The Agent Skills Directory

[COMMAND_EXECUTION]: The skill defines and uses a shell alias ci_fb that executes local Go code via go run . within the repository's .github/skills/ci-feedback-loop directory. This is used for core orchestration tasks like pausing CI, reading findings, and tagging PRs.
[INDIRECT_PROMPT_INJECTION]: This skill ingests and processes untrusted data from external sources, specifically GitHub pull request diffs and Copilot review comments.
Ingestion points: PR diffs and Copilot finding summaries are read into the agent context in steps 2 and 5.
Boundary markers: None explicitly defined in the instructions for separating untrusted review content from the agent's instructions.
Capability inventory: The skill has significant capabilities including file writing (pushing fixes), network operations (GitHub API calls), and CI management (canceling/rerunning jobs).
Sanitization: There is no evidence of sanitization for the review comments before the agent triages them or interpolates them into API calls.
[EXTERNAL_DOWNLOADS]: The skill depends on standard external tools like gh (GitHub CLI) and go (Go programming language) being pre-installed on the host system to function.

review-loop