triage
Triage a GitHub issue and present findings on validity, code locations, and optionally create a failing test or resolution plan.
Install / Use
npx skills add mend-detection-qa/payloadInstalls into whichever agent you are using.
Claude Commands
Claude Code slash commands
Quality Score
Category
Development & EngineeringSupported Platforms
Skill content
View source on GitHubdescription: Triage a GitHub issue and present findings on validity, code locations, and optionally create a failing test or resolution plan. argument-hint: <issue-number-or-url> allowed-tools: Bash(gh issue view:*), Task, TodoWrite, Write, AskUserQuestion
Triage GitHub Issue
Quickly triage GitHub issues to determine validity and identify where the issue exists in code. Do not write code during triage.
Process
Step 1: Validate & Fetch
- Check if
$ARGUMENTSprovided - if missing, ask user and stop - Create TodoWrite with: Fetch and validate issue data | Investigate codebase | Analyze validity and generate report | Offer next steps
- Fetch issue data:
gh issue view $ARGUMENTS --json author,title,number,body,comments,labels,url - If command fails: Inform user and stop
- Extract title, body, comments, labels, state, URL
- Verify issue has sufficient info to investigate
Step 2: Investigate
- Use Task tool with subagent_type=Explore to examine relevant parts of codebase
- Think step by step through the code paths involved until you understand the relevant files, functions, and logic
- Focus on understanding the problem space before judging validity
Step 3: Analyze & Report
- Think step by step about whether this issue is valid - be critical and consider alternative explanations
- Determine verdict: Valid | Invalid | Needs Info
- Determine confidence: High | Medium | Low
- If valid: Think deeply about the root cause before proposing fix direction
- Generate findings using output structure below
- Auto-save to
.claude/artifacts/triage-<issue-number>.md - Present findings to user
Step 4: Offer Next Steps
Use AskUserQuestion with multiSelect: true:
"What would you like to do next?"
- Create failing test - Write integration test that reproduces the issue
- Create resolution plan - Generate full plan using
/superpowers:write-plan - Done for now - End triage
If user selects test: Write an integration test that reproduces the issue.
If user selects plan: Use /superpowers:write-plan to create detailed implementation plan.
Output Structure
# Triage: #<issue-number> - <title>
**Verdict:** Valid | Invalid | Needs Info
**Confidence:** High | Medium | Low
## Problem
1-2 sentences on what's broken and why.
## Reproduction
Steps to reproduce from investigation:
1. ...
2. ...
## Code Locations
- `path/file.ts:123` - what's here
- `path/other.ts:456` - related
## Fix Direction
Brief hints on potential approach (not full plan):
- Consider X
- May need to change Y
## Issue Link
<url>
When to Stop
STOP immediately when:
$ARGUMENTSnot provided (ask for it)gh issue viewcommand fails- Issue not found or inaccessible
- Cannot understand issue after investigation
- Insufficient information to determine validity
Ask for clarification rather than guessing.
Remember
- Create TodoWrite with 4 steps upfront
- Mark steps in_progress before starting, completed after finishing
- Use Explore agent for investigation
- Be critical about issue validity
- Auto-save findings to
.claude/artifacts/ - Single question at end with multi-select options
- DO NOT write code during triage (tests come after, if requested)
- Stop when blocked, don't guess
Related Skills
claude-mem
92.6kPersistent Context Across Sessions for Every Agent – Captures everything your agent does during sessions, compresses it with AI, and injects relevant context back into future sessions. Works with Claude Code, OpenClaw, Codex, Gemini, Hermes, Copilot, OpenCode + More
Understand-Anything
81.1kGraphs that teach > graphs that impress. Turn any code into an interactive knowledge graph you can explore, search, and ask questions about. Works with Claude Code, Codex, Cursor, Copilot, Gemini CLI, and more.
Agent-Reach
76.7kGive your AI agent eyes to see the entire internet. Read & search Twitter, Reddit, YouTube, GitHub, Bilibili, XiaoHongShu — one CLI, zero API fees.
ruflo
69.8k🌊 The original agent meta-harness. Deploy intelligent multi-player swarms, coordinate autonomous workflows, and build conversational AI systems. Features adaptive memory, self-learning intelligence, RAG integration, and native Claude Code / Codex / Hermes and many more Integrated
Security Score
Audited on Jul 1, 2026
