The Agent Skills Directory

[COMMAND_EXECUTION]: The skill makes extensive use of the gh CLI to interact with GitHub's API for fetching pull request data and submitting review results. It also executes a local TypeScript script, scripts/validate-review-payload.ts, using the bun runtime to perform structural validation of the generated JSON payloads before they are sent to GitHub.
[EXTERNAL_DOWNLOADS]: The skill fetches data from external GitHub repositories to perform its reviews. This includes retrieving pull request metadata, unified diffs, and existing review threads via the GitHub REST API or Model Context Protocol (MCP).
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection as it processes untrusted input from GitHub pull requests, such as PR descriptions and code comments.
Ingestion points: PR titles, descriptions, and code diffs are fetched from GitHub and provided to the agent for analysis as described in references/github-routing.md.
Boundary markers: The instructions do not define specific delimiters or instructions to the agent to treat the PR content as untrusted data.
Capability inventory: The agent has the ability to execute shell commands (gh) and run local scripts (bun), and it interacts with the filesystem to save and validate payloads.
Sanitization: No explicit sanitization of the PR content is performed prior to the agent's evaluation.

github-pr-review