The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local test suites using identified runners (e.g., npm, pytest, go test). This is the intended behavior of the skill and is required to surface test results and coverage metrics.
[PROMPT_INJECTION]: The skill processes untrusted data from pull request diffs and comments (Indirect Prompt Injection).
Ingestion points: Reads code changes via gh pr diff and git diff, and analyzes PR comments retrieved via gh api.
Boundary markers: Explicit delimiters for the injected content are not defined in the instructions.
Capability inventory: The agent can run arbitrary shell commands through various package managers and version control tools specified in allowed-tools.
Sanitization: The skill incorporates a detailed trust-policy.md which mandates verifying the commenter's organizational membership or repository permissions before acting on any feedback, providing a strong defense against malicious instructions from third-party contributors.
[DATA_EXFILTRATION]: The skill queries the GitHub API to determine the trust status of users and retrieves code differences. These operations use well-known services (GitHub) for legitimate metadata retrieval and do not represent a security risk.

review-test