The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/sync_docs.py uses subprocess.run to execute git diff and git status to identify changes. The commands use fixed argument lists without shell interpolation, mitigating command injection risks.
[INDIRECT_PROMPT_INJECTION]: The skill processes markdown files from the repository which could contain untrusted instructions if the repository is shared or public.
Ingestion points: Files are read by scripts/audit_repo.py and scripts/sync_docs.py during scanning and validation.
Boundary markers: Absent; the scripts read full file contents into memory for regex analysis.
Capability inventory: The skill uses git status tools and relies on the agent's native file-writing capabilities for remediation.
Sanitization: None; however, the script logic is restricted to structural validation (regex) rather than execution of file content.
Mitigation: The skill's workflow instructions explicitly forbid auto-fixing and mandate using a question tool to obtain user approval for every individual change, preventing automated exploitation of the agent's capabilities.

repo-maintenance