The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill communicates with Bitbucket's official API (api.bitbucket.org) to retrieve pull request data and comments. This is a standard integration with a well-known service.
[COMMAND_EXECUTION]: Executes local git commands (git remote, git rev-parse) to determine repository context and invokes project-specific verification scripts (e.g., npm run lint) after applying code changes. These are standard developer operations.
[SAFE]: Recommends managing credentials via .env files or environment variables, which aligns with standard security practices for secret management.
[SAFE]: Includes platform-specific configuration instructions (e.g., --sandbox danger-full-access on macOS) specifically to enable the network connectivity required for the Bitbucket API.
[PROMPT_INJECTION]: The skill identifies and processes external data from Bitbucket PR comments to derive and apply source code modifications, creating a surface for indirect instructions to influence AI behavior.
Ingestion points: External comments fetched from the Bitbucket API in scripts/coderabbit-fetch.mjs.
Boundary markers: None; the skill parses raw markdown content from comment bodies.
Capability inventory: The skill has permissions to write to local files and execute shell commands (npm run lint).
Sanitization: The extraction script uses regular expressions to identify CodeRabbit-specific markdown blocks, but does not sanitize the content against adversarial instructions designed to bypass the AI's logic during the code generation step.

coderabbit-fix