Bug hunt
Reproduce first, find the real cause, and do not call it fixed until the checks say so.
3 of 4 available · 1 paused
From failing check to verified fix
Choose one investigation skill to start. These are alternatives, not four skills you need to run at once.
1. Choose your starting skill
Start here for a reproducible failure
Diagnosing bugsBuild a fast failing loop, then narrow down the cause. Also covers flaky bugs.
Available. Check the review, access labels and source before installing.
When you need a structured investigation
Systematic debuggingWork through root-cause investigation before trying a fix.
Paused: Scanner findings require review for this source version. Read the review; no install command is available.
When a change may have caused the regression
Debugging and error recoveryTriage the failure and use repository history to find where it began.
Available. Check the review, access labels and source before installing.
2. Install one and confirm it loaded
Open Diagnosing bugs’s installation instructions, or choose another available starter above. You need Node.js and npm for the Skills CLI. Choose your coding agent and project scope, then confirm the skill appears in that agent’s installed skills and ask it to identify the instructions it loaded. If it cannot, resolve installation before debugging.
Commands target the repository commits recorded with the automated scans. Scans cover only SKILL.md, not referenced files or runtime behavior. What the review covers.
3. Give the agent a useful starting prompt
Starter prompt for Diagnosing bugsUse the diagnosing-bugs skill to investigate this bug. Expected behavior: [what should happen] Actual behavior: [what happens instead] Reproduction: [steps, command, and a redacted error or log] Project and constraints: [stack, relevant paths, and what may be changed] First reproduce the failure and report the failing check. Trace the root cause with file and line references before proposing the smallest fix. Add a regression test where practical. Run the original reproduction and relevant project checks after the change. Report exact commands, results, and anything not verified. Do not merge, deploy, or change production.Replace the bracketed context and remove secrets from logs. If you chose a different available skill, change the skill name.
4. Ask for evidence, not just a fix
- A repeatable failure, or a clear account of what prevents reproduction.
- A root-cause explanation tied to code, logs or a controlled experiment.
- A small change and a regression test that fails before the fix and passes after it, where practical.
- The exact verification commands, their results and remaining gaps. A proposed test is not a passing test.
5. Verify before calling it done
Rerun the original failure, the regression test and the relevant project checks. Check nearby behavior and the final diff; leave any blocked checks visibly unverified.
For a separate finishing pass, install Verification before completion after reading its review.
Verification promptUse the verification-before-completion skill to check this fix. Rerun the original reproduction and relevant checks against the final code. Report the commands, exit results, regression coverage, and anything unverified. Do not claim success from earlier output.
Optional: install the available skills
You can start with one skill. This block collects all currently available members if you want the wider kit.
1 skill paused and excluded from the commands: Systematic debugging.
Commands target the repository commits recorded with the automated scans. Scans cover only SKILL.md, not referenced files or runtime behavior. What the review covers.
npx skills@latest add https://github.com/mattpocock/skills/tree/5863806dd4e15c63174bee4e7515a4256792ecfa --skill diagnosing-bugs
npx skills@latest add https://github.com/addyosmani/agent-skills/tree/39b723ceebf81474d0557c1fecd01eddab585563 --skill debugging-and-error-recovery
npx skills@latest add https://github.com/obra/superpowers/tree/22f0b463e7a5e579904dd6211a520321035526f7 --skill verification-before-completionEach command asks which agents to install for; add -g to install globally. Confirm installation in your chosen agent.
In this kit
Diagnosing bugs
Builds a fast failing feedback loop first, then narrows to the cause.
What it can touch: Edits code · Runs commands
Automated scan recorded · evidence and limits
- Automated scan: Checked 16 Nov 2026. Checks SKILL.md patterns and frontmatter; does not run the skill.
- Source revision: 5863806dd4e1 (opens in a new tab). The install command targets this commit. Only SKILL.md was scanned.
- Human reading: No full human-reading date recorded.
- Runtime evidence: No runtime-test evidence recorded here.
A review is not a security audit or a guarantee.
Commands target the repository commits recorded with the automated scans. Scans cover only SKILL.md, not referenced files or runtime behavior. What the review covers.
936Kinstalls+2.5K today
npx skills@latest add https://github.com/mattpocock/skills/tree/5863806dd4e15c63174bee4e7515a4256792ecfa --skill diagnosing-bugsSystematic debugging
Root cause before fixes, in four phases with clear stop signs.
What it can touch: Runs commands
Installation paused · evidence and limits
- Automated scan: Scanner findings require review for this source version.
- Source revision: 22f0b463e7a5 (opens in a new tab). Installation is paused for this revision. Only SKILL.md was scanned.
- Human reading: No full human-reading date recorded.
- Runtime evidence: No runtime-test evidence recorded here.
A review is not a security audit or a guarantee.
335Kinstalls+1.6K today
Review pausedDebugging and error recovery
A triage checklist: reproduce, localise, fix, then guard against the regression.
What it can touch: Runs commands · Uses git
Automated scan recorded · evidence and limits
- Automated scan: Checked 16 Nov 2026. Checks SKILL.md patterns and frontmatter; does not run the skill.
- Source revision: 39b723ceebf8 (opens in a new tab). The install command targets this commit. Only SKILL.md was scanned.
- Human reading: No full human-reading date recorded.
- Runtime evidence: No runtime-test evidence recorded here.
A review is not a security audit or a guarantee.
Commands target the repository commits recorded with the automated scans. Scans cover only SKILL.md, not referenced files or runtime behavior. What the review covers.
76Kinstalls+560 today
npx skills@latest add https://github.com/addyosmani/agent-skills/tree/39b723ceebf81474d0557c1fecd01eddab585563 --skill debugging-and-error-recoveryVerification before completion
No claim of done until the checks have run and the output has been read.
What it can touch: Runs commands
Automated scan recorded · evidence and limits
- Automated scan: Checked 16 Nov 2026. Checks SKILL.md patterns and frontmatter; does not run the skill.
- Source revision: 22f0b463e7a5 (opens in a new tab). The install command targets this commit. Only SKILL.md was scanned.
- Human reading: No full human-reading date recorded.
- Runtime evidence: No runtime-test evidence recorded here.
A review is not a security audit or a guarantee.
Commands target the repository commits recorded with the automated scans. Scans cover only SKILL.md, not referenced files or runtime behavior. What the review covers.
278Kinstalls+1.5K today
npx skills@latest add https://github.com/obra/superpowers/tree/22f0b463e7a5e579904dd6211a520321035526f7 --skill verification-before-completion