Skip to content
Back to skills

Do Patch

ASecurity

Diagnose and fix a test failure, regression, or review blocker with a focused change and causal verification.

  • 26 stars
  • 0 votes
  • 0 copies
  • 0 views
  • Added September 23, 2026
ai-agentsrefactoringgit

Security analysis

A100/100

Scanned September 23, 2026

npx -y skills add tomcounsell/ai --skill do-patch --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Do Patch?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for Do Patch
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/tomcounsell-do-patch-ai/badge)](https://www.skillsdirectory.com/skills/tomcounsell-do-patch-ai)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: do-patch
description: "Diagnose and fix a test failure, regression, or review blocker with a focused change and causal verification."
---

# Do Patch

Recover the original plan/issue, current branch/worktree, previous changes, full failure output, and relevant review comments. Verify the actual failing path before editing.
Trace input to expected output through component boundaries and identify where real values diverge. For a behavioral bug, reproduce it with a meaningful failing check when feasible; it must fail because of the defect, not setup. Fix the cause with the smallest coherent change, including directly affected integration points. Do not bury the failure with mocks, weakened assertions, or unrelated refactoring.
Run the regression and relevant surrounding checks; compare expected and observed values after the fix. Keep plan completion boxes consistent with demonstrated behavior. Honor the caller's commit/PR ownership when delegated; standalone work can commit according to the current workflow. Every patch is a fresh commit; never `git commit --amend`, since amending destroys the commit a reviewer already judged and defeats head-SHA freshness gates.
After repeated failures, stop repeating the same hypothesis and use $zoom-out or additional investigation. Report the cause, changed behavior, validation, and remaining uncertainty. A patch changes the reviewed head, so a managed lane needs a fresh review before merge. This procedure never opens a pull request.

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments

Loading comments…