The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from GitHub pull request metadata (body, comments, reviews), issue descriptions, and source code diffs.
Ingestion points: Pull request details and issue content are fetched using gh pr view and gh issue view commands as described in Step 1.
Boundary markers: The instructions lack explicit delimiters or specific directives for the agent to isolate untrusted content from the skill's own operational logic.
Capability inventory: The agent can execute shell commands for testing (Step 3) and perform network-based actions such as gh pr comment and gh pr review (Step 7).
Sanitization: No specific validation or sanitization routines are defined for data ingested from pull requests before it is processed or interpolated into review drafts.
[COMMAND_EXECUTION]: The skill extensively utilizes command-line interfaces to interact with the repository and environment.
Uses git fetch, git worktree, and gh tools to manage and inspect pull request data.
Executes yarn install, yarn build, yarn mocha, and yarn jest to build and test code changes directly from pull request branches.
Employs rg --hidden for searching repository content and pnpm kb for project-specific testing.
[EXTERNAL_DOWNLOADS]: The skill fetches code and metadata from external GitHub repositories.
Downloads pull request head branches from the ianstormtaylor/slate and udecode/plate repositories on GitHub to perform local analysis and testing.
[REMOTE_CODE_EXECUTION]: The skill retrieves code from external pull request heads and executes it locally using test runners.
Fetches content from remote branches and runs yarn mocha or yarn jest on that code in a detached worktree to verify findings.
[DYNAMIC_EXECUTION]: The skill involves the runtime execution of tests and potentially scripts associated with external pull requests.
Uses yarn and mocha/jest to execute test files retrieved from pull requests, which can involve running arbitrary code defined by the PR author.

review-comment