The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill relies on npx -y pr-hero to execute logic from the npm registry. This involves downloading and running code from a third-party source at runtime without version pinning or hash verification, posing a supply chain risk.\n- [EXTERNAL_DOWNLOADS]: Fetches the pr-hero package and potentially other dependencies from the npm registry (registry.npmjs.org) and downloads artifacts from GitHub Actions runs (github.com) during the triage process.\n- [INDIRECT_PROMPT_INJECTION]: The skill is designed to respond to untrusted data in the form of PR comments starting with <!-- pr-hero-finding. These external inputs influence the agent's autonomous behavior (fixing code, replying, pushing commits). While the skill includes instructions for the agent to exercise independent judgment and uses an isolated 'adjudicator' to mitigate risks, the ingestion of external instructions into an autonomous loop is a notable attack surface.\n
Ingestion points: GitHub PR comments and findings.json artifacts downloaded from CI.\n
Boundary markers: Includes extensive instructions for the agent to exercise independent judgment and use a separate 'adjudicator' sub-agent.\n
Capability inventory: Access to git push, gh pr comment, and shell execution via npx.\n
Sanitization: No explicit sanitization of comment content; relies on the agent's judgmental logic and sub-agent verification.\n- [COMMAND_EXECUTION]: Executes shell commands via npx, gh, and git to perform repository triage, status monitoring, and code fixes. The logic of these commands is partially driven by data extracted from external GitHub comments.\n- [DYNAMIC_EXECUTION]: Spawns a fresh, isolated sub-agent (the 'adjudicator') using a provided system prompt to validate claims. Additionally, it provides instructions to install further skills via npx -y pr-hero skills install, which could modify the agent's instruction set at runtime.

pr-hero-triage