Debugging

SkillDev tools

Active bug investigation — reproduce, diagnose, root-cause analysis. Produces debug report for /unipi:fix.

Available today. Use it from your connected AI after setup.

Connect ahel once, and every AI you use reads what you have installed.

Then ask your AI: use the Debugging skill

What this skill tells your AI

The instructions your AI receives, as published by neuron-mr-white/unipi in packages/workflow/skills/debug/SKILL.md and read by ahel’s review.

Active investigation to reproduce, diagnose, and root-cause bugs. Produces a structured debug report that /unipi:fix can consume.

Boundaries

This skill MAY: read codebase, run diagnostic commands, spawn subagents, write debug report to .unipi/docs/debug/. This skill MAY NOT: edit code, fix issues, run tests that modify state, deploy.

Write-permission clarification: "Read-only" means source/project code is read-only for diagnosis. It does not prohibit creating or updating the debug report. The agent should use available file-writing tools to write exactly the report under .unipi/docs/debug/. If writing that report fails, report the write failure explicitly.

This is diagnosis only — not fixing.

Command Format

/unipi:debug <string(greedy)>
  • string(greedy) — bug description, error message, or reproduction steps
  • Source/code read-only sandbox, with explicit permission to write the debug report to .unipi/docs/debug/
  • Spawns subagents if @unipi/subagents extension is installed

Output Path

.unipi/docs/debug/YYYY-MM-DD-<topic>-debug.md

Process

Phase 1: Understand the Bug

  1. Read the bug description carefully
  2. If unclear, ask clarifying questions (one at a time):
    • "What's the expected behavior?"
    • "What's the actual behavior?"
    • "Steps to reproduce?"
    • "When did this start happening?"
  3. Identify the scope:
    • Single function/module
    • Integration issue
    • System-wide problem

Exit: Bug understood, scope defined.

Phase 2: Reproduce

Attempt to reproduce the issue:

  1. Find reproduction path:

    • Trace the code flow from entry point
    • Identify the failing code path
    • Look for error messages, stack traces
  2. Run diagnostic commands:

    • grep for error messages in code
    • find related files
    • Check logs if available
    • Run read-only test commands if safe
  3. If subagents available:

    • Spawn explore agents to trace different code paths in parallel
    • Each agent reports findings

Exit: Reproduction steps documented (or confirmed unreproducible).

Phase 3: Diagnose

Deep dive into root cause:

  1. Trace the failure:

    • Start from the error/symptom
    • Work backwards to find origin
    • Document each step in the chain
  2. Check common causes:

    • Null/undefined values
    • Type mismatches
    • Race conditions
    • Missing error handling
    • Incorrect assumptions
    • Off-by-one errors
    • State corruption
  3. Analyze data flow:

    • Input validation
    • Transformation steps
    • Output format
    • Edge cases
  4. Check dependencies:

    • Recent changes to related code
    • External service issues
    • Configuration changes
    • Version mismatches

Exit: Root cause identified (or hypotheses listed if uncertain).

Phase 4: Document Findings

Write debug report to .unipi/docs/debug/YYYY-MM-DD-<topic>-debug.md using the available file-writing tool. Do not skip this because the investigation is otherwise read-only:

---
title: "{Bug Title} — Debug Report"
type: debug
date: YYYY-MM-DD
severity: {critical|high|medium|low}
status: {root-caused|needs-investigation|unreproducible}
---

# {Bug Title} — Debug Report

## Summary
{One-line description of the bug}

## Expected Behavior
{What should happen}

## Actual Behavior
{What actually happens}

## Reproduction Steps
1. {Step 1}
2. {Step 2}
3. {Step 3}

## Environment
- {Relevant context: OS, version, config}

## Root Cause Analysis

### Failure Chain
1. {Entry point / trigger}
2. {Intermediate step}
3. {Failure point}

### Root Cause
{Detailed explanation of why the bug occurs}

### Evidence
- File: `{file}:{line}` — {what's wrong}
- File: `{file}:{line}` — {related code}

## Affected Files
- `{file}` — {role in the bug}
- `{file}` — {role in the bug}

## Suggested Fix
{High-level approach to fixing — NOT implementation}

### Fix Strategy
1. {Step 1}
2. {Step 2}

### Risk Assessment
- {Risk 1}: {mitigation}
- {Risk 2}: {mitigation}

## Verification Plan
How to verify the fix works:
1. {Test case 1}
2. {Test case 2}

## Related Issues
- {Link to related bugs, PRs, or discussions}

## Notes
{Any additional context, gotchas, or observations}

Phase 5: Present & Handoff

Present summary to user:

"Debug report written to .unipi/docs/debug/YYYY-MM-DD-<topic>-debug.md"

Root Cause: {brief summary} Suggested Fix: {brief summary}

Then suggest:

If root cause found:

/unipi:fix debug:YYYY-MM-DD-<topic>-debug

If needs more investigation:

"Root cause unclear. Need to investigate {area} further."

/unipi:gather-context {area}

If unreproducible:

"Cannot reproduce with current steps. Can you provide more details?"

  • Wait for user input

Differences from scan-issues

Aspect/unipi:debug/unipi:scan-issues
PurposeInvestigate specific bugFind potential issues
InputBug report / error messageScope / category
OutputDebug report with root causeIssue list with priorities
DepthDeep single-issue analysisBroad codebase scan
Handoff/unipi:fix/unipi:quick-work or /unipi:brainstorm

Notes

  • Diagnosis only — fixes happen via /unipi:fix
  • Debug reports are reusable — /unipi:fix can autocomplete them
  • Subagent support enables parallel code path tracing
  • Always document the failure chain — helps verify fixes
  • If bug is complex, suggest /unipi:consultant for expert analysis

Signals

GitHub stars
64
Forks
15
Last commit
Sep 2026
Advanced
Catalog kind
skill
Gateway key
debug-neuron-mr-white
Source
github.com/neuron-mr-white/unipi