The Agent Skills Directory

[Command Execution] (LOW): The skill executes bun run pre-commit, which runs arbitrary scripts defined in the project's local configuration. While this is standard for a development workflow, it executes code based on the project's state.
[Indirect Prompt Injection] (LOW): The skill analyzes local Markdown files and git history to generate documentation and PR descriptions, creating a surface for indirect injection.
Ingestion points: Project *.md files, git log output, and git diff output are read and analyzed by subagents.
Boundary markers: No explicit delimiters are used when passing file content to subagents for analysis.
Capability inventory: Includes git add, git commit, git push, gh pr create, and bun run pre-commit.
Sanitization: The skill correctly uses quoted HEREDOCs (cat <<'EOF') when executing git commit and gh pr create. This prevents the shell from interpreting any special characters or commands that might be present in the generated commit messages or PR bodies.
[Data Exposure & Exfiltration] (SAFE): The skill performs network operations (git push, gh pr create), but these are limited to the project's configured remote repository and are consistent with the skill's primary purpose.

finish