commit-work
Pass
Audited by Gen Agent Trust Hub on Oct 1, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: Employs standard git and gh CLI tools for repository operations. Security is enhanced by a requirement for 'policy-readable' commands, explicitly forbidding the use of shell wrappers, command substitution, backticks, and redirections that could hide malicious arguments from platform filters.
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from the local environment. Ingestion points: Reads code diffs via 'git diff' and project instructions via 'CLAUDE.md'. Boundary markers: None explicitly defined. Capability inventory: Includes file system staging ('git add'), local commits ('git commit'), and remote network transport ('git push'). Sanitization: Includes explicit instructions to detect and report secrets or tokens before commitment.
- [SAFE]: Implements a proactive safety check for secrets and credentials within code diffs, instructing the agent to stop and report rather than proceeding with a commit.
Audit Metadata