| { |
| "observed_at": "2026-09-09T17:54:49Z", |
| "source_root": "/home/smlflg/Projekte/AgentProjectMinenraeumer/ProjectIntakeTower", |
| "pending_count": 139, |
| "pending_without_project_path": 10, |
| "pending_projects_missing": 1, |
| "pending_with_hai_mission": 0, |
| "done_directories": 3, |
| "sources": { |
| "NIGHT_STOP.md": { |
| "path": "/home/smlflg/Projekte/AgentProjectMinenraeumer/ProjectIntakeTower/NIGHT_STOP.md", |
| "sha256": "dd4b718d37ff4e411a23242c365c3785342c95a0b6c15ebc5aee7ed6463ba278", |
| "text": "# Night Stop\n\nTimestamp: 2026-05-04T20:17:05+02:00\n\n## Decision\n\nStop active agent expansion for tonight. Do not start new agents, operators, sub-agents, GitHub issues, PRs, pushes, deletes, refactors, or source-project edits.\n\nThe current value is in the artifacts already produced by `ProjectIntakeTower`, not in additional token burn tonight.\n\n## Current Control Surface\n\nUse only this directory as the restart point:\n\n```text\n/home/smlflg/Projekte/AgentProjectMinenraeumer/ProjectIntakeTower\n```\n\nRead first tomorrow:\n\n1. `OWNER_DIGEST.md`\n2. `triage/TOP_10_NEXT_ACTIONS.md`\n3. `triage/HUMAN_DECISION_REQUIRED.md`\n4. `triage/PROJECT_PRIORITIZATION.md`\n5. `visual_summary/project_learning_graphic.html`\n6. `jobs/done/`\n7. `jobs/pending/`\n\n## State Snapshot\n\n- `jobs/` currently contains 143 files.\n- `jobs/done/` exists and contains completed review artifacts.\n- `jobs/running/` exists and should be checked before launching anything new.\n- Recent triage artifacts include `KEEP.md`, `MERGE.md`, `ARCHIVE.md`, `PAUSE.md`, `URGENT.md`, `TOKEN_BURN_CANDIDATE.md`, `PROJECT_PRIORITIZATION.md`, and `READ_ONLY_AUDIT.md`.\n- Recent visual artifacts include `visual_summary/project_learning_graphic.html`, `.png`, and `learning_graphic_data.json`.\n\n## Observed Active Agent Process Groups\n\nObserved around 2026-05-04T20:17+02:00:\n\n- Long-running `claude -r` with MCP servers.\n- Several `hermes` profiles: `meta-new-agent`, `operator`, `operator-extreme`, and `meta`.\n- Several `codex` sessions with companion bridges.\n- `opencode` and two `opencode run --agent plan` verifier calls for the ACT consolidation plan.\n\nDo not treat these as proof that more work should be launched. If they are still active tomorrow, inspect before trusting their output.\n\n## Stop Message For Any Still-Running Agent\n\n```text\nSTOP NOW AFTER CURRENT STEP.\n\nDo not spawn more agents.\nDo not create operators or sub-operators.\nDo not modify source projects.\nDo not create GitHub issues, PRs, pushes, deletes, migrations, or refactors.\n\nWrite or summarize final state only.\nIf no explicit ProjectIntakeTower job_id exists, stop with NEEDS_HUMAN.\n\nAllowed output path:\nProjectIntakeTower/jobs/done/<job_id>/\n```\n\n## Hard Rule Until Resume\n\nNo more token-burn work tonight. Sleep before interpreting the output.\n\n" |
| }, |
| "PANIC_QUOTA_STOPLOSS.md": { |
| "path": "/home/smlflg/Projekte/AgentProjectMinenraeumer/ProjectIntakeTower/PANIC_QUOTA_STOPLOSS.md", |
| "sha256": "2c3a966b3a3a0450c4c8497f767fc3b3bbba6c9b5337eeee9e725ff3086c239c", |
| "text": "# PANIC QUOTA STOP-LOSS — 2026-05-04\n\nStatus: ACTIVE\nOwner: Samuel\nOperator stance: reduce chaos, do not reduce output.\n\n## Current observed state\n\nSystem scan at 2026-05-04 19:14:43 CEST showed:\n\n- `focused_category_scan.py` running at ~99% CPU under:\n `/home/smlflg/Projekte/AgentProjectMinenraeumer/ProjectIntakeTower/verifier/`\n- multiple Codex processes active\n- OpenCode active\n- multiple Hermes operator/profile processes active\n- Project Intake Tower already produced concrete artifacts:\n - `OWNER_DIGEST.md`\n - `triage/TOP_10_NEXT_ACTIONS.md`\n - `triage/BACKLOG.md`\n - `github/issue_drafts/`\n - `jobs/pending/`\n\n## Hard correction\n\nThe pyramid idea is not leverage anymore. It is orchestration debt.\n\nOperator², Operator³, and agents spawning agents only make sense if each layer compresses uncertainty.\nIf the layer mostly creates more tasks, more workers, and more unclear ownership, it is not an operator layer. It is entropy with a title.\n\n## Sunk-cost rule\n\nExpiring tokens are already paid.\nThe only rational question now is:\n\n> Can the next token produce an artifact that is useful tomorrow?\n\nUseful artifacts:\n- inventory\n- digest\n- backlog\n- issue draft\n- review report\n- test result\n- bounded implementation diff\n- verification report\n\nBad artifacts:\n- new agent hierarchy\n- new repo without owner decision\n- broad refactor\n- speculative operator layer\n- unreviewed code churn\n- parallel workers touching the same scope\n\n## Stop spawning rule\n\nUntil the current Project Intake Tower verifier finishes:\n\n- No new Operator² / Operator³ agents.\n- No new subagent pyramids.\n- No new source-project edits.\n- No parallel builders in the same repo.\n- No GitHub issue creation from drafts yet.\n- No deleting/archive moves.\n\nAllowed while verifier runs:\n\n- read-only audits\n- summaries\n- review of produced artifacts\n- issue-draft quality checks\n- deduplication plans\n- writing operator policy docs\n\n## Project Intake Tower operating contract\n\nProject Intake Tower is allowed to do exactly this:\n\n1. Inventory projects.\n2. Generate project cards.\n3. Generate evidence snippets.\n4. Generate owner digest.\n5. Generate triage files.\n6. Generate local GitHub issue drafts.\n7. Generate pending review-only job contracts.\n8. Verify artifact consistency.\n9. Stop.\n\nIt is NOT allowed to:\n\n- modify source projects\n- spawn new operator hierarchies\n- open GitHub issues automatically\n- archive/delete projects automatically\n- decide owner intent\n- turn every project into active work\n\n## Next 30-minute plan\n\n1. Let current verifier finish or inspect it if it stalls.\n2. Read verifier output / generated consistency reports.\n3. Produce one final `OWNER_DECISION_PACKET.md` with:\n - top 10\n - 3 merge candidates\n - 3 archive candidates\n - 3 must-not-touch projects\n - which repo, if any, gets a bounded implementation next\n4. Only then spend tokens on one of:\n - PromptGarage curation fix\n - Project Intake Tower verification/report cleanup\n - CodexDashBoard baseline setup\n\n## Max parallelism rule\n\nAt most 2 active coding agents touching source code.\nAt most 4 read-only agents producing reports.\nExactly 1 owner-facing synthesizer.\n\nIf more are active, the next task is not more work. The next task is convergence.\n\n## Best token-burning prompts that are still useful\n\n### Read-only issue draft reviewer\n\n```text\nReview the local issue drafts under ProjectIntakeTower/github/issue_drafts.\nDo not create GitHub issues.\nDo not modify source projects.\nClassify each draft as: publishable, needs human decision, duplicate, too vague, unsafe.\nReturn a compact table and top 10 publishable drafts.\n```\n\n### Owner packet synthesizer\n\n```text\nRead ProjectIntakeTower/OWNER_DIGEST.md and triage/*.md.\nProduce OWNER_DECISION_PACKET.md for Samuel.\nDo not modify source projects.\nCompress the current chaos into exactly 10 decisions:\n- 3 do now\n- 3 defer\n- 3 archive/merge candidates\n- 1 must-not-touch warning\nEach decision must cite the artifact path that supports it.\n```\n\n### Verifier failure analyst\n\n```text\nRead ProjectIntakeTower/verifier outputs and logs.\nDo not modify source projects.\nFind only consistency failures in the intake artifact set:\nmissing project cards, broken evidence paths, job contracts without cards, issue drafts without evidence.\nReturn exact file paths and minimal repair tasks.\n```\n\n## Emergency rule\n\nIf panic rises again, do not spawn.\nWrite a report.\nReports are how panic becomes control.\n" |
| }, |
| "MORNING_RESUME.md": { |
| "path": "/home/smlflg/Projekte/AgentProjectMinenraeumer/ProjectIntakeTower/MORNING_RESUME.md", |
| "sha256": "de1d69a6ed04be150ae415554f5023109b35afd857ad22acd3e8627fa0db285d", |
| "text": "# Morning Resume\n\nStart here after sleep. Do not try to reconstruct the full agent cascade from memory.\n\n## First 10 Minutes\n\n1. Read `NIGHT_STOP.md`.\n2. Read `OWNER_DIGEST.md`.\n3. Open `visual_summary/project_learning_graphic.html`.\n4. Read `triage/HUMAN_DECISION_REQUIRED.md`.\n5. Read only the first page of `triage/TOP_10_NEXT_ACTIONS.md`.\n\n## First Decision\n\nChoose exactly one anchor for the next work block:\n\n- `ProjectIntakeTower` if the goal is to understand and organize yesterday's output.\n- `ACT-agentcommandotower` if the goal is to consolidate the Agent Control Tower direction.\n- One job from `jobs/pending/` if the goal is controlled token burn.\n\nDefault: choose `ProjectIntakeTower`.\n\n## Allowed Work Block\n\nFor the first resumed block:\n\n- One operator.\n- Maximum two workers.\n- Each worker gets exactly one `jobs/pending/*.json`.\n- Workers write only to `jobs/done/<job_id>/`.\n- No worker may spawn sub-agents.\n- No source project edits until a human decision is written down.\n\n## Morning Verification\n\nRun these from `ProjectIntakeTower` before trusting the map:\n\n```bash\npython3 verifier/verify_inventory.py\npython3 verifier/verify_job_contracts.py\npython3 verifier/verify_artifacts.py\n```\n\nThen inspect:\n\n```bash\nfind jobs/done -maxdepth 2 -type f | sort\nfind jobs/running -maxdepth 2 -type f | sort\nfind jobs/pending -maxdepth 1 -type f | sort | head -40\n```\n\n## Do Not Do First\n\n- Do not open 20 terminals.\n- Do not launch recursive operators.\n- Do not merge, archive, delete, or push.\n- Do not try to understand all 500 project clusters at once.\n- Do not use token consumption as the success metric.\n\n## Success Criteria For The Morning\n\nWithin 30 minutes, produce one human-readable owner note:\n\n```text\nWhat happened:\nWhat artifacts are trustworthy:\nWhat is noise:\nOne chosen anchor:\nNext one job:\n```\n\n" |
| } |
| }, |
| "scope": "Read-only inventory; no old job authorized or started." |
| } |
