The Agent Skills Directory

[PROMPT_INJECTION]: The skill documents architectures that ingest untrusted external data, creating a potential surface for indirect prompt injection.
Ingestion points: The 'Ralphinho' pattern ingests RFC and PRD documents, and the 'Continuous Gemini' pattern ingests CI logs via gh run view.
Boundary markers: The provided documentation does not specify the use of delimiters or 'ignore embedded instructions' warnings for these external inputs.
Capability inventory: The loops involve powerful capabilities including shell command execution (bash), Git operations (gh, git worktree), and file system writes.
Sanitization: The instructions do not mention sanitizing or validating the content of external specifications or logs before processing.
[EXTERNAL_DOWNLOADS]: The skill references external tools such as continuous-gemini and gh for GitHub integration.
Evidence: The documentation explicitly mentions installing continuous-gemini and provides examples of using gh pr create and gh run list.
Note: The skill includes a security warning advising users to review external code and avoid piping scripts directly to bash, demonstrating awareness of remote code execution risks.

autonomous-loops