git-release
Pass
Audited by Gen Agent Trust Hub on Oct 8, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes data from potentially untrusted local sources, including Git metadata and project-level documentation files (e.g., AGENTS.md, CONTRIBUTING, .gitflow/workflow.json).
- Ingestion points: The skill reads repository state via 'git' commands and parses project documentation in 'references/contract.md'.
- Boundary markers: The 'SKILL.md' instructions require the agent to distinguish between facts, rules, and suggestions, providing a logical boundary.
- Capability inventory: The skill can execute local Git commands and suggests paths for other tools, but requires explicit user authorization for state-changing actions.
- Sanitization: The 'inspect_repository.py' script cleans the environment of sensitive Git configuration variables and uses structured list arguments for subprocess calls.
- [COMMAND_EXECUTION]: The skill uses a Python script to execute local 'git' commands for repository inspection.
- Evidence: Found in 'scripts/inspect_repository.py' using 'subprocess.Popen'.
- Evaluation: The implementation is robust, using non-shell execution (list of arguments), clearing sensitive 'GIT_*' environment variables, and implementing strict timeouts and output limits to prevent resource exhaustion or credential leakage.
Audit Metadata