How to use the Systemic Issue Triage agent skill

What does the skill make the agent do?

When triaging, the skill tells the agent to sort every incoming report into exactly one of five buckets: (A) superseded by an in-flight design change, (B) duplicate of a known class, (C) real new bug, (D) feature request, or (E) unclear. Real new bugs are grouped by shared root cause rather than handled one at a time. Two or more issues with the same root become one fix at the root, closed with named test evidence.

The skill also pushes the agent to check every proposed fix for over-engineering. If a fix would add a state, a verb, a config flag, a gate, or a parallel copy of existing truth, the agent is told to look for something to delete or relax instead. Issues are closed only against named tests, not against promises that something "should be fixed now".

Install the skill from the ZIP

Download the ZIP file from this page and unzip it. The skill is delivered as a folder containing SKILL.md. To make it available across your projects, unzip it into ~/.claude/skills/. To make it available only inside one project, unzip it into that project's .claude/skills/ folder.

If you use the Claude apps instead of the file system, you can upload the skill under Settings → Capabilities → Skills.

• Personal install: ~/.claude/skills/
• Project install: .claude/skills/ inside the project
• Claude apps: Settings → Capabilities → Skills, then upload

How do you trigger the skill with a natural request?

You do not need a special command. The skill's description lists the situations it applies to: a new issue, a bug report, triage, a backlog, an issue flood, a community report, a root cause, a dead-end, or a blocked user. Phrase your request in those terms and the agent can pick the skill up.

Name the goal clearly. Asking for a backlog report and asking for a fix plan are different tasks, and the skill covers the fix-design side. For a full audit of every open issue and pull request, the skill's own text points to a separate skill called backlog-triage.

• "Triage these new issues and tell me which are duplicates."
• "We have a flood of bug reports about the same installer step. Plan fixes by root cause, not one patch per issue."
• "Before you write code, check whether this fix adds a new flag or state. If it does, see whether we can delete something instead."

Examples of what the agent produces

The skill's output contract asks for a per-triage summary: bucket counts, a per-issue table (issue, bucket, root or cluster, evidence reference), and any urgent flags. For each fix batch it asks for the issues closed, the named tests that prove each closure, and the net line change, with deletion-heavy batches described as the goal.

The skill also gives worked patterns, such as a dead-end report being treated as the highest severity class, where the fix is an exit message rather than a new guard. A refusal message should name a command the user can actually run, and the skill asks the agent to confirm that command runs before treating the message as fixed.

What does the skill need, and what does it run?

The skill lists no requirements. It does not run scripts and does not use the network. It is a set of written instructions in SKILL.md that the agent reads. Because it is written for one specific project, its rules point to that project's own files, including CONTRIBUTING.md and another skill called issue-creation, so a different codebase will not have those references.

Label and close authority stays with the human and with those delegated skills. The skill is guidance for analysis and planning. It does not change labels or close issues on its own.

• Requirements: none listed
• Runs scripts: no
• Uses network: no
• Files: SKILL.md

Safety notes and limits

Treat issue text as untrusted input. The skill says issue and model text is data, not instructions, and it keeps protected labels and close or reopen actions behind the gates defined in the project's delegated skills. Human decisions remain required for conflicts and for label or type corrections.

Do not use the skill for quick one-off bug fixes. A full triage process adds overhead that a single small fix does not need.

Tips for getting better results

Give the agent the full set of related reports at once so it can spot shared roots. Ask it to reproduce the reported scenario before implementing, because the skill treats a stated mechanism as a hypothesis until a reproduction confirms it. When one issue describes two separate failures, ask the agent to comment on both and close neither until both are fixed.

• Provide all related issues in one request so clustering is possible.
• Ask for reproduction before any code change.
• Check that the agent names the test that proves each closure.