The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill instructs the agent to execute local scripts located at .claude/scripts/lib/commit-review-policy.cjs and .claude/hooks/lib/review-receipt.cjs. These scripts are not part of the skill's distribution and reside in the project tree, which could lead to the execution of untrusted code if the project environment is compromised.
[COMMAND_EXECUTION]: The instructions explicitly state to 'never reconfirm commit/push'. By bypassing the human-in-the-loop confirmation typically required for git operations, the skill significantly increases its autonomy, allowing it to push changes to remote repositories without user review.
[INDIRECT_PROMPT_INJECTION]: The skill has a significant attack surface for indirect prompt injection as it processes untrusted data (the code being committed).
Ingestion points: Reads the content of pending commits and project configuration files.
Boundary markers: No specific delimiters or instructions are used to distinguish between code and agent instructions during the scan.
Capability inventory: The agent can execute Node.js scripts, perform git commits, and push to remote repositories.
Sanitization: While the skill includes a safety check to scan for hardcoded secrets and .env files, it does not include measures to prevent the agent from following malicious instructions embedded within the code content itself.

commit