Issue Triage (Live)
An untriaged backlog is a pile no one trusts. In Claude Cowork this skill reads the real open issues, classifies each, applies the safe labels and priorities in place, and hands back a report — turning a heap into a queue, while leaving the judgement calls to a human.
What This Skill Produces
- The triaged backlog — each open issue typed (bug / feature / question / duplicate), severity-tagged, and prioritised
- Applied changes — labels and priority set via the connector; duplicates linked to their canonical issue
- A triage-report artifact — what was applied, the duplicate clusters, and the issues that need a human decision
Required Inputs
Ask for these if not provided:
- The repo/project — which GitHub repo or Linear team, and the filter (all open, untriaged only)
- The label & priority scheme — existing labels and what P0–P3 mean here
- Autonomy — apply labels/priority automatically, or preview first (default: apply labels, preview closes/merges)
Framework: The Triage Pass
- Type it — bug / feature / question / duplicate / needs-info.
- Severity for bugs — blocking / major / minor, from impact and reproducibility.
- Priority — severity × reach × strategic pull → P0–P3.
- Dedupe — cluster near-identical issues; pick the canonical, link the rest.
- Needs-info — issues too thin to action get the label and a templated ask.
Execution (Cowork)
- Read issues — via the GitHub/Linear connector, list open issues in scope with bodies and existing labels.
- Classify — apply type, severity, priority using the project's scheme; detect duplicates by title/body similarity, not exact match.
- Apply the safe changes — set labels and priority through the connector; add a
needs-info comment where required; link duplicates to the canonical issue. Do not close issues automatically.
- Escalate — anything ambiguous, strategic, or close-worthy is listed for the human, not actioned.
- Emit the artifact — the triage report with applied changes, duplicate clusters, and the human queue.
Guardrails: never auto-close issues (only propose); apply only labels/priority/links automatically; use the project's real label scheme, not invented labels; if the connector is unauthorised, produce the triage plan without applying and say so.
Output Format
An Issue Triage Report:
Summary
N open · B bugs · F features · Q questions · D duplicate clusters · applied to M
Applied
| Issue |
Type |
Severity |
Priority |
Labels added |
Duplicate clusters
Needs a human
- #N — why (close-worthy / strategic / ambiguous)
Quality Checks
Anti-Patterns
- Auto-closing issues — propose, don't close.
- Inventing labels the project doesn't use.
- Exact-match dedupe that misses reworded duplicates.
- Labelling a thin issue P2 instead of asking for repro/info.
Example Trigger Phrases
- "Triage my open GitHub issues in Cowork."
- "Label and prioritise the backlog, and cluster the duplicates."
- "Sort my Linear issues — types, severity, priority."
- "Clean up the issue tracker and flag what needs me."
1---2name: issue-triage-live3description: Triage the user's REAL issue tracker — read open issues via the GitHub/Linear connector, label / prioritise / dedupe / flag them, and apply the safe changes — not advice on triage. Use when asked to triage my issues, clean up the backlog, label and prioritise open issues, or sort my GitHub issues in Cowork. Reads open issues via the connector, classifies by type / severity / duplicate, applies labels and priorities, and produces a triage-report artifact with the applied changes and the ones needing a human call.4---5
6# Issue Triage (Live)
7
8An untriaged backlog is a pile no one trusts. In Claude Cowork this skill reads the *real* open issues, classifies each, applies the safe labels and priorities in place, and hands back a report — turning a heap into a queue, while leaving the judgement calls to a human.
9
10## What This Skill Produces
11
12- **The triaged backlog** — each open issue typed (bug / feature / question / duplicate), severity-tagged, and prioritised
13- **Applied changes** — labels and priority set via the connector; duplicates linked to their canonical issue
14- **A triage-report artifact** — what was applied, the duplicate clusters, and the issues that need a human decision
15
16## Required Inputs
17
18Ask for these if not provided:
19- **The repo/project** — which GitHub repo or Linear team, and the filter (all open, untriaged only)
20- **The label & priority scheme** — existing labels and what P0–P3 mean here
21- **Autonomy** — apply labels/priority automatically, or preview first (default: apply labels, preview closes/merges)
22
23## Framework: The Triage Pass
24
251. **Type it** — bug / feature / question / duplicate / needs-info.
262. **Severity for bugs** — blocking / major / minor, from impact and reproducibility.
273. **Priority** — severity × reach × strategic pull → P0–P3.
284. **Dedupe** — cluster near-identical issues; pick the canonical, link the rest.
295. **Needs-info** — issues too thin to action get the label and a templated ask.
30
31## Execution (Cowork)
32
331. **Read issues** — via the GitHub/Linear connector, list open issues in scope with bodies and existing labels.
342. **Classify** — apply type, severity, priority using the project's scheme; detect duplicates by title/body similarity, not exact match.
353. **Apply the safe changes** — set labels and priority through the connector; add a `needs-info` comment where required; link duplicates to the canonical issue. **Do not close** issues automatically.
364. **Escalate** — anything ambiguous, strategic, or close-worthy is listed for the human, not actioned.
375. **Emit the artifact** — the triage report with applied changes, duplicate clusters, and the human queue.
38
39Guardrails: never auto-close issues (only propose); apply only labels/priority/links automatically; use the project's real label scheme, not invented labels; if the connector is unauthorised, produce the triage plan without applying and say so.
40
41## Output Format
42
43An **Issue Triage Report**:
44
45### Summary
46`N open · B bugs · F features · Q questions · D duplicate clusters · applied to M`
47
48### Applied
49| Issue | Type | Severity | Priority | Labels added |
50|---|---|---|---|---|
51
52### Duplicate clusters
53- canonical #X ← #Y, #Z
54
55### Needs a human
56- #N — why (close-worthy / strategic / ambiguous)
57
58## Quality Checks
59- [ ] No issue was auto-closed
60- [ ] Labels/priorities use the project's real scheme
61- [ ] Duplicate clusters name a canonical issue and link the rest
62- [ ] Priority reflects severity × reach, not gut feel
63- [ ] Ambiguous/strategic issues were escalated, not force-labelled
64
65## Anti-Patterns
66- **Auto-closing** issues — propose, don't close.
67- **Inventing labels** the project doesn't use.
68- **Exact-match dedupe** that misses reworded duplicates.
69- **Labelling a thin issue P2** instead of asking for repro/info.
70
71## Example Trigger Phrases
72- "Triage my open GitHub issues in Cowork."
73- "Label and prioritise the backlog, and cluster the duplicates."
74- "Sort my Linear issues — types, severity, priority."
75- "Clean up the issue tracker and flag what needs me."