Linear Triage

Workflow

1. Find the team — mcp__claude_ai_Linear__list_teams with a query.
2. Resolve the triage state ID — mcp__claude_ai_Linear__list_issue_statuses for the team; pick the entry with type === "triage".
3. List issues — mcp__claude_ai_Linear__list_issues with team, state (the triage state ID), limit: 50, includeArchived: false. Paginate if hasNextPage.
4. Categorize each issue into one of:
   • Straightforward — clear fix, unambiguous desired behavior, no open product questions
   • Needs discussion or investigation — requires a product decision, repro steps, design input, or further research
   • Stale / duplicate candidate — empty description, very old with no activity, or clearly duplicates a newer issue
5. Scope the straightforward issues — spawn parallel Agent calls with subagent_type: Explore (read-only). Split issues into groups and assign each sub-agent a group. Each sub-agent should:
   • Read the issue title, description, and any linked context
   • Return an effort rating for each issue in its group:
     • Small — well-understood, self-contained change
     • Medium — moderate complexity or cross-cutting concern
     • Large — significant scope or architectural impact
     • Unknown — needs a product decision or more repro before estimating
6. Recommend Oz delegation candidates — the Small-rated issues from step 5.
7. Confirm before delegating — present the user with a summary table of all issues proposed for delegation (issue ID, title, effort rating) and ask for explicit approval of the full list. Do not proceed until the user confirms.
8. (On confirmation) Delegate — mcp__claude_ai_Linear__save_issue with id and delegate: "Oz", in parallel.

Gotchas

• Oz is an agent, not a user. "Assign to Oz" means setting the delegate field, not assignee. Confirm with the user before delegating.
• list_issues with large limits can fail with a proxy error. Use limit: 50; if the result is too large to read inline, parse it from the persisted tool-result file (see step 3).
• Don't modify issues until the user explicitly authorizes it.
• Always identify yourself as an agent if you comment in linear. Write: "Comment by {agent} running /linear-triage"