ChatKit (chatkit-python/js Apache-2.0); openai-guardrails-python (MIT),D15,1,,security@openai.com contact for MCP abuse; no advisories for the AgentKit components; 3P Guardrails bypasses without CVE.,[from-official-docs] https://developers.openai.com/api/docs/guides/tools-connectors-mcp; [third-party-report] https://www.helpnetsecurity.com/2025/11/03/zenity-openai-runtime-protection/,rater1 | |
| openai-agents-sdk,C,0.24.0,D01,1,,Tier C: needs_approval defaults False and MCP require_approval defaults None; approval callbacks are opt-in.,[from-official-docs] https://github.com/openai/openai-agents-python/blob/main/src/agents/tool.py; [from-official-docs] https://github.com/openai/openai-agents-python/blob/main/docs/guardrails.md,rater1 | |
| openai-agents-sdk,C,0.24.0,D02,1,,Tier C: Docker and hosted sandbox clients are optional; the local client adds no confinement.,[from-official-docs] https://github.com/openai/openai-agents-python/blob/main/docs/sandbox_agents.md,rater1 | |
| openai-agents-sdk,C,0.24.0,D03,N/A,N/A (not documented: no egress facts collected).,N/A (not documented: no egress facts collected).,,rater1 | |
| openai-agents-sdk,C,0.24.0,D04,N/A,N/A (tier C library: loads no workspace configuration).,N/A (tier C library: loads no workspace configuration).,,rater1 | |
| openai-agents-sdk,C,0.24.0,D05,1,,Tier C: require_approval and tool_filter exist but default to no approval.,[from-official-docs] https://github.com/openai/openai-agents-python/blob/main/docs/mcp.md,rater1 | |
| openai-agents-sdk,C,0.24.0,D06,N/A,N/A (not documented: no extension facts collected).,N/A (not documented: no extension facts collected).,,rater1 | |
| openai-agents-sdk,C,0.24.0,D07,1,,Tier C: keys from the environment; sensitive tool data included in exported traces by default.,[from-official-docs] https://github.com/openai/openai-agents-python/blob/main/src/agents/run_config.py; [from-official-docs] https://github.com/openai/openai-agents-python/blob/main/src/agents/_config.py,rater1 | |
| openai-agents-sdk,C,0.24.0,D08,N/A,N/A (not documented: no unattended-run facts collected).,N/A (not documented: no unattended-run facts collected).,,rater1 | |
| openai-agents-sdk,C,0.24.0,D09,N/A,N/A (not documented: no memory facts collected).,N/A (not documented: no memory facts collected).,,rater1 | |
| openai-agents-sdk,C,0.24.0,D10,1,,Tier C: handoff input filters exist (opt-in context scoping); permission scoping between agents not documented.,[from-official-docs] https://github.com/openai/openai-agents-python/blob/main/src/agents/extensions/handoff_filters.py,rater1 | |
| openai-agents-sdk,C,0.24.0,D11,2,,"Tier C: built-in tracing of generations, tool calls, handoffs and guardrails on by default.",[from-official-docs] https://github.com/openai/openai-agents-python/blob/main/docs/tracing.md,rater1 | |
| openai-agents-sdk,C,0.24.0,D12,1,,Tracing of prompts and tool I/O to OpenAI is on by default with an env opt-out.,[from-official-docs] https://github.com/openai/openai-agents-python/blob/main/src/agents/tracing/processors.py,rater1 | |
| openai-agents-sdk,C,0.24.0,D13,N/A,N/A (not documented: no distribution facts collected).,N/A (not documented: no distribution facts collected).,,rater1 | |
| openai-agents-sdk,C,0.24.0,D14,N/A,N/A (not documented: no managed-policy facts collected).,N/A (not documented: no managed-policy facts collected).,,rater1 | |
| openai-agents-sdk,C,0.24.0,D15,1,,SECURITY.md with CVD policy; no published advisories.,[from-official-docs] https://github.com/openai/openai-agents-python/blob/main/SECURITY.md,rater1 | |
| openclaw,A,2026.10.3,D01,1,,"Rich deny/allowlist/ask model, but the shipped default for host exec is security=full with ask=off. Patched allowlist and approval bypass GHSAs noted.",[from-official-docs] https://github.com/openclaw/openclaw/blob/main/docs/tools/exec-approvals.md; [from-official-docs] https://github.com/openclaw/openclaw/blob/main/docs/gateway/permission-modes.md; [from-official-docs] https://github.com/openclaw/openclaw/security/advisories/GHSA-gv46-4xfq-jv58,rater1 | |
| openclaw,A,2026.10.3,D02,1,,Docker/Podman/SSH sandbox backends exist but sandboxing is off by default; Gateway always on host. Patched sandbox-escape GHSA noted.,[from-official-docs] https://github.com/openclaw/openclaw/blob/main/docs/gateway/sandboxing.md; [from-official-docs] https://github.com/openclaw/openclaw/security/advisories/GHSA-9p3r-hh9g-5cmg,rater1 | |
| openclaw,A,2026.10.3,D03,1,,Secret egress proxy and traffic allowlist are documented as default-off. Patched SSRF GHSAs noted.,[from-official-docs] https://github.com/openclaw/openclaw/blob/main/docs/gateway/secrets.md; [from-official-docs] https://api.github.com/repos/openclaw/openclaw/security-advisories,rater1 | |
| openclaw,A,2026.10.3,D04,0,,"No repository trust concept (trust is gateway auth, pairing and scopes, a D16 topic); patched CVE showed cwd .env injection enabling config takeover.",[from-official-docs] https://github.com/openclaw/openclaw/blob/main/docs/gateway/config-gateway.md; [from-official-docs] https://github.com/openclaw/openclaw/security/advisories/GHSA-8rh7-6779-cjqq,rater1 | |
| openclaw,A,2026.10.3,D05,1,,"Per-server approval modes (approve/prompt/auto) and per-tool grants, but no managed allow/deny list or change detection.",[from-official-docs] https://github.com/openclaw/openclaw/blob/main/docs/tools/mcp.md,rater1 | |
| openclaw,A,2026.10.3,D06,1,,"ClawHub registry with a 'skills verify' trust envelope (would support 2), lowered by a third-party report of hundreds of malicious skills on the registry; patched plugin-install path traversal noted.",[from-official-docs] https://github.com/openclaw/openclaw/blob/main/docs/tools/skills.md; [from-official-docs] https://github.com/openclaw/openclaw/security/advisories/GHSA-qrq5-wjgg-rvqw; [third-party-report] https://thehackernews.com/2026/02/researchers-find-341-malicious-clawhub.html,rater1 | |
| openclaw,A,2026.10.3,D07,1,,700/600 permissions are only recommended and plaintext credentials remain agent-readable; patched token-leak GHSA; 3P mass exposure of misconfigured servers.,[from-official-docs] https://github.com/openclaw/openclaw/blob/main/docs/gateway/security/secrets-and-storage.md; [third-party-report] https://forklog.com/en/critical-vulnerabilities-found-in-clawdbot-ai-agent-for-cryptocurrency-theft/; [from-official-docs] https://github.com/openclaw/openclaw/security/advisories/GHSA-g8p2-7wf7-98mq,rater1 | |
| openclaw,A,2026.10.3,D08,1,,"agent exec is a CI entry with ephemeral state, pinned config and --auth-env-only, but inherits host exec security=full/ask=off and has no official CI action.",[from-official-docs] https://github.com/openclaw/openclaw/blob/main/docs/cli/agent.md; [from-official-docs] https://github.com/openclaw/openclaw/blob/main/docs/tools/exec-approvals.md,rater1 | |
| openclaw,A,2026.10.3,D09,1,,"Agent-written MEMORY.md and daily notes in the workspace are fed back into prompts; flush runs are append-only, but no confirmation or off switch is documented.",[from-official-docs] https://github.com/openclaw/openclaw/blob/main/docs/concepts/memory.md; [from-official-docs] https://github.com/openclaw/openclaw/blob/main/SECURITY.md,rater1 | |
| openclaw,A,2026.10.3,D10,1,,Sub-run inheritance is not documented beyond a patched critical heartbeat-inheritance sandbox bypass.,[from-official-docs] https://github.com/openclaw/openclaw/security/advisories/GHSA-g5cg-8x5w-7jpm,rater1 | |
| openclaw,A,2026.10.3,D11,2,,Metadata-only durable audit ledger plus OpenTelemetry/Prometheus exporters. Not 3: policy enforcement and admin-change coverage not documented.,[from-official-docs] https://github.com/openclaw/openclaw/blob/main/docs/gateway/audit.md; [from-official-docs] https://github.com/openclaw/openclaw/blob/main/docs/gateway/opentelemetry.md,rater1 | |
| openclaw,A,2026.10.3,D12,1,,"Daily update-check request by default (metadata only), feature stats opt-in; retention not documented.",[from-official-docs] https://github.com/openclaw/openclaw/blob/main/docs/gateway/telemetry.md,rater1 | |
| openclaw,A,2026.10.3,D13,1,,"npm with release dist-tags and explicit --allow-scripts; no signature, checksum or provenance documented.",[from-official-docs] https://registry.npmjs.org/openclaw; [from-official-docs] https://github.com/openclaw/openclaw/blob/main/README.md,rater1 | |
| openclaw,A,2026.10.3,D14,1,,"Gateway operator roles and scopes exist, but no managed-settings layer with documented precedence; repeated scope-escalation advisories.",[from-official-docs] https://github.com/openclaw/openclaw/blob/main/docs/gateway/config-gateway.md; [from-official-docs] https://github.com/openclaw/openclaw/security/advisories/GHSA-hf68-49fm-59cq,rater1 | |
| openclaw,A,2026.10.3,D15,2,,SECURITY.md with private GHSA/email and 722 published advisories (39 with CVE); no paid bounty or safe harbour documented.,[from-official-docs] https://api.github.com/repos/openclaw/openclaw/security-advisories; [from-official-docs] https://github.com/openclaw/openclaw/blob/main/SECURITY.md,rater1 | |
| openclaw,A,2026.10.3,D16,1,,"Gateway binds loopback and requires auth by default (token generated at onboarding), but Origin/Host validation is not documented and a patched Control UI token-leak plus repeated scope-escalation advisories exist.",[from-official-docs] https://github.com/openclaw/openclaw/blob/main/docs/gateway/config-gateway.md; [from-official-docs] https://github.com/openclaw/openclaw/security/advisories/GHSA-g8p2-7wf7-98mq; [from-official-docs] https://github.com/openclaw/openclaw/security/advisories/GHSA-hf68-49fm-59cq,rater1 | |
| opencode,A,1.19.0,D01,1,,"Default ruleset is '*': allow (verified), so shell and edits run without asking; project config can change rules; --auto/--yolo flags.",[from-official-docs] https://github.com/anomalyco/opencode/blob/v1.19.0/packages/opencode/src/agent/agent.ts; [from-official-docs] https://github.com/anomalyco/opencode/blob/v1.19.0/packages/web/src/content/docs/permissions.mdx; [from-official-docs] https://github.com/anomalyco/opencode/blob/v1.19.0/packages/opencode/src/cli/cmd/run.ts,rater1 | |
| opencode,A,1.19.0,D02,0,,SECURITY.md: OpenCode does not sandbox the agent.,[from-official-docs] https://github.com/anomalyco/opencode/blob/v1.19.0/SECURITY.md,rater1 | |
| opencode,A,1.19.0,D03,0,,webfetch/websearch allowed by the default ruleset; no egress control.,[from-official-docs] https://github.com/anomalyco/opencode/blob/v1.19.0/packages/opencode/src/agent/agent.ts; [from-official-docs] https://github.com/anomalyco/opencode/blob/v1.19.0/SECURITY.md,rater1 | |
| opencode,A,1.19.0,D04,0,,"Project opencode.json, .opencode plugins and MCP definitions load with no trust prompt (verified in the loader); malicious config is out of scope.",[from-official-docs] https://github.com/anomalyco/opencode/blob/v1.19.0/packages/opencode/src/config/config.ts; [from-official-docs] https://github.com/anomalyco/opencode/blob/v1.19.0/packages/web/src/content/docs/plugins.mdx; [from-official-docs] https://github.com/anomalyco/opencode/blob/v1.19.0/SECURITY.md,rater1 | |
| opencode,A,1.19.0,D05,1,,Project MCP servers start without approval; managed config can override entries; no allowlist or change detection.,[from-official-docs] https://github.com/anomalyco/opencode/blob/v1.19.0/packages/web/src/content/docs/mcp-servers.mdx; [from-official-docs] https://github.com/anomalyco/opencode/blob/v1.19.0/packages/web/src/content/docs/config.mdx,rater1 | |
| opencode,A,1.19.0,D06,0,,"npm plugins listed in config (including project config) are installed automatically at startup and project .opencode/plugins/*.ts auto-load, with no review.",[from-official-docs] https://github.com/anomalyco/opencode/blob/v1.19.0/packages/web/src/content/docs/plugins.mdx,rater1 | |
| opencode,A,1.19.0,D07,1,,"auth.json written 0600 (verified) and .env reads ask, but log retention and redaction are not documented.",[from-official-docs] https://github.com/anomalyco/opencode/blob/v1.19.0/packages/opencode/src/auth/index.ts; [from-official-docs] https://github.com/anomalyco/opencode/blob/v1.19.0/packages/opencode/src/agent/agent.ts,rater1 | |
| opencode,A,1.19.0,D08,1,,Headless runs inherit the allow-all default with --auto/--yolo available; no official CI integration. Patched unauthenticated-server CVE-2026-22812 noted.,[from-official-docs] https://github.com/anomalyco/opencode/blob/v1.19.0/packages/opencode/src/cli/cmd/run.ts; [from-official-docs] https://github.com/anomalyco/opencode/blob/v1.19.0/packages/opencode/src/agent/agent.ts; [from-official-docs] https://github.com/anomalyco/opencode/security/advisories/GHSA-vxw4-wv6m-9hhh,rater1 | |
| opencode,A,1.19.0,D09,N/A,N/A (feature absent: no cross-session agent memory; AGENTS.md is user-authored).,N/A (feature absent: no cross-session agent memory; AGENTS.md is user-authored).,[from-official-docs] https://github.com/anomalyco/opencode/blob/v1.19.0/packages/web/src/content/docs/cli.mdx,rater1 | |
| opencode,A,1.19.0,D10,1,,"Subagents inherit parent deny and external_directory rules and have their own rulesets (verified), but propagation of --auto to subagents is not documented.",[from-official-docs] https://github.com/anomalyco/opencode/blob/v1.19.0/packages/opencode/src/agent/subagent-permissions.ts,rater1 | |
| opencode,A,1.19.0,D11,1,,Sessions in local SQLite; no export or audit log.,[from-official-docs] https://github.com/anomalyco/opencode/blob/v1.19.0/packages/opencode/src/auth/index.ts,rater1 | |
| opencode,A,1.19.0,D12,1,,"Share is manual by default, but auto-share publishes conversations publicly and retention is not documented.",[from-official-docs] https://github.com/anomalyco/opencode/blob/v1.19.0/packages/web/src/content/docs/share.mdx,rater1 | |
| opencode,A,1.19.0,D13,1,,curl|bash installer without checksum or signature (verified) and autoupdate on by default.,[from-official-docs] https://github.com/anomalyco/opencode/blob/v1.19.0/install,rater1 | |
| opencode,A,1.19.0,D14,2,,Managed files and macOS MDM merged last and override user/project config (precedence verified). Not 3: no sandbox or telemetry controls exist to cover.,[from-official-docs] https://github.com/anomalyco/opencode/blob/v1.19.0/packages/opencode/src/config/config.ts; [from-official-docs] https://github.com/anomalyco/opencode/blob/v1.19.0/packages/web/src/content/docs/config.mdx,rater1 | |
| opencode,A,1.19.0,D15,2,,SECURITY.md with GHSA intake; 3 GHSAs (2 with CVE); no bounty.,[from-official-docs] https://github.com/anomalyco/opencode/blob/v1.19.0/SECURITY.md; [from-official-docs] https://github.com/anomalyco/opencode/security/advisories/GHSA-vxw4-wv6m-9hhh; [from-official-docs] https://github.com/anomalyco/opencode/security/advisories/GHSA-c83v-7274-4vgp; [from-official-docs] https://github.com/anomalyco/opencode/security/advisories/GHSA-632h-h47v-g4x4,rater1 | |
| opencode,A,1.19.0,D16,1,,"opencode serve is opt-in and protected only if OPENCODE_SERVER_PASSWORD is set; patched unauthenticated-RCE, XSS and cross-site upgrade advisories.",[from-official-docs] https://github.com/anomalyco/opencode/blob/v1.19.0/SECURITY.md; [from-official-docs] https://github.com/anomalyco/opencode/security/advisories/GHSA-vxw4-wv6m-9hhh; [from-official-docs] https://github.com/anomalyco/opencode/security/advisories/GHSA-632h-h47v-g4x4; [from-official-docs] https://github.com/anomalyco/opencode/security/advisories/GHSA-c83v-7274-4vgp,rater1 | |
| openhands,A,Agent Canvas 1.27.1 / SDK 1.54.0 / CLI 1.16.0,D01,1,,"TUI defaults to AlwaysConfirm, but the only controls are coarse confirmation policies (always/never/risk threshold) with no allow/deny rules; headless and --yolo disable confirmation.",[from-official-docs] https://github.com/OpenHands/OpenHands-CLI/blob/main/openhands_cli/tui/textual_app.py; [from-official-docs] https://github.com/OpenHands/software-agent-sdk/blob/main/openhands-sdk/openhands/sdk/security/confirmation_policy.py,rater1 | |
| openhands,A,Agent Canvas 1.27.1 / SDK 1.54.0 / CLI 1.16.0,D02,1,,Docker runtime is optional; the quickstart runs the agent-server on the host with full filesystem access.,[from-official-docs] https://github.com/OpenHands/OpenHands/blob/main/README.md,rater1 | |
| openhands,A,Agent Canvas 1.27.1 / SDK 1.54.0 / CLI 1.16.0,D03,0,,No outbound policy; only inbound firewall guidance.,[from-official-docs] https://github.com/OpenHands/OpenHands/blob/main/docs/SELF_HOSTING.md,rater1 | |
| openhands,A,Agent Canvas 1.27.1 / SDK 1.54.0 / CLI 1.16.0,D04,0,,Project .openhands/hooks.json is loaded from the work dir on CLI start with no trust prompt.,[from-official-docs] https://github.com/OpenHands/software-agent-sdk/blob/main/openhands-sdk/openhands/sdk/hooks/config.py,rater1 | |
| openhands,A,Agent Canvas 1.27.1 / SDK 1.54.0 / CLI 1.16.0,D05,1,,User mcp.json with enable/disable commands; skills/plugins can bundle MCP servers; no allowlist or change detection.,[from-official-docs] https://github.com/OpenHands/OpenHands-CLI/blob/main/README.md,rater1 | |
| openhands,A,Agent Canvas 1.27.1 / SDK 1.54.0 / CLI 1.16.0,D06,1,,Plugins discovered from user and project dirs run plugin code; no vetting or signing.,[from-official-docs] https://github.com/OpenHands/software-agent-sdk/blob/main/openhands-sdk/openhands/sdk/plugin/discovery.py,rater1 | |
| openhands,A,Agent Canvas 1.27.1 / SDK 1.54.0 / CLI 1.16.0,D07,1,,CLI persists the LLM api_key in agent_settings.json with expose_secrets=True; Fernet cipher only optional.,[from-official-docs] https://github.com/OpenHands/OpenHands-CLI/blob/main/openhands_cli/stores/agent_store.py; [from-official-docs] https://github.com/OpenHands/software-agent-sdk/blob/main/openhands-sdk/openhands/sdk/utils/cipher.py,rater1 | |
| openhands,A,Agent Canvas 1.27.1 / SDK 1.54.0 / CLI 1.16.0,D08,0,,Headless mode uses NeverConfirm (auto-approves everything) and the agent-server is unsecured without a session key; no untrusted-input guidance.,[from-official-docs] https://github.com/OpenHands/software-agent-sdk/blob/main/openhands-agent-server/openhands/agent_server/config.py; [from-official-docs] https://github.com/OpenHands/OpenHands-CLI/blob/main/openhands_cli/tui/textual_app.py,rater1 | |
| openhands,A,Agent Canvas 1.27.1 / SDK 1.54.0 / CLI 1.16.0,D09,1,,Project and user MEMORY.md injected into the system prompt; no confirmation or off switch documented.,[from-official-docs] https://github.com/OpenHands/software-agent-sdk/blob/main/openhands-sdk/openhands/sdk/context/memory.py,rater1 | |
| openhands,A,Agent Canvas 1.27.1 / SDK 1.54.0 / CLI 1.16.0,D10,1,,Delegate and subagent tools exist; inheritance not documented.,[from-official-docs] https://github.com/OpenHands/software-agent-sdk/blob/main/openhands-tools/openhands/tools/delegate,rater1 | |
| openhands,A,Agent Canvas 1.27.1 / SDK 1.54.0 / CLI 1.16.0,D11,1,,Local persisted conversations; no structured export documented in the facts.,[from-official-docs] https://github.com/OpenHands/OpenHands-CLI/blob/main/openhands_cli/locations.py,rater1 | |
| openhands,A,Agent Canvas 1.27.1 / SDK 1.54.0 / CLI 1.16.0,D12,1,,"Agent-server telemetry requires recorded consent (opt-in), but retention is not documented.",[from-official-docs] https://github.com/OpenHands/software-agent-sdk/blob/main/openhands-agent-server/openhands/agent_server/telemetry/policy.py,rater1 | |
| openhands,A,Agent Canvas 1.27.1 / SDK 1.54.0 / CLI 1.16.0,D13,1,,"curl installer with no documented checksum, signature or provenance.",[from-official-docs] https://github.com/OpenHands/OpenHands-CLI/blob/main/README.md,rater1 | |
| openhands,A,Agent Canvas 1.27.1 / SDK 1.54.0 / CLI 1.16.0,D14,N/A,N/A (not documented: no managed-policy facts collected).,N/A (not documented: no managed-policy facts collected).,,rater1 | |
| openhands,A,Agent Canvas 1.27.1 / SDK 1.54.0 / CLI 1.16.0,D15,1,,"A GHSA/CVE was published, but no SECURITY.md or policy is documented, and a third-party CVE's original issue was deleted.",[from-official-docs] https://github.com/OpenHands/OpenHands/security/advisories; [from-official-docs] https://nvd.nist.gov/vuln/detail/CVE-2026-19022,rater1 | |
| openhands,A,Agent Canvas 1.27.1 / SDK 1.54.0 / CLI 1.16.0,D16,0,,Empty session_api_keys means the server is unsecured; only bind/firewall guidance.,[from-official-docs] https://github.com/OpenHands/software-agent-sdk/blob/main/openhands-agent-server/openhands/agent_server/config.py; [from-official-docs] https://github.com/OpenHands/OpenHands/blob/main/docs/SELF_HOSTING.md,rater1 | |
| openhands-sdk,C,1.54.0,D01,1,,"Tier C: AlwaysConfirm/ConfirmRisky policies and security analyzers exist, but ConversationState defaults to NeverConfirm.",[from-official-docs] https://github.com/OpenHands/software-agent-sdk/blob/main/openhands-sdk/openhands/sdk/security/confirmation_policy.py; [from-official-docs] https://github.com/OpenHands/software-agent-sdk/blob/main/openhands-sdk/openhands/sdk/security/analyzer.py,rater1 | |
| openhands-sdk,C,1.54.0,D02,1,,Tier C: Docker/Apptainer/remote workspaces optional; default LocalWorkspace runs on the host.,[from-official-docs] https://github.com/OpenHands/software-agent-sdk/blob/main/openhands-sdk/openhands/sdk/conversation/conversation.py,rater1 | |
| openhands-sdk,C,1.54.0,D03,N/A,N/A (not documented: absence of egress control is only unverified).,N/A (not documented: absence of egress control is only unverified).,,rater1 | |
| openhands-sdk,C,1.54.0,D04,0,,"Runtime helper loaders read repo .openhands/hooks.json (shell command hooks), skills and AGENTS.md as always-active context with no consent step.",[from-official-docs] https://github.com/OpenHands/software-agent-sdk/blob/main/openhands-sdk/openhands/sdk/hooks/config.py,rater1 | |
| openhands-sdk,C,1.54.0,D05,1,,Tier C: MCP config supplied in code with OAuth support; only the generic confirmation policy (default NeverConfirm) gates calls.,[from-official-docs] https://github.com/OpenHands/software-agent-sdk/blob/main/openhands-sdk/openhands/sdk/mcp/config.py; [from-official-docs] https://github.com/OpenHands/software-agent-sdk/blob/main/openhands-sdk/openhands/sdk/security/confirmation_policy.py,rater1 | |
| openhands-sdk,C,1.54.0,D06,1,,Tier C: built-in skills/plugins marketplace modules with an installed-skills dir; no vetting or signing documented.,[from-official-docs] https://github.com/OpenHands/software-agent-sdk/blob/main/openhands-sdk/openhands/sdk/skills/installed.py,rater1 | |
| openhands-sdk,C,1.54.0,D07,1,,"Tier C: SecretRegistry masks secrets in observations, but conversation secrets are stored unencrypted unless a Cipher is configured.",[from-official-docs] https://github.com/OpenHands/software-agent-sdk/blob/main/openhands-sdk/openhands/sdk/conversation/state.py,rater1 | |
| openhands-sdk,C,1.54.0,D08,N/A,"N/A (not documented: the in-dimension fact concerns agent-server authentication, scored under D16).","N/A (not documented: the in-dimension fact concerns agent-server authentication, scored under D16).",[from-official-docs] https://github.com/OpenHands/software-agent-sdk/blob/main/openhands-agent-server/openhands/agent_server/config.py,rater1 | |
| openhands-sdk,C,1.54.0,D09,N/A,N/A (not documented: absence of memory is only unverified).,N/A (not documented: absence of memory is only unverified).,,rater1 | |
| openhands-sdk,C,1.54.0,D10,1,,"Tier C: subagents inherit the parent permission_mode unless overridden, so a permissive parent mode propagates.",[from-official-docs] https://github.com/OpenHands/software-agent-sdk/blob/main/openhands-sdk/openhands/sdk/subagent/schema.py,rater1 | |
| openhands-sdk,C,1.54.0,D11,2,,Tier C: conversation events persisted by default; Laminar tracing one setting away.,[from-official-docs] https://github.com/OpenHands/software-agent-sdk/blob/main/openhands-sdk/openhands/sdk/observability/laminar.py,rater1 | |
| openhands-sdk,C,1.54.0,D12,N/A,N/A (no vendor telemetry unless the user configures a Laminar key; rubric: libraries with no telemetry are N/A).,N/A (no vendor telemetry unless the user configures a Laminar key; rubric: libraries with no telemetry are N/A).,[from-official-docs] https://github.com/OpenHands/software-agent-sdk/blob/main/openhands-sdk/openhands/sdk/observability/laminar.py,rater1 | |
| openhands-sdk,C,1.54.0,D13,N/A,N/A (not documented: provenance not checked; only an unverified fact).,N/A (not documented: provenance not checked; only an unverified fact).,,rater1 | |
| openhands-sdk,C,1.54.0,D14,N/A,N/A (not documented: only an unverified fact).,N/A (not documented: only an unverified fact).,,rater1 | |
| openhands-sdk,C,1.54.0,D15,0,,No SECURITY.md and no GHSAs for the SDK repository.,[from-official-docs] https://api.github.com/repos/OpenHands/software-agent-sdk/security-advisories,rater1 | |
| openhands-sdk,C,1.54.0,D16,0,,agent-server with no session API key is unsecured on 127.0.0.1 (reachable with no authentication).,[from-official-docs] https://github.com/OpenHands/software-agent-sdk/blob/main/openhands-agent-server/openhands/agent_server/config.py,rater1 | |
| pi,A,1.2.0,D01,0,,No per-action approval exists by design; bash/edit/write run unprompted; only whole-tool selection flags.,[from-official-docs] https://github.com/earendil-works/pi/blob/main/packages/coding-agent/docs/security.md; [from-official-docs] https://github.com/earendil-works/pi/blob/main/packages/coding-agent/docs/cli.md,rater1 | |
| pi,A,1.2.0,D02,0,,No sandbox; docs direct users to containers/VMs.,[from-official-docs] https://github.com/earendil-works/pi/blob/main/packages/coding-agent/docs/security.md,rater1 | |
| pi,A,1.2.0,D03,0,,No egress control; PI_OFFLINE affects only Pi's own requests.,[from-official-docs] https://github.com/earendil-works/pi/blob/main/packages/coding-agent/docs/environment-variables.md,rater1 | |
| pi,A,1.2.0,D04,2,,"Project trust (default 'ask') gates .pi settings, MCP, extensions, skills, prompts and system-prompt files; non-interactive modes skip them unless configured. Not 3: env/git-helper coverage and central policy not documented. Patched CVE-2026-54325 noted.",[from-official-docs] https://github.com/earendil-works/pi/blob/main/packages/coding-agent/docs/security.md; [from-official-docs] https://github.com/earendil-works/pi/security/advisories/GHSA-mqxh-6gq7-558m,rater1 | |
| pi,A,1.2.0,D05,1,,"Project MCP read only after trust with per-server/tool exposure controls, but no managed allowlist or change detection.",[from-official-docs] https://github.com/earendil-works/pi/blob/main/packages/coding-agent/docs/mcp.md,rater1 | |
| pi,A,1.2.0,D06,1,,"Packages are pinnable and project packages trust-gated, but extensions run in-process with no vetting or signing.",[from-official-docs] https://github.com/earendil-works/pi/blob/main/packages/coding-agent/docs/packages.md; [from-official-docs] https://github.com/earendil-works/pi/security/advisories/GHSA-jfgx-wxx8-mp94,rater1 | |
| pi,A,1.2.0,D07,1,,"auth.json 0600 in a 0700 dir (source), but sessions may contain credentials with no redaction or retention documented. Patched write-race CVE noted.",[from-official-docs] https://github.com/earendil-works/pi/blob/main/packages/coding-agent/src/core/auth-storage.ts; [from-official-docs] https://github.com/earendil-works/pi/security/advisories/GHSA-r95r-rj6r-c39x; [from-official-docs] https://github.com/earendil-works/pi/blob/main/packages/coding-agent/docs/security.md,rater1 | |
| pi,A,1.2.0,D08,1,,"Print/JSON/RPC modes skip untrusted project resources, but tools run unprompted and there is no official CI integration.",[from-official-docs] https://github.com/earendil-works/pi/blob/main/packages/coding-agent/docs/security.md,rater1 | |
| pi,A,1.2.0,D09,N/A,N/A (not documented: absence of memory is only unverified).,N/A (not documented: absence of memory is only unverified).,,rater1 | |
| pi,A,1.2.0,D10,N/A,N/A (not documented: absence of sub-agents is only unverified).,N/A (not documented: absence of sub-agents is only unverified).,,rater1 | |
| pi,A,1.2.0,D11,1,,JSONL sessions saved by default; no telemetry export (score-2 clause unmet).,[from-official-docs] https://github.com/earendil-works/pi/blob/main/packages/coding-agent/docs/sessions.md; [from-official-docs] https://github.com/earendil-works/pi/security/advisories/GHSA-7v5m-pr3q-6453,rater1 | |
| pi,A,1.2.0,D12,1,,Anonymous install/update telemetry on by default with env opt-outs; retention not documented.,[from-official-docs] https://github.com/earendil-works/pi/blob/main/packages/coding-agent/docs/settings.md,rater1 | |
| pi,A,1.2.0,D13,1,,"Installer pins dependencies and npm route uses --ignore-scripts, but no checksum, signature or provenance is documented.",[from-official-docs] https://github.com/earendil-works/pi/blob/main/packages/coding-agent/docs/quickstart.md,rater1 | |
| pi,A,1.2.0,D14,N/A,N/A (not documented: absence of an admin layer is only unverified).,N/A (not documented: absence of an admin layer is only unverified).,,rater1 | |
| pi,A,1.2.0,D15,2,,SECURITY.md with email/GHSA route and four GHSAs with CVEs; no bounty.,[from-official-docs] https://github.com/earendil-works/pi/blob/main/SECURITY.md; [from-official-docs] https://api.github.com/repos/earendil-works/pi/security-advisories,rater1 | |
| pydantic-ai,C,2.56.0,D01,1,,Tier C: deferred-tool approval primitive exists but is opt-in per tool; nothing is gated by default.,[from-official-docs] https://github.com/pydantic/pydantic-ai/blob/main/docs/deferred-tools.md; [from-official-docs] https://github.com/pydantic/pydantic-ai/blob/main/docs/tools.md,rater1 | |
| pydantic-ai,C,2.56.0,D02,1,,Tier C: Bubblewrap/Modal/E2B sandboxes are opt-in add-ons; LocalWorkspace runs on the host.,[from-official-docs] https://github.com/pydantic/pydantic-ai/blob/main/docs/workspace.md; [from-official-docs] https://github.com/pydantic/pydantic-ai/blob/main/docs/harness/bubblewrap-sandbox.md; [from-official-docs] https://github.com/pydantic/pydantic-ai/blob/main/docs/harness/shell.md,rater1 | |
| pydantic-ai,C,2.56.0,D03,2,,Tier C: built-in web_fetch has domain allow/block lists and an SSRF private-IP/metadata blocklist by default. Five patched bypass CVEs recorded (rubric: patched bypasses do not change the score).,[from-official-docs] https://github.com/pydantic/pydantic-ai/security/advisories; [from-official-docs] https://github.com/pydantic/pydantic-ai/blob/main/docs/harness/bubblewrap-sandbox.md,rater1 | |
| pydantic-ai,C,2.56.0,D04,N/A,N/A (library: no repository-level config is auto-loaded; MCP config only via an explicit path).,N/A (library: no repository-level config is auto-loaded; MCP config only via an explicit path).,[from-official-docs] https://github.com/pydantic/pydantic-ai/blob/main/docs/cli.md,rater1 | |
| pydantic-ai,C,2.56.0,D05,1,,Tier C: MCP config only by explicit path; approval only via the generic opt-in deferred-tool primitive; no allowlist or change detection.,[from-official-docs] https://github.com/pydantic/pydantic-ai/blob/main/docs/mcp/client.md; [from-official-docs] https://github.com/pydantic/pydantic-ai/blob/main/docs/deferred-tools.md,rater1 | |
| pydantic-ai,C,2.56.0,D06,N/A,N/A (feature absent: no marketplace or registry; capabilities are plain Python packages).,N/A (feature absent: no marketplace or registry; capabilities are plain Python packages).,[from-official-docs] https://github.com/pydantic/pydantic-ai/blob/main/docs/harness/index.md,rater1 | |
| pydantic-ai,C,2.56.0,D07,1,,Tier C: no credential store (application responsibility); env scrubbing for shell and content-free OTel mode exist (redaction gaps patched).,[from-official-docs] https://github.com/pydantic/pydantic-ai/security/advisories; [from-official-docs] https://github.com/pydantic/pydantic-ai/blob/main/docs/harness/shell.md,rater1 | |
| pydantic-ai,C,2.56.0,D08,N/A,N/A (not documented: only an unverified fact).,N/A (not documented: only an unverified fact).,,rater1 | |
| pydantic-ai,C,2.56.0,D09,2,,"Opt-in Markdown memory notebook with namespaces chosen by application code. Not 3: no confirmation, provenance or central disable.",[from-official-docs] https://github.com/pydantic/pydantic-ai/blob/main/docs/harness/memory.md,rater1 | |
| pydantic-ai,C,2.56.0,D10,N/A,N/A (not documented: sub-agent facts are unverified).,N/A (not documented: sub-agent facts are unverified).,,rater1 | |
| pydantic-ai,C,2.56.0,D11,2,,Tier C: built-in OpenTelemetry GenAI instrumentation one setting away (instrument=True).,[from-official-docs] https://github.com/pydantic/pydantic-ai/blob/main/docs/logfire.md,rater1 | |
| pydantic-ai,C,2.56.0,D12,N/A,N/A (library sends nothing unless the developer configures Logfire/OTel; rubric: no telemetry -> N/A).,N/A (library sends nothing unless the developer configures Logfire/OTel; rubric: no telemetry -> N/A).,[from-official-docs] https://github.com/pydantic/pydantic-ai/blob/main/docs/logfire.md,rater1 | |
| pydantic-ai,C,2.56.0,D13,N/A,N/A (not documented: provenance not checked; only an unverified fact).,N/A (not documented: provenance not checked; only an unverified fact).,,rater1 | |
| pydantic-ai,C,2.56.0,D14,N/A,N/A (not documented: only an unverified fact).,N/A (not documented: only an unverified fact).,,rater1 | |
| pydantic-ai,C,2.56.0,D15,2,,Org SECURITY.md with private reporting and 17 GHSAs with CVEs; no bounty.,[from-official-docs] https://api.github.com/advisories?affects=pydantic-ai&ecosystem=pip,rater1 | |
| qoder,B,CLI 1.1.71 (docs/npm; not executed),D01,1,,"Default mode asks for sensitive actions with allow/ask/deny (deny first), but non-default modes including bypass_permissions take effect from settings in trusted directories, i.e. approve-everything can come from a trusted repo's project settings.",[from-official-docs] https://docs.qoder.com/en/cli/permissions,rater1 | |
| qoder,B,CLI 1.1.71 (docs/npm; not executed),D02,1,,"No sandbox in the CLI; only the IDE 'Experts Mode' sandbox, not on by default for the product.",[from-official-docs] https://docs.qoder.com/user-guide/quest/terminal-and-sandbox.md; [from-official-docs] https://docs.qoder.com/qoder/approval-and-sandbox.md,rater1 | |
| qoder,B,CLI 1.1.71 (docs/npm; not executed),D03,1,,"Desktop asks before internet access, but there is no domain allowlist or egress policy; CLI covers only proxy settings.",[from-official-docs] https://docs.qoder.com/cli/network.md; [from-official-docs] https://docs.qoder.com/qoder/approval-and-sandbox.md,rater1 | |
| qoder,B,CLI 1.1.71 (docs/npm; not executed),D04,2,,"Folder trust on by default: untrusted cwd loads only user config (project settings with hooks/MCP/permissions ignored) and project memory only in trusted workspaces. Not 3: headless, env and central folder policy not documented.",[from-official-docs] https://docs.qoder.com/cli/settings.md; [from-official-docs] https://docs.qoder.com/cli/memory.md,rater1 | |
| qoder,B,CLI 1.1.71 (docs/npm; not executed),D05,1,,"mcp.allowed/excluded and project-server enable lists exist, but no approval step for project .mcp.json is documented and trust:true skips tool confirmation.",[from-official-docs] https://docs.qoder.com/cli/settings-reference; [from-official-docs] https://docs.qoder.com/cli/mcp-reference.md,rater1 | |
| qoder,B,CLI 1.1.71 (docs/npm; not executed),D06,1,,Built-in plugin marketplace; users told to review scripts; no vetting or signing.,[from-official-docs] https://docs.qoder.com/release-notes/qoder-cli.md; [from-official-docs] https://docs.qoder.com/qoder/security-and-authorization.md,rater1 | |
| qoder,B,CLI 1.1.71 (docs/npm; not executed),D07,N/A,N/A (not documented: credential storage location and format undocumented).,N/A (not documented: credential storage location and format undocumented).,[from-official-docs] https://docs.qoder.com/cli/settings-reference,rater1 | |
| qoder,B,CLI 1.1.71 (docs/npm; not executed),D08,1,,"Print mode denies anything that would prompt, but the official action's token permissions and actor checks are not documented.",[from-official-docs] https://docs.qoder.com/cli/run-in-scripts; [from-official-docs] https://docs.qoder.com/cli/qoder-action.md,rater1 | |
| qoder,B,CLI 1.1.71 (docs/npm; not executed),D09,2,,Auto-memory is opt-in markdown saved locally with env toggles. Not 3: scanning and central disable not documented.,[from-official-docs] https://docs.qoder.com/cli/memory.md,rater1 | |
| qoder,B,CLI 1.1.71 (docs/npm; not executed),D10,1,,Subagents declaring bypass are downgraded only when disableYoloMode=true; inheritance otherwise undocumented.,[from-official-docs] https://docs.qoder.com/en/cli/permissions,rater1 | |
| qoder,B,CLI 1.1.71 (docs/npm; not executed),D11,1,,Session persistence only; no audit log or OTel export found.,[from-official-docs] https://docs.qoder.com/cli/cli-reference,rater1 | |
| qoder,B,CLI 1.1.71 (docs/npm; not executed),D12,1,,Usage and AI-code statistics on by default with a privacy-mode opt-out; retention and training not documented.,[from-official-docs] https://docs.qoder.com/cli/settings-reference; [third-party-report] https://tostracker.app/document/qodercom-privacy,rater1 | |
| qoder,B,CLI 1.1.71 (docs/npm; not executed),D13,1,,curl|bash installer with auto-update on by default; no integrity verification documented.,[from-official-docs] https://docs.qoder.com/cli/installation.md,rater1 | |
| qoder,B,CLI 1.1.71 (docs/npm; not executed),D14,1,,"Policy keys (allowManagedPermissionRulesOnly, disableYoloMode) exist but the managed file path/MDM delivery and precedence are undocumented.",[from-official-docs] https://docs.qoder.com/en/cli/permissions,rater1 | |
| qoder,B,CLI 1.1.71 (docs/npm; not executed),D15,N/A,N/A (not documented: only an unverified fact).,N/A (not documented: only an unverified fact).,,rater1 | |
| qwen-code,A,0.25.1,D01,1,,"Schema default approval mode is AUTO (LLM classifier approves 'safe' actions without asking; verified), so the default runs shell without asking the user; allow/ask/deny rules and --yolo exist.",[from-official-docs] https://github.com/QwenLM/qwen-code/blob/main/packages/cli/src/config/settingsSchema.ts,rater1 | |
| qwen-code,A,0.25.1,D02,1,,"bwrap/Landlock execution sandbox (workspace-write, network closed) is opt-in; repo settings cannot change it.",[from-official-docs] https://github.com/QwenLM/qwen-code/blob/main/docs/users/features/sandbox.md,rater1 | |
| qwen-code,A,0.25.1,D03,1,,Network 'closed' only inside the opt-in execution sandbox; no domain allowlist.,[from-official-docs] https://github.com/QwenLM/qwen-code/blob/main/docs/users/features/sandbox.md,rater1 | |
| qwen-code,A,0.25.1,D04,1,,"Folder trust exists but is off by default in code (verified), so project settings, .env and extensions load; some keys are operator-scope only.",[from-official-docs] https://github.com/QwenLM/qwen-code/blob/main/packages/cli/src/config/trustedFolders.ts; [from-official-docs] https://github.com/QwenLM/qwen-code/blob/main/docs/users/configuration/settings.md,rater1 | |
| qwen-code,A,0.25.1,D05,1,,Project MCP servers are gated only if folder trust is enabled (off by default); no managed list or change detection.,[from-official-docs] https://github.com/QwenLM/qwen-code/blob/main/docs/users/features/mcp.md; [from-official-docs] https://github.com/QwenLM/qwen-code/blob/main/packages/cli/src/config/trustedFolders.ts,rater1 | |
| qwen-code,A,0.25.1,D06,1,,"Extensions, skills and project hooks load from the workspace; HTTP hook URL allowlists are operator-only; no vetting.",[from-official-docs] https://github.com/QwenLM/qwen-code/blob/main/docs/users/features/hooks.md; [from-official-docs] https://github.com/QwenLM/qwen-code/blob/main/docs/users/configuration/settings.md,rater1 | |
| qwen-code,A,0.25.1,D07,1,,API keys may be stored plaintext in settings.json env; no keychain or redaction documented.,[from-official-docs] https://github.com/QwenLM/qwen-code/blob/main/docs/users/configuration/auth.md,rater1 | |
| qwen-code,A,0.25.1,D08,1,,"Official action and qwen -p exist, but headless inherits the AUTO classifier default and the action's permissions/actor checks are not documented.",[from-official-docs] https://github.com/QwenLM/qwen-code-action; [from-official-docs] https://github.com/QwenLM/qwen-code/blob/main/packages/cli/src/config/settingsSchema.ts,rater1 | |
| qwen-code,A,0.25.1,D09,1,,"Auto-memory plus git-tracked, secret-scanned team memory; off switch and write confirmation not documented.",[from-official-docs] https://github.com/QwenLM/qwen-code/blob/main/docs/users/configuration/settings.md,rater1 | |
| qwen-code,A,0.25.1,D10,N/A,N/A (not documented: no sub-agent facts collected).,N/A (not documented: no sub-agent facts collected).,,rater1 | |
| qwen-code,A,0.25.1,D11,2,,"Opt-in OpenTelemetry export with prompt logging opt-in, plus local session storage.",[from-official-docs] https://github.com/QwenLM/qwen-code/blob/main/docs/users/configuration/settings.md,rater1 | |
| qwen-code,A,0.25.1,D12,1,,Usage statistics on by default with opt-out; retention not documented.,[from-official-docs] https://github.com/QwenLM/qwen-code/blob/main/docs/users/configuration/settings.md,rater1 | |
| qwen-code,A,0.25.1,D13,1,,"curl|bash from an OSS bucket, npm and Homebrew; no verification documented.",[from-official-docs] https://github.com/QwenLM/qwen-code/blob/main/README.md,rater1 | |
| qwen-code,A,0.25.1,D14,2,,System settings and system-defaults files with operator-only keys a repo cannot set; coverage of MCP/telemetry and verification missing.,[from-official-docs] https://github.com/QwenLM/qwen-code/blob/main/docs/users/configuration/settings.md,rater1 | |
| qwen-code,A,0.25.1,D15,1,,SECURITY.md routes to an Aliyun portal; no published advisories.,[from-official-docs] https://github.com/QwenLM/qwen-code/blob/main/SECURITY.md,rater1 | |
| qwen-code,A,0.25.1,D16,0,,qwen serve binds 127.0.0.1:4170 with bearer auth off by default; any local process can drive code execution.,[from-official-docs] https://github.com/QwenLM/qwen-code/blob/main/docs/users/qwen-serve.md,rater1 | |
| raven-evermind,C,0.2.4,D01,1,,"Tier C: default permissions.mode is 'smart' (an LLM reviewer approves without asking the user), treated like classifier defaults; per-tool allow/ask/deny and a builtin deny list that holds in every mode exist; headless refuses ask-tier calls.",[from-official-docs] https://github.com/EverMind-AI/Raven/blob/main/docs-site/docs/permissions.md,rater1 | |
| raven-evermind,C,0.2.4,D02,1,,"Tier C: boxlite microVM is opt-in (backend none by default, host execution, network allowed).",[from-official-docs] https://github.com/EverMind-AI/Raven/blob/main/docs-site/docs/sandbox.md; [from-official-docs] https://github.com/EverMind-AI/Raven/blob/main/raven/config/schema.py,rater1 | |
| raven-evermind,C,0.2.4,D03,2,,Tier C: fetch SSRF guard with redirect re-validation and DNS pinning on by default; sandbox domain allowlist exists but egress is open by default.,[from-official-docs] https://github.com/EverMind-AI/Raven/blob/main/raven/security/network.py,rater1 | |
| raven-evermind,C,0.2.4,D04,N/A,N/A (tier C: config is user-global only; no repository-scoped configuration is read).,N/A (tier C: config is user-global only; no repository-scoped configuration is read).,[from-official-docs] https://github.com/EverMind-AI/Raven/blob/main/raven/security/trust.py,rater1 | |
| raven-evermind,C,0.2.4,D05,2,,Tier C: unknown MCP tools default to ask and auth modes (api key/OAuth) are documented. Not 3: no platform tool allowlist.,[from-official-docs] https://github.com/EverMind-AI/Raven/blob/main/raven/config/schema.py,rater1 | |
| raven-evermind,C,0.2.4,D06,1,,Built-in SkillHub and plugin market plus self-evolving tools/skills; no vetting documented.,[from-official-docs] https://github.com/EverMind-AI/Raven/blob/main/README.md,rater1 | |
| raven-evermind,C,0.2.4,D07,0,,"Provider api_key stored in ~/.raven/config.json, which the source itself describes as world-readable (score-0 clause); MCP tokens in a credentials dir; redaction only in ACP/bug reports.",[from-official-docs] https://github.com/EverMind-AI/Raven/blob/main/raven/config/schema.py,rater1 | |
| raven-evermind,C,0.2.4,D08,2,,Tier C: unattended runs fail closed on approvals and tool tiers are enforceable; documented gap: DAG confirm:true fails open without an ask channel.,[from-official-docs] https://github.com/EverMind-AI/Raven/blob/main/docs-site/docs/permissions.md,rater1 | |
| raven-evermind,C,0.2.4,D09,1,,Agent-writable cross-session memory and self-adopted improvements; no confirmation or off switch documented.,[from-official-docs] https://github.com/EverMind-AI/Raven/blob/main/README.md,rater1 | |
| raven-evermind,C,0.2.4,D10,1,,"Deny rules reach Raven agents, but ask tier and mode are not carried to third-party backends and the ACP client auto-prefers allow.",[from-official-docs] https://github.com/EverMind-AI/Raven/blob/main/docs-site/docs/permissions.md,rater1 | |
| raven-evermind,C,0.2.4,D11,2,,Tier C: structured tracing spans carry permission decisions; trajectory store and local logs.,[from-official-docs] https://github.com/EverMind-AI/Raven/blob/main/docs-site/docs/permissions.md,rater1 | |
| raven-evermind,C,0.2.4,D12,N/A,N/A (no telemetry code; bug-report upload only when a URL is configured).,N/A (no telemetry code; bug-report upload only when a URL is configured).,[from-official-docs] https://github.com/EverMind-AI/Raven/blob/main/docs-site/docs/trajectory-debugging.md,rater1 | |
| raven-evermind,C,0.2.4,D13,1,,curl|bash installer verifies only the Node tarball; no signature or provenance for the Raven package.,[from-official-docs] https://github.com/EverMind-AI/Raven/blob/main/install.sh,rater1 | |
| raven-evermind,C,0.2.4,D14,N/A,N/A (not documented: only an unverified fact).,N/A (not documented: only an unverified fact).,,rater1 | |
| raven-evermind,C,0.2.4,D15,1,,SECURITY.md with private GitHub reporting; no published advisories.,[from-official-docs] https://github.com/EverMind-AI/Raven/blob/main/SECURITY.md,rater1 | |
| raven-evermind,C,0.2.4,D16,1,,A2A server is off by default and uses a shared bearer token when on; Origin/Host validation not documented.,[from-official-docs] https://github.com/EverMind-AI/Raven/blob/main/docs-site/docs/permissions.md,rater1 | |
| replit-agent,B,cloud service (no client version),D01,0,,No per-action approval: Build mode applies changes directly; Plan mode and checkpoints are not per-action approval.,[from-official-docs] https://docs.replit.com/help/agent-and-ai.md,rater1 | |
| replit-agent,B,cloud service (no client version),D02,N/A,N/A (not documented: isolation details are unverified).,N/A (not documented: isolation details are unverified).,,rater1 | |
| replit-agent,B,cloud service (no client version),D03,N/A,N/A (not documented: no egress facts).,N/A (not documented: no egress facts).,,rater1 | |
| replit-agent,B,cloud service (no client version),D04,N/A,N/A (SaaS; no local workspace configuration facts).,N/A (SaaS; no local workspace configuration facts).,,rater1 | |
| replit-agent,B,cloud service (no client version),D05,1,,"Curated/custom MCP servers configured in the UI with only trust advice; no approval, allowlist or change detection documented.",[from-official-docs] https://docs.replit.com/chat/connect-through-mcp.md,rater1 | |
| replit-agent,B,cloud service (no client version),D06,1,,Skills and integrations exist; no vetting documented.,[from-official-docs] https://docs.replit.com/chat/agent-skills.md,rater1 | |
| replit-agent,B,cloud service (no client version),D07,1,,"A Secrets store is recommended, but agent access to secrets, redaction and retention are not documented.",[from-official-docs] https://docs.replit.com/learn/security-checklist.md,rater1 | |
| replit-agent,B,cloud service (no client version),D08,N/A,N/A (no headless or CI mode documented).,N/A (no headless or CI mode documented).,,rater1 | |
| replit-agent,B,cloud service (no client version),D09,N/A,N/A (not documented).,N/A (not documented).,,rater1 | |
| replit-agent,B,cloud service (no client version),D10,N/A,N/A (not documented: only an unverified fact).,N/A (not documented: only an unverified fact).,,rater1 | |
| replit-agent,B,cloud service (no client version),D11,2,,"Enterprise audit logs (65+ events incl. Agent activity) with SIEM streaming and a compliance API; policy enforcement not documented and Enterprise-only, so not 3.",[from-official-docs] https://docs.replit.com/teams/identity-and-access-management/audit-logs.md,rater1 | |
| replit-agent,B,cloud service (no client version),D12,1,,Content used to improve models by default; opt-out only for Pro (Enterprise excluded).,[from-official-docs] https://docs.replit.com/legal-and-security-info/model-improvement.md,rater1 | |
| replit-agent,B,cloud service (no client version),D13,N/A,N/A (SaaS; no distributable client).,N/A (SaaS; no distributable client).,,rater1 | |
| replit-agent,B,cloud service (no client version),D14,2,,"Enterprise admins enforce approved model providers and security policies (Auto-Protect, required scans, publish blocking); no control over agent actions.",[from-official-docs] https://docs.replit.com/teams/modes-and-model-providers.md,rater1 | |
| replit-agent,B,cloud service (no client version),D15,1,,Security email only; no bounty or advisories.,[from-official-docs] https://docs.replit.com/legal-and-security-info/security.md,rater1 | |
| roo-code,A,3.54.0 (final); CLI 0.1.17,D01,1,,"The extension has allow/deny command lists and alwaysAllow toggles, but the CLI auto-approves by default and repo .roo/mcp.json alwaysAllow '*' can approve MCP tools from repository files. Patched allow-list bypass CVEs noted.",[from-official-docs] https://github.com/RooCodeInc/Roo-Code/blob/main/packages/types/src/global-settings.ts; [from-official-docs] https://github.com/RooCodeInc/Roo-Code/blob/main/src/package.json; [from-official-docs] https://github.com/RooCodeInc/Roo-Code/blob/main/apps/cli/README.md; [from-official-docs] https://github.com/RooCodeInc/Roo-Code/security/advisories; [from-official-docs] https://github.com/RooCodeInc/Roo-Code/blob/main/src/services/mcp/McpHub.ts,rater1 | |
| roo-code,A,3.54.0 (final); CLI 0.1.17,D02,0,,No sandbox in extension or CLI.,[from-official-docs] https://github.com/RooCodeInc/Roo-Code/blob/main/README.md,rater1 | |
| roo-code,A,3.54.0 (final); CLI 0.1.17,D03,0,,No domain allowlist or network policy.,[from-official-docs] https://github.com/RooCodeInc/Roo-Code/blob/main/README.md,rater1 | |
| roo-code,A,3.54.0 (final); CLI 0.1.17,D04,0,,No trust concept: project .roo/mcp.json is watched and connected automatically; the protected-file list only stops auto-approved agent writes. Patched settings-write RCE CVEs noted.,[from-official-docs] https://github.com/RooCodeInc/Roo-Code/blob/main/src/core/protect/RooProtectedController.ts; [from-official-docs] https://github.com/RooCodeInc/Roo-Code/security/advisories; [from-official-docs] https://github.com/RooCodeInc/Roo-Code/blob/main/src/services/mcp/McpHub.ts,rater1 | |
| roo-code,A,3.54.0 (final); CLI 0.1.17,D05,1,,"Per-server alwaysAllow (wildcard allowed) and disabledTools exist, but project servers spawn without approval.",[from-official-docs] https://github.com/RooCodeInc/Roo-Code/blob/main/src/services/mcp/McpHub.ts; [from-official-docs] https://github.com/RooCodeInc/Roo-Code/security/advisories,rater1 | |
| roo-code,A,3.54.0 (final); CLI 0.1.17,D06,1,,Custom modes and skills load from the repo; marketplace without signing or vetting.,[from-official-docs] https://github.com/RooCodeInc/Roo-Code/blob/main/src/services/skills/SkillsManager.ts,rater1 | |
| roo-code,A,3.54.0 (final); CLI 0.1.17,D07,1,,.rooignore read protection exists (patched symlink bypass); credential storage not documented in the facts.,[from-official-docs] https://github.com/RooCodeInc/Roo-Code/security/advisories,rater1 | |
| roo-code,A,3.54.0 (final); CLI 0.1.17,D08,0,,CLI --print auto-approves by default with no guidance; no official CI action.,[from-official-docs] https://github.com/RooCodeInc/Roo-Code/blob/main/apps/cli/README.md; [from-official-docs] https://github.com/RooCodeInc/Roo-Code/security/advisories,rater1 | |
| roo-code,A,3.54.0 (final); CLI 0.1.17,D09,N/A,N/A (feature absent: no built-in persistent agent memory).,N/A (feature absent: no built-in persistent agent memory).,,rater1 | |
| roo-code,A,3.54.0 (final); CLI 0.1.17,D10,1,,Subtasks gated by alwaysAllowSubtasks; inheritance not documented.,[from-official-docs] https://github.com/RooCodeInc/Roo-Code/blob/main/packages/types/src/global-settings.ts,rater1 | |
| roo-code,A,3.54.0 (final); CLI 0.1.17,D11,1,,Local task history only.,[from-official-docs] https://github.com/RooCodeInc/Roo-Code/blob/main/apps/cli/README.md,rater1 | |
| roo-code,A,3.54.0 (final); CLI 0.1.17,D12,1,,Telemetry on by default with opt-out; retention not documented.,[from-official-docs] https://github.com/RooCodeInc/Roo-Code/blob/main/PRIVACY.md,rater1 | |
| roo-code,A,3.54.0 (final); CLI 0.1.17,D13,1,,curl|sh CLI installer without checksum; repository archived.,[from-official-docs] https://github.com/RooCodeInc/Roo-Code/blob/main/apps/cli/README.md; [from-official-docs] https://github.com/RooCodeInc/Roo-Code/blob/main/README.md,rater1 | |
| roo-code,A,3.54.0 (final); CLI 0.1.17,D14,1,,Only an auto-import settings path (defaults users can override).,[from-official-docs] https://github.com/RooCodeInc/Roo-Code/blob/main/src/package.json,rater1 | |
| roo-code,A,3.54.0 (final); CLI 0.1.17,D15,2,,SECURITY.md and 11 GHSAs with CVEs; no bounty.,[from-official-docs] https://github.com/RooCodeInc/Roo-Code/blob/main/SECURITY.md; [from-official-docs] https://github.com/RooCodeInc/Roo-Code/security/advisories,rater1 | |
| smolagents,C,1.27.0.dev0 (1.26.0 latest release),D01,0,,Tier C: no approval primitive; CodeAgent runs LLM-written Python unattended (only an import/function allowlist).,[from-official-docs] https://github.com/huggingface/smolagents/blob/main/docs/source/en/tutorials/secure_code_execution.md,rater1 | |
| smolagents,C,1.27.0.dev0 (1.26.0 latest release),D02,1,,Tier C: default local AST interpreter is documented as not a secure sandbox; docker/e2b/modal executors are opt-in. Patched escape CVE-2025-5120; CVE-2026-4963 without listed fix.,[from-official-docs] https://github.com/huggingface/smolagents/blob/main/docs/source/en/tutorials/secure_code_execution.md; [from-official-docs] https://github.com/huggingface/smolagents/security/advisories,rater1 | |
| smolagents,C,1.27.0.dev0 (1.26.0 latest release),D03,0,,No egress control in the local executor; SSRF CVE-2026-2654 with no fix listed.,[from-official-docs] https://github.com/advisories/GHSA-jxgv-6j54-wwc7,rater1 | |
| smolagents,C,1.27.0.dev0 (1.26.0 latest release),D04,N/A,N/A (library; only an unverified fact).,N/A (library; only an unverified fact).,,rater1 | |
| smolagents,C,1.27.0.dev0 (1.26.0 latest release),D05,1,,Tier C: stdio MCP requires an explicit trust_remote_code=True flag; no per-call approval or allowlist.,[from-official-docs] https://github.com/huggingface/smolagents/blob/main/src/smolagents/tools.py,rater1 | |
| smolagents,C,1.27.0.dev0 (1.26.0 latest release),D06,1,,"Tier C: Hub tools refuse to load unless trust_remote_code=True; no vetting, signing or pinning documented.",[from-official-docs] https://github.com/huggingface/smolagents/blob/main/src/smolagents/tools.py,rater1 | |
| smolagents,C,1.27.0.dev0 (1.26.0 latest release),D07,1,,Tier C: keys from env or --api-key on the command line (process-list exposure); remote executors get no secrets.,[from-official-docs] https://github.com/huggingface/smolagents/blob/main/src/smolagents/cli.py,rater1 | |
| smolagents,C,1.27.0.dev0 (1.26.0 latest release),D08,N/A,N/A (not documented: only an unverified fact).,N/A (not documented: only an unverified fact).,,rater1 | |
| smolagents,C,1.27.0.dev0 (1.26.0 latest release),D09,N/A,"N/A (feature absent: only in-run AgentMemory, no persistent store).","N/A (feature absent: only in-run AgentMemory, no persistent store).",[from-official-docs] https://github.com/huggingface/smolagents/blob/main/docs/source/en/tutorials/memory.md,rater1 | |
| smolagents,C,1.27.0.dev0 (1.26.0 latest release),D10,1,,Managed agents share the process and its authority; isolation only by running the whole system in a sandbox.,[from-official-docs] https://github.com/huggingface/smolagents/blob/main/docs/source/en/tutorials/secure_code_execution.md,rater1 | |
| smolagents,C,1.27.0.dev0 (1.26.0 latest release),D11,1,,Tier C: tracing requires third-party OpenInference instrumentation (opt-in).,[from-official-docs] https://github.com/huggingface/smolagents/blob/main/docs/source/en/tutorials/inspect_runs.md,rater1 | |
| smolagents,C,1.27.0.dev0 (1.26.0 latest release),D12,N/A,N/A (no telemetry in the library; opt-in OTel only).,N/A (no telemetry in the library; opt-in OTel only).,[from-official-docs] https://github.com/huggingface/smolagents/blob/main/docs/source/en/tutorials/inspect_runs.md,rater1 | |
| smolagents,C,1.27.0.dev0 (1.26.0 latest release),D13,N/A,N/A (not documented: only an unverified fact).,N/A (not documented: only an unverified fact).,,rater1 | |
| smolagents,C,1.27.0.dev0 (1.26.0 latest release),D14,N/A,N/A (not documented: only an unverified fact).,N/A (not documented: only an unverified fact).,,rater1 | |
| smolagents,C,1.27.0.dev0 (1.26.0 latest release),D15,2,,SECURITY.md with private reporting and security email; five GHSAs with CVEs (some without fixed versions); no bounty.,[from-official-docs] https://github.com/huggingface/smolagents/security/advisories,rater1 | |
| strands-agents,C,strands-py 1.61.0 / strands-harness 0.3.0,D01,1,,"Tier C: HITL presets (ask/smart) and Cedar policies exist, but harness interventions default to None (every call runs); 3P: BYPASS_TOOL_CONSENT env.",[from-official-docs] https://github.com/strands-agents/harness-sdk/blob/main/harness-py/src/strands_harness/agent.py; [third-party-report] https://github.com/strands-agents/harness-sdk/blob/main/site/src/content/docs/user-guide/evals-sdk/eval-sop.mdx,rater1 | |
| strands-agents,C,strands-py 1.61.0 / strands-harness 0.3.0,D02,1,,Tier C: default NotASandboxLocalEnvironment runs on the host; Docker/SSH sandboxes opt-in.,[from-official-docs] https://github.com/strands-agents/harness-sdk/blob/main/strands-py/src/strands/sandbox/not_a_sandbox_local_environment.py,rater1 | |
| strands-agents,C,strands-py 1.61.0 / strands-harness 0.3.0,D03,N/A,N/A (not documented: no egress facts collected).,N/A (not documented: no egress facts collected).,,rater1 | |
| strands-agents,C,strands-py 1.61.0 / strands-harness 0.3.0,D04,N/A,N/A (not documented: no workspace-config facts collected).,N/A (not documented: no workspace-config facts collected).,,rater1 | |
| strands-agents,C,strands-py 1.61.0 / strands-harness 0.3.0,D05,N/A,N/A (not documented: no MCP facts collected).,N/A (not documented: no MCP facts collected).,,rater1 | |
| strands-agents,C,strands-py 1.61.0 / strands-harness 0.3.0,D06,N/A,N/A (not documented: no extension facts collected).,N/A (not documented: no extension facts collected).,,rater1 | |
| strands-agents,C,strands-py 1.61.0 / strands-harness 0.3.0,D07,1,,"Tier C: no credential store; session store was world-readable in /tmp until 1.48.0 (patched, now project dir).",[from-official-docs] https://github.com/strands-agents/harness-sdk/security/advisories,rater1 | |
| strands-agents,C,strands-py 1.61.0 / strands-harness 0.3.0,D08,N/A,N/A (not documented: no unattended-run facts collected).,N/A (not documented: no unattended-run facts collected).,,rater1 | |
| strands-agents,C,strands-py 1.61.0 / strands-harness 0.3.0,D09,1,,Project-scoped memory and session dirs; no off switch or write confirmation documented; patched session-injection GHSA.,[from-official-docs] https://github.com/strands-agents/harness-sdk/blob/main/harness-py/src/strands_harness/defaults.py; [from-official-docs] https://github.com/strands-agents/harness-sdk/security/advisories/GHSA-3w55-933p-8pv5,rater1 | |
| strands-agents,C,strands-py 1.61.0 / strands-harness 0.3.0,D10,1,,Subagents share the parent's sandbox and hooks (depth-capped at 2); no per-agent scoping documented.,[from-official-docs] https://github.com/strands-agents/harness-sdk/blob/main/harness-py/src/strands_harness/defaults.py,rater1 | |
| strands-agents,C,strands-py 1.61.0 / strands-harness 0.3.0,D11,2,,Tier C: built-in OTel tracing one setting away (OTEL_TRACES_EXPORTER).,[from-official-docs] https://github.com/strands-agents/harness-sdk/blob/main/harness-py/src/strands_harness/telemetry.py,rater1 | |
| strands-agents,C,strands-py 1.61.0 / strands-harness 0.3.0,D12,N/A,N/A (library with no vendor telemetry; export only when configured).,N/A (library with no vendor telemetry; export only when configured).,[from-official-docs] https://github.com/strands-agents/harness-sdk/blob/main/harness-py/src/strands_harness/telemetry.py,rater1 | |
| strands-agents,C,strands-py 1.61.0 / strands-harness 0.3.0,D13,N/A,N/A (not documented: no distribution facts collected).,N/A (not documented: no distribution facts collected).,,rater1 | |
| strands-agents,C,strands-py 1.61.0 / strands-harness 0.3.0,D14,N/A,N/A (not documented: no managed-policy facts collected).,N/A (not documented: no managed-policy facts collected).,,rater1 | |
| strands-agents,C,strands-py 1.61.0 / strands-harness 0.3.0,D15,2,,SECURITY.md and two published GHSAs; no bounty documented.,[from-official-docs] https://github.com/strands-agents/harness-sdk/security/advisories,rater1 | |
| trae,B,TRAE v3.6.0 (docs changelog 2026-10-08; not executed). trae-agent 0.1.0 (source read),D01,1,,"Manual, LLM Guardian, Full Access and custom rule modes exist, but the default mode is not stated, Auto-Run 'bypasses all security checks' and an 'Automatically Run MCP' toggle exists; no deny-in-every-mode or central control.",[from-official-docs] https://docs.trae.ai/ide/permission-and-approval; [from-official-docs] https://docs.trae.ai/ide/auto-run-and-security; [third-party-report] https://docs.trae.ai/ide/auto-run-and-security,rater1 | |
| trae,B,TRAE v3.6.0 (docs changelog 2026-10-08; not executed). trae-agent 0.1.0 (source read),D02,1,,"OS sandbox exists (Seatbelt/Windows/bubblewrap) with workspace write scope, but its default state is not documented, so 'on by default' cannot be confirmed.",[from-official-docs] https://docs.trae.ai/ide/sandbox,rater1 | |
| trae,B,TRAE v3.6.0 (docs changelog 2026-10-08; not executed). trae-agent 0.1.0 (source read),D03,1,,Domain/IP allow/deny lists exist but network is allowed to all by default.,[from-official-docs] https://docs.trae.ai/ide/permission-and-approval,rater1 | |
| trae,B,TRAE v3.6.0 (docs changelog 2026-10-08; not executed). trae-agent 0.1.0 (source read),D04,1,,"Project MCP needs an explicit toggle with confirmation, but project hooks (including imported .claude settings) have no documented trust gate.",[from-official-docs] https://docs.trae.ai/ide/add-mcp-servers,rater1 | |
| trae,B,TRAE v3.6.0 (docs changelog 2026-10-08; not executed). trae-agent 0.1.0 (source read),D05,1,,"Project MCP opt-in, first-use authorization and per-server/tool mcpRules, but no managed list or change detection; TRAE does not review MCP servers.",[from-official-docs] https://docs.trae.ai/ide/auto-run-and-security; [from-official-docs] https://docs.trae.ai/ide/add-mcp-servers,rater1 | |
| trae,B,TRAE v3.6.0 (docs changelog 2026-10-08; not executed). trae-agent 0.1.0 (source read),D06,1,,Plugin marketplace without documented vetting or signing.,[from-official-docs] https://docs.trae.ai/ide/changelog,rater1 | |
| trae,B,TRAE v3.6.0 (docs changelog 2026-10-08; not executed). trae-agent 0.1.0 (source read),D07,N/A,N/A (not documented: no credential facts collected).,N/A (not documented: no credential facts collected).,,rater1 | |
| trae,B,TRAE v3.6.0 (docs changelog 2026-10-08; not executed). trae-agent 0.1.0 (source read),D08,N/A,N/A (not documented: only an unverified fact).,N/A (not documented: only an unverified fact).,,rater1 | |
| trae,B,TRAE v3.6.0 (docs changelog 2026-10-08; not executed). trae-agent 0.1.0 (source read),D09,1,,Rules & Memories with global and project rules; memory write controls and off switch not documented.,[from-official-docs] https://docs.trae.ai/ide/rules,rater1 | |
| trae,B,TRAE v3.6.0 (docs changelog 2026-10-08; not executed). trae-agent 0.1.0 (source read),D10,N/A,N/A (not documented: only an unverified fact).,N/A (not documented: only an unverified fact).,,rater1 | |
| trae,B,TRAE v3.6.0 (docs changelog 2026-10-08; not executed). trae-agent 0.1.0 (source read),D11,N/A,N/A (not documented: only an unverified fact).,N/A (not documented: only an unverified fact).,,rater1 | |
| trae,B,TRAE v3.6.0 (docs changelog 2026-10-08; not executed). trae-agent 0.1.0 (source read),D12,0,,"Chat content may be used for training unless Privacy mode is on (would be 1), lowered to 0 by a third-party report, with the vendor's own statement, that the telemetry toggle covers only VS Code-framework telemetry and collection continued when disabled.",[from-official-docs] https://docs.trae.ai/ide/privacy-mode; [from-official-docs] https://docs.trae.ai/ide/data-storage-policy-for-remote-agents; [third-party-report] https://www.theregister.com/2025/07/28/bytedance_trae_telemetry/,rater1 | |
| trae,B,TRAE v3.6.0 (docs changelog 2026-10-08; not executed). trae-agent 0.1.0 (source read),D13,N/A,N/A (not documented: no distribution facts collected).,N/A (not documented: no distribution facts collected).,,rater1 | |
| trae,B,TRAE v3.6.0 (docs changelog 2026-10-08; not executed). trae-agent 0.1.0 (source read),D14,N/A,N/A (not documented: only an unverified fact).,N/A (not documented: only an unverified fact).,,rater1 | |
| trae,B,TRAE v3.6.0 (docs changelog 2026-10-08; not executed). trae-agent 0.1.0 (source read),D15,N/A,N/A (not documented: only an unverified fact).,N/A (not documented: only an unverified fact).,,rater1 | |
| trueforge,C,0.179.1,D01,2,,Tier C: the built-in loop pauses for approval before MCP tools labelled write/destructive by default; unlabelled tools run unless named or @all. Not 3: no platform policy enforcement documented.,[from-official-docs] https://github.com/truefoundry/trueforge/blob/main/docs/create-agent/overview.mdx,rater1 | |
| trueforge,C,0.179.1,D02,2,,"Tier C: code runs in a Daytona or srt sandbox with the agent loop and credentials outside it; sandbox egress is unrestricted by default, so not 3.",[from-official-docs] https://github.com/truefoundry/trueforge/blob/main/docs/sandbox.mdx,rater1 | |
| trueforge,C,0.179.1,D03,1,,"Domain allowlist, block-all network and server outbound host lists exist, but sandbox egress is unrestricted by default (opt-in controls).",[from-official-docs] https://github.com/truefoundry/trueforge/blob/main/docs/sandbox.mdx,rater1 | |
| trueforge,C,0.179.1,D04,N/A,N/A (server product; only an unverified fact).,N/A (server product; only an unverified fact).,,rater1 | |
| trueforge,C,0.179.1,D05,2,,"Tier C: remote MCP only with documented header/OAuth auth, per-tool approval switches, write/destructive tools approved by default. Not 3: no platform allowlist or change detection.",[from-official-docs] https://github.com/truefoundry/trueforge/blob/main/docs/mcp-servers.mdx; [from-official-docs] https://github.com/truefoundry/trueforge/blob/main/docs/create-agent/overview.mdx,rater1 | |
| trueforge,C,0.179.1,D06,1,,Git-backed skills registered by admins in Settings; no vetting or pinning documented.,[from-official-docs] https://github.com/truefoundry/trueforge/blob/main/docs/skills.mdx,rater1 | |
| trueforge,C,0.179.1,D07,1,,"Tier C: keys in the DB with redacted API/UI stand-ins and never inside the sandbox, but no secret manager, managed identity or at-rest encryption.",[from-official-docs] https://github.com/truefoundry/trueforge/blob/main/docs/sandbox.mdx,rater1 | |
| trueforge,C,0.179.1,D08,N/A,"N/A (not documented: the in-dimension fact concerns deployment authentication, scored under D16).","N/A (not documented: the in-dimension fact concerns deployment authentication, scored under D16).",[from-official-docs] https://github.com/truefoundry/trueforge/blob/main/docs/authentication/overview.mdx,rater1 | |
| trueforge,C,0.179.1,D09,N/A,N/A (not documented: only an unverified fact).,N/A (not documented: only an unverified fact).,,rater1 | |
| trueforge,C,0.179.1,D10,N/A,N/A (not documented: only an unverified fact).,N/A (not documented: only an unverified fact).,,rater1 | |
| trueforge,C,0.179.1,D11,2,,"Tier C: sessions record turns, tool calls, subagents, tokens and timing by default in SQLite/Postgres.",[from-official-docs] https://github.com/truefoundry/trueforge/blob/main/docs/sessions.mdx,rater1 | |
| trueforge,C,0.179.1,D12,1,,"Sentry error reporting is off by default, but no retention is documented (score-1 clause).",[from-official-docs] https://github.com/truefoundry/trueforge/blob/main/packages/trueforge/src/config.ts,rater1 | |
| trueforge,C,0.179.1,D13,2,,Tier C: npm publication with SLSA v1 provenance attestation.,[from-official-docs] https://registry.npmjs.org/-/npm/v1/attestations/@truefoundry%2ftrueforge@0.179.1,rater1 | |
| trueforge,C,0.179.1,D14,1,,OIDC admin/user roles and admin-only Settings; no policy layer with documented coverage.,[from-official-docs] https://github.com/truefoundry/trueforge/blob/main/docs/authentication/overview.mdx,rater1 | |
| trueforge,C,0.179.1,D15,1,,SECURITY.md with email and 3-day ack; no advisories.,[from-official-docs] https://github.com/truefoundry/trueforge/blob/main/SECURITY.md,rater1 | |
| trueforge,C,0.179.1,D16,0,,"Standalone mode has no login and a fixed well-known service key; Helm defaults have no OIDC, a placeholder API key and unauthenticated Redis.",[from-official-docs] https://github.com/truefoundry/trueforge/blob/main/docs/authentication/overview.mdx,rater1 | |
| vertex-ai-agent-engine,C,"Agent Runtime (renamed 2026-04-22); Agent Identity GA claim on 2026-04-22, API in Preview per 2026-06-18 note",D01,1,,"Tier C: approval is delegated to the agent framework; the platform offers IAM allow/deny on agent identities and Preview semantic governance, but no default tool approval.",[from-official-docs] https://docs.cloud.google.com/gemini-enterprise-agent-platform/release-notes; [from-official-docs] https://docs.cloud.google.com/gemini-enterprise-agent-platform/scale/runtime/agent-identity,rater1 | |
| vertex-ai-agent-engine,C,"Agent Runtime (renamed 2026-04-22); Agent Identity GA claim on 2026-04-22, API in Preview per 2026-06-18 note",D02,3,,Tier C: provider-managed shell/computer-use sandboxes (GA) with VPC Service Controls and private connectivity.,[from-official-docs] https://docs.cloud.google.com/gemini-enterprise-agent-platform/release-notes,rater1 | |
| vertex-ai-agent-engine,C,"Agent Runtime (renamed 2026-04-22); Agent Identity GA claim on 2026-04-22, API in Preview per 2026-06-18 note",D03,3,,Tier C: Agent Gateway enforces VPC-SC perimeter rules (gateways created after 2026-09-08); the conflicting access-doc note is unverified and ignored.,[from-official-docs] https://docs.cloud.google.com/gemini-enterprise-agent-platform/release-notes,rater1 | |
| vertex-ai-agent-engine,C,"Agent Runtime (renamed 2026-04-22); Agent Identity GA claim on 2026-04-22, API in Preview per 2026-06-18 note",D04,N/A,N/A (cloud-deployed; no repository-level configuration).,N/A (cloud-deployed; no repository-level configuration).,,rater1 | |
| vertex-ai-agent-engine,C,"Agent Runtime (renamed 2026-04-22); Agent Identity GA claim on 2026-04-22, API in Preview per 2026-06-18 note",D05,2,,"Tier C: Agent Gateway, Agent Registry and Model Armor on the gateway are GA; the facts do not state that the registry enforces a tool allowlist, so not 3.",[from-official-docs] https://docs.cloud.google.com/gemini-enterprise-agent-platform/release-notes,rater1 | |
| vertex-ai-agent-engine,C,"Agent Runtime (renamed 2026-04-22); Agent Identity GA claim on 2026-04-22, API in Preview per 2026-06-18 note",D06,N/A,N/A (not documented: no extension facts collected).,N/A (not documented: no extension facts collected).,,rater1 | |
| vertex-ai-agent-engine,C,"Agent Runtime (renamed 2026-04-22); Agent Identity GA claim on 2026-04-22, API in Preview per 2026-06-18 note",D07,1,,"Tier C: managed identity integration exists (2 on the ladder), but deployed agents get all their service account's permissions and Agent Identity is Preview; third-party Unit 42 report of excessive default service-agent permissions and non-editable broad scopes lowers to 1.",[from-official-docs] https://docs.cloud.google.com/gemini-enterprise-agent-platform/scale/runtime/manage-agent-access; [from-official-docs] https://docs.cloud.google.com/gemini-enterprise-agent-platform/release-notes; [third-party-report] https://unit42.paloaltonetworks.com/double-agents-vertex-ai/,rater1 | |
| vertex-ai-agent-engine,C,"Agent Runtime (renamed 2026-04-22); Agent Identity GA claim on 2026-04-22, API in Preview per 2026-06-18 note",D08,N/A,N/A (not documented: no unattended-run facts).,N/A (not documented: no unattended-run facts).,,rater1 | |
| vertex-ai-agent-engine,C,"Agent Runtime (renamed 2026-04-22); Agent Identity GA claim on 2026-04-22, API in Preview per 2026-06-18 note",D09,1,,"Memory Bank is GA; scoping, review and off-switch controls not documented.",[from-official-docs] https://docs.cloud.google.com/gemini-enterprise-agent-platform/release-notes,rater1 | |
| vertex-ai-agent-engine,C,"Agent Runtime (renamed 2026-04-22); Agent Identity GA claim on 2026-04-22, API in Preview per 2026-06-18 note",D10,N/A,N/A (not documented: no multi-agent facts).,N/A (not documented: no multi-agent facts).,,rater1 | |
| vertex-ai-agent-engine,C,"Agent Runtime (renamed 2026-04-22); Agent Identity GA claim on 2026-04-22, API in Preview per 2026-06-18 note",D11,2,,Tier C: Agent Observability is GA.,[from-official-docs] https://docs.cloud.google.com/gemini-enterprise-agent-platform/release-notes,rater1 | |
| vertex-ai-agent-engine,C,"Agent Runtime (renamed 2026-04-22); Agent Identity GA claim on 2026-04-22, API in Preview per 2026-06-18 note",D12,N/A,N/A (not documented: Google Cloud data terms not collected).,N/A (not documented: Google Cloud data terms not collected).,,rater1 | |
| vertex-ai-agent-engine,C,"Agent Runtime (renamed 2026-04-22); Agent Identity GA claim on 2026-04-22, API in Preview per 2026-06-18 note",D13,N/A,N/A (not documented: managed service; SDK provenance not collected).,N/A (not documented: managed service; SDK provenance not collected).,,rater1 | |
| vertex-ai-agent-engine,C,"Agent Runtime (renamed 2026-04-22); Agent Identity GA claim on 2026-04-22, API in Preview per 2026-06-18 note",D14,2,,Tier C: IAM allow/deny and principalSet grants across all agents in a project/organisation; coverage of tools and telemetry not documented.,[from-official-docs] https://docs.cloud.google.com/gemini-enterprise-agent-platform/scale/runtime/agent-identity,rater1 | |
| vertex-ai-agent-engine,C,"Agent Runtime (renamed 2026-04-22); Agent Identity GA claim on 2026-04-22, API in Preview per 2026-06-18 note",D15,1,,"GHSA/CVEs were issued for the Vertex AI SDK, but no policy or bounty is documented in the facts, so score 2 is not met.",[from-official-docs] https://github.com/advisories/GHSA-wh2j-26j7-9728,rater1 | |
| warp,A,stable v0.2026.06.03.09.49.stable_00,D01,1,,"GUI default asks before commands, but code diffs are 'AgentDecides' (writes without a user prompt), auto-approve bypasses the denylist by default and the CLI profile allows everything. Patched denylist/approval bypass CVEs noted.",[from-official-docs] https://github.com/warpdotdev/warp/blob/master/crates/cloud_object_models/src/ai_execution_profile.rs; [from-official-docs] https://github.com/warpdotdev/warp/blob/master/app/src/settings/ai.rs; [from-official-docs] https://github.com/warpdotdev/warp/security/advisories,rater1 | |
| warp,A,stable v0.2026.06.03.09.49.stable_00,D02,1,,--sandboxed/--environment are opt-in; local agent runs unsandboxed.,[from-official-docs] https://github.com/warpdotdev/warp/blob/master/crates/warp_cli/src/agent.rs,rater1 | |
| warp,A,stable v0.2026.06.03.09.49.stable_00,D03,1,,"Only a default command denylist (curl, wget, ssh, ...), a weak control that is not a destination allowlist.",[from-official-docs] https://github.com/warpdotdev/warp/blob/master/crates/cloud_object_models/src/ai_execution_profile.rs,rater1 | |
| warp,A,stable v0.2026.06.03.09.49.stable_00,D04,0,,Warp-native project MCP config is always detected and spawned with no trust concept; only third-party MCP file import is off by default.,[from-official-docs] https://github.com/warpdotdev/warp/blob/master/app/src/settings/ai.rs,rater1 | |
| warp,A,stable v0.2026.06.03.09.49.stable_00,D05,1,,"Per-profile MCP allow/deny lists (deny wins), but project .warp MCP servers spawn without approval and mcp_permissions defaults to AgentDecides.",[from-official-docs] https://github.com/warpdotdev/warp/blob/master/crates/cloud_object_models/src/ai_execution_profile.rs; [from-official-docs] https://docs.warp.dev/agents/using-agents/agent-profiles-permissions; [from-official-docs] https://github.com/warpdotdev/warp/blob/master/app/src/settings/ai.rs,rater1 | |
| warp,A,stable v0.2026.06.03.09.49.stable_00,D06,1,,Skills/workflows from a home skills dir; no vetting or signing documented.,[from-official-docs] https://github.com/warpdotdev/warp/blob/master/crates/warp_core/src/paths.rs,rater1 | |
| warp,A,stable v0.2026.06.03.09.49.stable_00,D07,1,,"Secret-redaction crate and keys kept out of settings/sync payloads, but credential storage mechanism and log retention are not documented.",[from-official-docs] https://github.com/warpdotdev/warp/blob/master/app/src/settings/privacy.rs; [from-official-docs] https://github.com/warpdotdev/warp/blob/master/specs/APP-5380/PRODUCT.md,rater1 | |
| warp,A,stable v0.2026.06.03.09.49.stable_00,D08,0,,The default CLI profile 'will never ask the user for permission'; no CI guidance.,[from-official-docs] https://github.com/warpdotdev/warp/blob/master/crates/cloud_object_models/src/ai_execution_profile.rs,rater1 | |
| warp,A,stable v0.2026.06.03.09.49.stable_00,D09,1,,Cloud-synced Drive rules/knowledge and memory store; no confirmation or scoping documented.,[from-official-docs] https://github.com/warpdotdev/warp/blob/master/app/src/settings/ai.rs,rater1 | |
| warp,A,stable v0.2026.06.03.09.49.stable_00,D10,1,,run_agents is AlwaysAsk in the GUI but AlwaysAllow in the CLI; inheritance not documented.,[from-official-docs] https://github.com/warpdotdev/warp/blob/master/crates/cloud_object_models/src/ai_execution_profile.rs,rater1 | |
| warp,A,stable v0.2026.06.03.09.49.stable_00,D11,1,,Local block history; agent commands excluded from history by default; no export documented.,[from-official-docs] https://github.com/warpdotdev/warp/blob/master/app/src/settings/ai.rs,rater1 | |
| warp,A,stable v0.2026.06.03.09.49.stable_00,D12,1,,Telemetry and crash reporting on by default with opt-out; retention not documented.,[from-official-docs] https://github.com/warpdotdev/warp/blob/master/app/src/settings/privacy.rs,rater1 | |
| warp,A,stable v0.2026.06.03.09.49.stable_00,D13,1,,"Native auto-update; no signing, checksum or update-pinning documented in the facts.",[from-official-docs] https://github.com/warpdotdev/warp/blob/master/app/src/autoupdate,rater1 | |
| warp,A,stable v0.2026.06.03.09.49.stable_00,D14,2,,Team Admin Panel denylist rules are never bypassed and provider keys are team-managed; coverage limited.,[from-official-docs] https://docs.warp.dev/agents/using-agents/agent-profiles-permissions; [from-official-docs] https://github.com/warpdotdev/warp/blob/master/specs/byok-byoe-enterprise/PRODUCT.md,rater1 | |
| warp,A,stable v0.2026.06.03.09.49.stable_00,D15,2,,SECURITY.md and 10 GHSAs with CVEs; bounty not confirmed.,[from-official-docs] https://github.com/warpdotdev/warp/blob/master/SECURITY.md; [from-official-docs] https://github.com/warpdotdev/warp/security/advisories,rater1 | |
| windsurf,B,Devin Desktop v3.10.67 (2026-10-12); docs-level,D01,1,,"Graded auto-execution levels with deny precedence and an admin maximum level, but the default level is not stated in the facts and Turbo runs everything except the deny list.",[from-official-docs] https://docs.devin.ai/desktop/terminal; [from-official-docs] https://docs.devin.ai/desktop/changelog.md,rater1 | |
| windsurf,B,Devin Desktop v3.10.67 (2026-10-12); docs-level,D02,1,,Optional sandbox for Devin Local/CLI only; none documented for Cascade.,[from-official-docs] https://docs.devin.ai/desktop/devin-local.md,rater1 | |
| windsurf,B,Devin Desktop v3.10.67 (2026-10-12); docs-level,D03,1,,Domain allow/deny only inside the optional sandbox; default network_mode full.,[from-official-docs] https://docs.devin.ai/cli/sandbox.md,rater1 | |
| windsurf,B,Devin Desktop v3.10.67 (2026-10-12); docs-level,D04,1,,"Restricted Mode disables agents and hooks until the workspace is trusted, but MCP and settings coverage are not documented; patched nested-settings bypass.",[from-official-docs] https://docs.devin.ai/desktop/changelog.md; [from-official-docs] https://github.com/advisories/GHSA-x6gf-c826-8c9m,rater1 | |
| windsurf,B,Devin Desktop v3.10.67 (2026-10-12); docs-level,D05,1,,"Admin MCP allowlist and registry URLs, but no project-server approval or per-tool permissions documented.",[from-official-docs] https://docs.devin.ai/desktop/cascade/mcp,rater1 | |
| windsurf,B,Devin Desktop v3.10.67 (2026-10-12); docs-level,D06,1,,"Admin publisher allowlist and marketplace URL, but no vetting or signing documented.",[from-official-docs] https://docs.devin.ai/desktop/enterprise-policies.md,rater1 | |
| windsurf,B,Devin Desktop v3.10.67 (2026-10-12); docs-level,D07,N/A,N/A (not documented: no credential facts collected).,N/A (not documented: no credential facts collected).,,rater1 | |
| windsurf,B,Devin Desktop v3.10.67 (2026-10-12); docs-level,D08,N/A,N/A (no headless mode in the desktop product; scored under devin).,N/A (no headless mode in the desktop product; scored under devin).,[from-official-docs] https://docs.devin.ai/cli/reference/commands.md,rater1 | |
| windsurf,B,Devin Desktop v3.10.67 (2026-10-12); docs-level,D09,1,,Cascade memories are auto-generated and agent-writable (workspace-scoped) with no documented review or off switch.,[from-official-docs] https://docs.devin.ai/desktop/cascade/memories,rater1 | |
| windsurf,B,Devin Desktop v3.10.67 (2026-10-12); docs-level,D10,1,,Session approvals cover subagents (inherited); per-agent scoping not documented.,[from-official-docs] https://docs.devin.ai/desktop/devin-local.md,rater1 | |
| windsurf,B,Devin Desktop v3.10.67 (2026-10-12); docs-level,D11,2,,"Custom analytics export, shareable transcripts with secret redaction and transcript hooks for auditing; policy enforcement not documented.",[from-official-docs] https://docs.devin.ai/desktop/devin-local.md; [from-official-docs] https://docs.devin.ai/desktop/cascade/hooks.md,rater1 | |
| windsurf,B,Devin Desktop v3.10.67 (2026-10-12); docs-level,D12,1,,"Admin TelemetryLevel policy exists, but retention and the training default are not documented.",[from-official-docs] https://docs.devin.ai/desktop/enterprise-policies.md,rater1 | |
| windsurf,B,Devin Desktop v3.10.67 (2026-10-12); docs-level,D13,1,,Auto-update controllable by policy; no signing or checksum documented.,[from-official-docs] https://docs.devin.ai/desktop/enterprise-policies.md,rater1 | |
| windsurf,B,Devin Desktop v3.10.67 (2026-10-12); docs-level,D14,2,,"MDM/GPO device policies override local settings and cover auto-execution cap, MCP, extensions and telemetry. Not 3: sandbox not covered and nothing verified for a docs-only harness.",[from-official-docs] https://docs.devin.ai/desktop/enterprise-policies.md; [from-official-docs] https://docs.devin.ai/desktop/terminal; [from-official-docs] https://docs.devin.ai/desktop/cascade/mcp,rater1 | |
| windsurf,B,Devin Desktop v3.10.67 (2026-10-12); docs-level,D15,1,,"security@ with PGP, CVD and safe harbour, but no advisories published (an unfixed CVE at disclosure).",[from-official-docs] https://docs.devin.ai/desktop/security/reporting.md; [from-official-docs] https://www.tenable.com/security/research/tra-2025-47,rater1 | |