The Agent Skills Directory

[COMMAND_EXECUTION] (LOW): The skill is designed to interact with the local filesystem and shell by executing git commands and repository-specific verification scripts (tests, lint, or build).
Evidence: Workflow step 7 in SKILL.md explicitly instructs the agent to 'Run the repo's fastest meaningful check (unit tests, lint, or build) before moving on.'
[PROMPT_INJECTION] (LOW): The skill processes untrusted data from the user's codebase, which creates an indirect prompt injection surface. Malicious instructions within code comments or files could attempt to influence the agent's behavior during the commit process.
Ingestion points: git diff and git diff --cached in SKILL.md steps 1 and 4.
Boundary markers: Absent. The instructions do not explicitly tell the agent to ignore instructions embedded within the diff output.
Capability inventory: Local command execution via git and arbitrary test scripts.
Sanitization: Absent; however, the skill includes a self-check step (Workflow step 4) to manually look for secrets and debug logs before committing.

commit-work