The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes Git commands to generate diffs and verify file tracking within a repository context. Evidence: scripts/package_phase_review.py calls subprocess.run for git ls-files and git diff. These calls are restricted to repository-relative paths with validation against directory traversal.
[INDIRECT_PROMPT_INJECTION]: The skill packages arbitrary artifacts and code diffs for external LLM review, creating an inherent attack surface for instructions embedded in untrusted data. 1. Ingestion points: scripts/package_phase_review.py collects data from evidence_artifacts, progress_context_artifacts, and changed_files into review ZIP bundles. 2. Boundary markers: The templates/PHASE_REVIEW_REQUEST.md template instructs reviewers to prioritize deterministic evidence artifacts over prose implementation claims. 3. Capability inventory: The skill has file system access, ZIP creation, Git command execution in scripts/package_phase_review.py, and local Unix socket communication. 4. Sanitization: The system performs SHA-256 integrity verification in scripts/phase_status.py and validates that all processed paths are relative and free of traversal characters.

plan-iterate