The Agent Skills Directory

[COMMAND_EXECUTION]: The skill makes extensive use of ctx_execute to perform shell operations involving git, gh, and uv. Commands are constructed dynamically using PR-specific variables such as PR numbers, branch names, and file paths.\n- [INDIRECT_PROMPT_INJECTION]: The skill's primary function involves analyzing untrusted external data from Pull Requests, which creates an attack surface for indirect prompt injection.\n
Ingestion points: Processes PR_DIFF, PR_COMMENTS, and metadata fetched via gh pr view in SKILL.md.\n
Boundary markers: The instructions include a 'Confused deputy mitigation' section (9.3.2) directing sub-agents to ignore instructions embedded in the diff or comments.\n
Capability inventory: The skill can execute shell commands, read repository files, and dispatch parallel sub-agents.\n
Sanitization: Relies on natural language instructional guardrails rather than programmatic sanitization of the input content.\n- [DYNAMIC_EXECUTION]: The skill assembles and executes multi-line shell scripts at runtime for tasks such as spike traversal and CI gate verification.\n- [EXTERNAL_DOWNLOADS]: Utilizes uv to manage environments and run tools like ruff, which may fetch remote dependencies. It also retrieves data from GitHub repositories using the gh CLI.

pr-council-review