Automated GitHub issue triage and labeling

Crongent triages GitHub issues on a schedule: an agent reads new issues, applies labels from your taxonomy, flags likely duplicates, asks for missing reproduction steps, and leaves everything it is unsure about for a human. Your rules live in a reusable skill, so every repo triages the same way.

The problem

Untriaged issues pile up and real bugs hide among questions and duplicates.

How to set it up in Crongent

  1. Write your label taxonomy as a skill.
  2. Create an hourly routine allowed to label and comment.
  3. Review the transcript for anything it skipped.

Example routine

Name
Issue triage
Trigger
Hourly
Model
anthropic/claude-haiku-5.5
Instructions
For issues opened since the last run: apply type and area labels, link likely duplicates, and ask for repro steps if missing.

Who uses it

Frequently asked questions

Will it re-label issues humans already triaged?
Tell it not to in the instructions; the transcript shows exactly what it changed so you can tune it.

Related use cases