git-commit-writer
Pass
Audited by Gen Agent Trust Hub on Sep 18, 2026
Risk Level: SAFEREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [REMOTE_CODE_EXECUTION]: The command scripts in
commands/droid/git-commit-writer.mdandcommands/opencode/git-commit-writer.mdinclude a conditional update feature triggered by a--update-skillflag. This feature downloads and executes an installation script (install.sh) from the vendor's official repository on GitHub (Alcyone-Labs/agent-skills).\n- [COMMAND_EXECUTION]: The skill executes localgitcommands (git diff,git log) to analyze the state of the repository and identify changes to be committed.\n- [INDIRECT_PROMPT_INJECTION]: The skill processes potentially untrusted data from staged git diffs and project files likeAGENTS.md.\n - Ingestion points: Reading staged changes via
git diffand project documentation files (AGENTS.md,README.md,CONTRIBUTING.md).\n - Boundary markers: The instructions in
commands/gemini/git-commit-writer.tomlestablish a specific expert persona and strict output constraints (returning only the commit message).\n - Capability inventory: Limited to
gitread-only analysis and repository metadata access.\n - Sanitization: Diff content and project guidelines are processed directly without explicit sanitization before being provided to the model context.
Audit Metadata