andreaswasita/copilot-agents-dojo/autonomous-bug-fix
Claude Code and agent skill by andreaswasita/copilot-agents-dojo
- community
Checking that the SKILL.md file is still in the repository.
Reproduces, diagnoses, fixes, and verifies bugs unaided.
Install
npx vskill@latest install andreaswasita/copilot-agents-dojo/autonomous-bug-fixSKILL.md
Autonomous Bug Fix Skill
Walks the agent through the full reproduce → diagnose → fix → verify cycle on its own, without asking the user for navigation help, log-reading instructions, or test commands. Does NOT cover net-new feature work — for that use plan-before-code and executing-plans.
When to Use
- User reports a bug, regression, or unexpected behavior.
- CI is red on a branch you own.
- An error or warning appears in logs.
- A previously working feature stopped working.
- The user says "this is broken," "it doesn't work," or similar.
Prerequisites
- Read access to the failing component (use
glob/grepto locate it). - The
powershelltool to run tests,git, and reproduction commands. - A clean working tree on a feature branch (use
using-git-worktreesif needed). verify-before-doneskill loaded — you'll invoke it at the end.
How to Run
Quick Reference
| Phase | Action | Tool | |
|---|---|---|---|
| Find code | Search by symbol or file pattern | grep, glob | |
| Read code | Inspect the failing path | view | |
| Reproduce | Run failing command/test | powershell | |
| Diagnose | Recent changes / blame | powershell → git log -20 --oneline, git blame <path> | |
| Edit | Minimal, root-cause fix | edit | |
| Add test | Regression test | create or edit under tests/ | |
| Verify | Re-run reproduction + full suite | powershell, then verify-before-done |
Procedure
Step 1: Reproduce
Find the relevant code yourself — grep for the symbol, glob for the path. Do not ask the user "which file is this in?"
Run the failing path via the powershell tool. If no failing test exists, write a minimal one that captures the bug. Confirm reproduction with concrete evidence (stack trace, exit code, log line).
If you can't reproduce:
- Check for environment-specific behavior (OS, Node/Python version).
- Look for race conditions or state-dependent paths.
- Ask the user for clarification only as a last resort, and only with specific questions.
Step 2: Diagnose Root Cause
Read the error message and stack trace before doing anything else — they exist for a reason. Trace from the user-visible symptom back to the root cause:
Stop at the root cause, not the first thing that looks suspicious. "Why did this break?" — not "what broke?"
Step 3: Fix at the Root
Edit with precision via the edit tool:
- Change only what needs to change. Bigger diffs hide bigger bugs.
- Fix the root cause, never the symptom. Swallowing an exception is not a fix.
- Add a regression test that fails without the fix and passes with it.
- Check whether the same root cause exists elsewhere (
grepfor the pattern). If yes, fix those too or log them intasks/lessons.mdfor follow-up.
Step 4: Verify
Hand off to the verify-before-done skill:
- Re-run the reproduction. The error must be gone.
- Run the full test suite. No regressions.
- Record evidence in
tasks/todo.mdunder the task.
Step 5: Rollback If Verification Fails
If verification fails after your fix:
- Reset the working tree (
git checkout -- .orgit stashviapowershell). - Log the failed attempt in
tasks/lessons.mdwith the root-cause analysis you had. - Re-plan with a different approach via
plan-before-code. - Never push broken code hoping someone catches it later.
Pitfalls
- DO NOT ask the user "which file is it in?" or "how do I run the tests?" Use
grep/glob/view/powershellto discover it. - DO NOT apply a band-aid (catch-and-swallow, ignore the warning). Fix the root cause.
- DO NOT fix the symptom. A login button that fails because of a missing provider is a provider bug, not a button bug.
- DO NOT skip the regression test. Without one, the bug comes back in 3 months.
- DO NOT push without verification. An unverified fix is often worse than no fix.
- DO NOT abandon failed attempts silently. Log them in
tasks/lessons.md.
Verification
- [ ] Reproduction was confirmed with concrete evidence before any fix.
- [ ] Root cause is identified in writing (in commit message or
tasks/todo.md). - [ ] Regression test added that fails on
mainand passes on the branch. - [ ]
verify-before-donechecklist passed. - [ ] If similar patterns exist elsewhere, they are fixed or logged.
Security
- Security scanPassed, 100 of 100Pattern checks on the skill's files.
- Model reviewNot runA model review is scheduled for the next rescan.
Quality evaluation
- Task evaluationNot runNo evaluation yet. Run one in Skill Studio.
Source
- SKILL.md in the repositoryCheckingskills/autonomous-bug-fix/SKILL.md
Popularity
- Repository stars
- 32
- Repository forks
- 12
- 7-day trend score
- 17
Stars and forks are for the whole GitHub repository, not this skill alone. Refreshed 30 Apr 2026, 6 months ago.
Works with
Any agent vskill installs to: 56 targets in all. These 39 have a page on this site.
- Claude Code
- Cursor
- GitHub Copilot
- Windsurf
- Codex
- Gemini CLI
- Cline
- Amp
- Roo Code
- Goose
- Continue
- Kilo Code
- Junie
- Trae
- OpenHands
- Qwen Code
- AI Agent
- Tabnine Chat
- Sourcegraph Cody
- Amazon Q
- Firebase Genkit
- Sweep
- Devin
- Phind
- Mentat
- Claude Desktop
- Aider
- GPT Pilot
- Plandex
- bolt.new
- v0
- Replit Agent
- Anse
- TypingMind
- BetterChatGPT
- LibreChat
- Warp AI
- Pieces
- Fig AI
Embed badge
[](https://verified-skill.com/skills/andreaswasita/copilot-agents-dojo/autonomous-bug-fix)