devtu-auto-discover-apis
Pass
Audited by Gen Agent Trust Hub on Apr 16, 2026
Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTIONREMOTE_CODE_EXECUTION
Full Analysis
- [PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection because it extracts API metadata from external web documentation and search results to generate code and configurations. \n
- Ingestion points: External web search results and scraped API documentation (SKILL.md, Step 1.3). \n
- Boundary markers: Human approval gates are mandated at four critical stages: post-discovery, post-creation, post-validation, and pre-PR (SKILL.md 'Quality Gates' and REFERENCE.md). \n
- Capability inventory: The skill writes Python scripts and JSON files, executes Python tests, and performs Git/GitHub CLI operations (SKILL.md, Phases 2, 3, and 4). \n
- Sanitization: Relies on manual human review of artifacts; no automated sanitization or filtering of scraped documentation is specified. \n- [REMOTE_CODE_EXECUTION]: The skill dynamically generates Python tool classes and JSON configurations derived from external sources and templates. These generated scripts are subsequently loaded and executed during the validation phase to verify API connectivity and schema compliance. \n- [COMMAND_EXECUTION]: The agent executes multiple shell commands to facilitate the development workflow, including running internal test scripts (python scripts/test_new_tools.py) and managing version control operations (git commit, git push, and gh pr create).
Audit Metadata