The Agent Skills Directory

Command Execution: The skill relies on several CLI tools, including node, pnpm, cargo, and gh (GitHub CLI). These tools are used to execute a local maintenance script (scripts/pr-status.js), run tests, and manage GitHub PR states. This pattern is standard for developer-oriented automation skills.
Indirect Prompt Injection Surface: The skill is designed to ingest and analyze external data such as CI job logs and PR review comments.
Ingestion points: Data is read from files generated by the triage script, specifically scripts/pr-status/index.md, job-{id}.md, and thread-{N}.md.
Capability inventory: The agent has the ability to execute shell commands (pnpm, node, gh) based on its analysis of these files.
Boundary markers: The instructions do not explicitly define boundary markers for the external content, though they guide the agent toward specific analytical steps (prioritization and matching environment variables).
Sanitization: There is no explicit sanitization step for the content of the PR threads or logs. While this is a common surface for indirect prompt injection, the risk is inherent to the skill's primary purpose of processing PR-related data.
Environment Variable Configuration: The skill provides instructions for mirroring CI environment variables locally to reproduce failures. This is a common and necessary practice for debugging, though users should ensure that variables being mirrored do not conflict with sensitive local configurations.

pr-status-triage