The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill facilitates the ingestion and processing of untrusted data from external sources, specifically GitHub Pull Request labels, descriptions, and automated review comments from CodeRabbit.
Ingestion points: Data enters the agent's context through gh pr view outputs and the review threads generated by external tools (CodeRabbit) as described in the Post-push / PR checklist and Updating an existing PR sections.
Boundary markers: The instructions do not specify any delimiters (e.g., XML tags or triple quotes) to wrap external data or explicit 'ignore instructions' warnings when parsing PR comments or labels.
Capability inventory: The skill allows the agent to execute shell commands via the GitHub CLI (gh), Git (git), and Node.js (npm), as well as perform file system operations to create blog posts (website/blog/*.mdx) and roadmap updates.
Sanitization: While the skill mentions delegating substantial reviews to a specialized agent that 'knows how to parse CR threads,' there are no explicit sanitization or validation steps within this orchestration skill itself to prevent malicious instructions embedded in PR data from being executed.

pull-request