The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted content from GitHub Pull Requests, Issues, and external documentation (Ingestion points). It implements a strong security boundary by explicitly instructing the agent to disregard any instructions or commands found within these untrusted external sources (Boundary markers). Its capabilities include executing git, test, lint, and build commands (Capability inventory). Sanitization is performed by the main agent which filters sub-agent findings and verifies the absence of secrets and unplanned file changes (Sanitization).
[DATA_EXFILTRATION]: The skill incorporates a mandatory security verification step to ensure that secrets, credentials, and unnecessary local information are not present in the code diffs being reviewed.
[COMMAND_EXECUTION]: The workflow requires executing build, test, and lint commands. While this involves shell interaction, it is a primary function of the skill's code validation purpose and the skill restricts these operations to the specific branch and SHA being audited.

review-branch