The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from the repository (via git diff and git log) to generate commit messages and PR descriptions. This creates an attack surface where instructions embedded in the code or commit history could theoretically influence the agent's behavior.
Ingestion points: Reads local file changes and history through git diff and git log.
Boundary markers: Absent. The skill does not define specific delimiters to wrap the untrusted code content.
Capability inventory: Performs file system staging (git add), local commit creation (git commit), and remote pull request creation via the gh CLI.
Sanitization: Absent. The content of the diff is processed directly for summarization.
[SAFE]: The skill implements proactive security guardrails by requiring the agent to scan for hardcoded secrets (e.g., API keys, private keys, .env files) before finalizing a commit. Additionally, it prohibits destructive actions like force pushing or rewriting history without explicit user confirmation.

smart-commit