The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from the local repository and external model responses, creating a surface for indirect prompt injection.
Ingestion points: The skill reads code changes via git diff and pull request metadata via gh pr view as described in SKILL.md (Sections 1 and 2).
Boundary markers: The skill defines a structured response template for the subagent review (VERDICT: APPROVE or CHANGES, etc.), but it does not specify clear delimiters or "ignore instructions" warnings for the code content being analyzed.
Capability inventory: The skill has the ability to execute shell commands (git, gh) and delegate tasks to other AI models (delegate_task).
Sanitization: The instructions require the agent to independently verify the validity and severity of findings ("Check each finding and severity yourself") before applying any fixes, which serves as a manual validation step.
[COMMAND_EXECUTION]: The skill executes several shell commands to interact with the local git repository and GitHub CLI.
Evidence: The skill uses git status, git fetch, git merge-base, git diff, and gh pr view to determine the scope of changes and retrieve PR information. These are standard operations for a code review tool.

sol-loop