Triage lets a maintainer manage an issue tracker in natural language. It shows what needs attention in three buckets, then takes one issue or PR at a time: it gathers context, recommends a category and state, verifies the claim, grills if the request needs shaping, and applies the outcome.
Every comment it posts starts with a disclaimer that it was generated by AI during triage.
What You Get
- A prioritized view of unlabeled, needs-triage and needs-info items
- A codebase check for an existing implementation and for earlier rejections in .out-of-scope
- A reproduction of the reported bug or a check of what a PR diff does
- An agent brief or triage notes posted as a comment
Common Use Cases
- Work through a backlog of unlabeled issues each morning
- Decide whether an external PR is ready for an agent or a human
- Close an already-built request with a pointer to where it lives
What Makes It Unique
It verifies the claim before any grilling, so a confirmed reproduction strengthens the agent brief. It keeps an .out-of-scope knowledge base for rejected enhancements, separates already-implemented requests from rejected ones, and lets the maintainer override a state directly.
Under The Hood
A SKILL.md with linked reference files for writing agent briefs and for the out-of-scope knowledge base. It reads the issue tracker through whatever setup-matt-pocock-skills configured, reads the codebase and domain glossary, can check out a PR and run tests, posts comments, applies labels and closes issues. It calls the grilling and domain-modeling skills when a request needs fleshing out.