Advanced Bug Fixing Workflow
Paste the following prompt into your AI chat to install this skill:
Please follow https://skillhub.cn/install/skillhub.md and install @user_458438e1/bug-fix-farming.
About this skill
What It Solves
Bug fixing often stalls between vague symptoms, ad-hoc guesses, and unconfirmed code changes. This skill turns defect repair into an inspectable engineering flow: build a bug card with symptoms, expected behavior, impact scope, and acceptance criteria, then proceed through minimal repro, root-cause analysis, repair, tests, and CodeReview. It fits online defects, regression issues, and post-CI-failure code investigation, especially when teams want reusable repair steps.
How It Works
The skill uses a step-file architecture, loading only the current step file to reduce context bloat. Its core principle is AI analyzes and proposes, while humans confirm and decide: AI must not bypass mandatory checkpoints to modify code. Key checkpoints include:
- CP1: confirm the repro script, test case, and run result match the reported bug.
- CP2: confirm candidate root causes, evidence chain, and final diagnosis direction.
- CP3: confirm the fix plan, including affected files, change approach, and risk assessment; if rejected, AI must provide at least two alternatives.
- CP4: confirm test coverage, edge cases, and run results.
Users can use y, r, or s to confirm, reject, or skip allowed steps; at CP3, p can trigger multi-perspective discussion and e can expand the fix scope. The final output is a fixed Bug Fix Report containing bug card, reproduction and root cause, fix and tests, review gate, and conclusion.
Boundaries
It is not for 'write a new feature,' generic code critique, or 'explain this module.' Execution forbids fixing before reproduction, treating a successful CI rerun as root-cause repair, or patching the same module repeatedly without global review. If a high-risk regression appears, prefer rollback and add regression tests first.
Use Cases
- A production API intermittently returns 500; reproduce minimally, identify root cause, and produce a review-ready fix report.
- CI fails and looks like a regression; decide whether code is defective, then run repair and test checkpoints.
- A payment callback never fires; convert symptoms, expectations, and impact into a bug card before proceeding.
- The proposed fix is challenged; require two alternative repair paths and wait for human confirmation.
Best For
- Backend engineers responsible for production stability who need to drive 500s, timeouts, or regressions to closure.
- Developers triaging CI failures who want to distinguish code defects from flakiness and add tests.
- Tech leads reviewing junior or contractor changes who need auditable human checkpoints before edits.
- Product engineers on payment or order risks who require reviewable and rollback-ready fix plans.
Related Skills
Track multiple GitHub projects, check releases and changelogs, and generate structured update summaries with impact analysis.
Upload local files to Qiniu Object Storage with server-side auto-delete days and public or private permanent URLs, ideal for temporary files and replacing Base64 storage.
Creates themed learning sites by generating flashcards, lessons, quizzes, index, and progress pages into a deployable website.
End-to-end mathematical optimization modeling with Gurobi and Python, covering data validation, code generation, solving, result analysis, and iterative revision.