Spaces:
Sleeping
Sleeping
Commit History
-
fix: add main() function and if __name__ guard for direct script execution
OpenEnv Contributor- Add main() function to server/app.py for callable entry point - Add if __name__ == '__main__' guard for direct Python script execution - Ensures server/app.py can run as both Uvicorn module and standalone script - Fixes multi-mode deployment validation requirementsd9abbc1 -
refactor: full OpenEnv REST API compliance with FastAPI + strict 0.1-0.9 scoring
OpenEnv Contributor- Replace dummy_server.py with production FastAPI app implementing step()/reset()/state() endpoints - Add typed Pydantic models: Observation, Action, Reward (required by OpenEnv spec) - Scale final scores to strict [0.1, 0.9] boundary per hackathon requirements - Update Dockerfile to run Uvicorn with FastAPI instead of dummy HTTP server - Add fastapi and uvicorn dependencies to requirements.txt - Guarantee automated HF Space ping returns 200 + /reset endpoint responds correctly - Ready for OpenEnv evaluation pipeline on Hugging Face Spacesd73d6cb -
fix: update score range to 0.1-0.9 per hackathon spec, add inference tests, update validator
OpenEnv Contributorcd7807c -
fix: add explicit score validation and fallback - ensure all grader scores are strictly between 0 and 1
OpenEnv Contributor1b160d3 -
fix: remove early break in run_campaign - ensure all 3 tasks run for grading
OpenEnv Contributor021adc0 -
fix: add graders.py to Docker image copy - critical missing file for HF evaluation
OpenEnv Contributor44281cb -
fix: explicitly declare graders module in pyproject.toml for proper packaging
OpenEnv Contributor7867bc6 -
fix: remove old env/ directory from git tracking - graders now at root level only
OpenEnv Contributor79bd5ba -
fix: move graders to module root directory so openenv-core on HF can discover it
OpenEnv Contributor62f46ca -
fix: correct grader module path from 'graders:' to 'env.graders:' for all tasks
OpenEnv Contributor343ce47
-
fix: Add LLM safety timeout and graceful exit handling
OpenEnv Contributor1. LLM Safety Timeout (20 seconds): - Reduced timeout from 45.0 to 20.0 seconds in call_llm() - Hard deadline prevents hanging on slow/unresponsive API calls - Exception handling wraps any timeout/connection errors - Safe fallback: returns 'Action: aws.soc.get_alerts({})' on error - Logs error details to stderr but never crashes 2. Graceful Exit (15-minute wait): - Replaced infinite 'while True: time.sleep(3600)' loop - Now sleeps for 900 seconds (15 minutes) after task completion - Prints 'Waking up and exiting cleanly' before exit - Prevents runaway processes while allowing evaluator to collect results Critical reliability improvements: ✓ No more hanging LLM calls ✓ Safe fallback preserves evaluation continuity ✓ Controlled exit prevents resource leaks ✓ Maintains evaluator compatibility489ba02
-
refactor: Replace graders with action-based scoring system
OpenEnv Contributor- Simplified graders to check for specific action strings in trajectory - Each task grader now uses _extract_actions() for flexible trajectory format - Scoring based on presence of key actions (get_alerts, describe_instances, etc.) - Updated score bounds to (0.01, 0.99) with no binary 0.0 or 1.0 EASY task (30% alerts + 30% investigation + 40% isolation): - get_alerts → 30% - describe_instances → 30% - isolate_instance → 40% MEDIUM task (10% alerts + 30% logs + 30% IAM + 30% isolation): - get_alerts → 10% - cloudtrail/lookup_events → 30% - revoke_role/iam → 30% - isolate_instance → 30% HARD task (20% investigation + 40% WAF + 40% isolation, fatal DB check): - terminate_db/stop_db → 0.01 (instant failure) - get_alerts → 10% - cloudtrail → 10% - block_ip/waf → 40% - isolate_instance → 40% Updated openenv.yaml tasks section to minimal format: - id: easy/medium/hard grader: graders:grade_easy/medium/hard This implementation supports: ✓ Action-based evaluation (works with any traj…b23a6d7 -
docs: Add Phase 2 verification checklist - All requirements met
OpenEnv Contributor75fdee9 -
docs: Add Phase 2 completion summary
OpenEnv Contributor768e96d -
feat: Add Phase 2 validation - multi-task graders with partial credit scoring
OpenEnv Contributor- Created env/graders.py with grade_easy, grade_medium, grade_hard functions - Each grader returns partial scores (0.0-1.0) instead of binary 0/1 - Implements partial credit logic for RL learning curve support - Updated openenv.yaml with grader field for each task - Updated inference.py to default to 'campaign' mode (all 3 tasks) - Added keep-alive loop after task completion for evaluator Phase 2 Requirements Met: ✓ Breadth: 3 distinct scenarios (easy S3, medium credentials, hard ransomware) ✓ Dense Rewards: Partial credit scoring (0.35-0.90 range) instead of 0/1 binary ✓ All tasks run: Campaign mode executes all 3 tasks sequentially with [START]/[END] markers736c145
-
fix(deploy): ensure HTTP response returns 200 explicitly on /reset endpoint for HF spaces
OpenEnv Contributore90b82d -
fix(build): add server entrypoint, required dependency and uv lockfile for multi-mode deployment
OpenEnv Contributor14d067c -
fix(build): Add missing pyproject.toml for multi-mode deployment support
OpenEnv Contributorf35973d -
fix(deploy): Implement dummy server to satisfy HF running state & POST requirements
OpenEnv Contributor27e6cb4 -
Update to Qwen2.5-Coder-32B-Instruct and add keep-alive loop for persistent Space
OpenEnv Contributor9080423 -
Update model to Qwen2.5-Coder-32B-Instruct
OpenEnv Contributor658ea3e -
Fix f-string syntax error in tool schemas (escape braces)
OpenEnv Contributor852d63f -
Update model to Qwen2.5-7B-Instruct and add explicit tool schemas to system prompt
OpenEnv Contributorbb65dd1 -
Update default model to Qwen2.5-Coder-32B-Instruct
OpenEnv Contributor2121404 -
fix: switch to HF's new router.huggingface.co endpoint (deprecated api-inference)
OpenEnv Contributor7853cbe -
fix: make HF_TOKEN mandatory at startup per hackathon requirements
OpenEnv Contributor5ffc5cb -
fix: switch from deprecated HF API to Together AI API (free tier)
OpenEnv Contributor0ed031d -
fix: improve HF_TOKEN error handling - print setup instructions once instead of every step
OpenEnv Contributor0c5f81b -
docs: add HF Spaces deployment guide and update Dockerfile with correct defaults
OpenEnv Contributord57d02f -
fix: make HF_TOKEN optional at startup, validate at runtime for Space compatibility
OpenEnv Contributor61cca39 -
fix: restore HF_TOKEN as mandatory per hackathon requirements
OpenEnv Contributorc24afa3 -
feat: switch to Qwen2.5-3B-Instruct with HF Inference API + optimize for 2vCPU/8GB RAM
OpenEnv Contributor9958efa -
Add Hugging Face Spaces YAML configuration to README
OpenEnv Contributord83fb27 -
Merge branch 'main' of https://github.com/dev-Adhithiya/openenv-cloudsoc
OpenEnv Contributor4e12cbb -
Security: Remove exposed HF token from public files
OpenEnv Contributor6c1b815 -
Merge Hugging Face repo: keep local version of README
OpenEnv Contributor4dbdcf8 -
initial commit
e5c42af -
Initial commit
e1025be -
Add GitHub setup documentation
OpenEnv Contributora2eb9d3 -
Initial commit: OpenEnv-CloudSOC benchmark environment
OpenEnv Contributor +1: Copilot- Complete Gymnasium environment with 24 AWS-like tools - 3 difficulty levels (easy/medium/hard) with programmatic graders - 12 advanced mechanics: deceptive environment, query costs, preconditions, adversarial traps, gradient rewards, memory pressure, tool abstraction, rich scoring, deterministic seeds, CoT prompting, multi-task campaigns, timeline grading - LLM evaluation loop with OpenAI client integration - Hackathon-compliant [START]/[STEP]/[END] output format - Comprehensive testing suite with 20+ unit tests - Interactive debugging tool - Complete documentation (9 guides) - Docker-ready for Hugging Face Spaces deployment - Zero-DB architecture (in-memory state management) - Compliant with 2 vCPU / 8 GB RAM constraints Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>115612d