How to triage a messy Linear backlog in 10 minutes
언제 쓸까: Monday morning. The backlog has 200+ untriaged issues and planning is in an hour.
사전 조건
- Linear workspace access — OAuth through
mcp-remote— first tool call opens browser to grant - Team slug (e.g.
ENG) — Look at any issue identifier — the prefix is your team slug
흐름
-
Pull every untriaged issue for the teamList all ENG issues in state 'Triage' or with no priority set, created in the last 30 days. Return id, title, reporter, and description length.✓ 복사됨→ Table of candidates with enough context to classify
-
Cluster by themeGroup these into 4-6 themes (bug / infra / onboarding / perf / etc.). For each cluster, propose a priority and suggested project.✓ 복사됨→ Themed clusters with rationale
-
Apply the decisionsFor each cluster, update the issues: set priority, add the matching label, and move to 'Backlog' state. Do NOT assign anyone.✓ 복사됨→ N issues updated, confirmations logged
결과: A planning-ready backlog, grouped and prioritized, with an audit trail of every change the agent made.
함정
- Agent mass-assigns issues to the wrong people based on 'who wrote similar code' — Explicitly tell it NOT to assign — assignment is a human call and should stay that way
- Free-tier Linear rate-limits you after ~1500 requests/h — Batch updates with
issueBatchUpdatewhen available; otherwise keep triage runs under 500 issues