What does this skill do?

The Defect Fix Orchestrator Skill resolves software bugs by ensuring that each defect is reproduced as a regression test before applying the fix. The workflow highlights the defect in red, applies the fix until it turns green, reviews the code, and confirms the commit with the prefix "fix:". It includes human checkpoints and an optional security reviewer for sensitive paths.

Reproducing bugs
Write a regression test that fails first, reproducing the exact defect before touching the production code.
Fix marked as "green"
Apply the fix until the test turns green, ensuring that the defect is actually resolved and that it isn't just a blind fix.
Control Doors
Pause execution at Gate 1 (generated plan) and Gate 2 (before commit) for human validation before proceeding.
Safety Review
It adds an automatic security checker when the defect is located in sensitive system paths.

Usage examples

🐛 API hang
The poller hangs with an empty API response; reproduce the bug with a test and fix it.
🧮 Incorrect calculation
The discount is applied twice to the total cart amount; write a regression test and fix it.
🔁 Regression detected
There's a regression after the last deployment; orchestrate the fix by running the test that fails and committing with a "fix" prefix.
📄 Malfunction
This documented behavior no longer works; please reproduce the issue as a test and fix it.

Features

Test in red first Require that the defect be reproduced as a failed regression test before applying any fixes to the code.
Phased Orchestration Run the engine with a phase mask of 0 → (2) → 4 → 5 → 6 and delegate each phase to specialized agents.
Validation Gates Gate 1 validates the generated plan, and Gate 2 confirms the commit, ensuring key milestones through human intervention.
Safety Inspector Enable an additional security check when the defect affects sensitive system paths.
Traceable commits Generate commits prefixed with "fix:" and a documented code review report in the repository.

Frequently asked questions

Phase 1 of the investigation is usually skipped by default. If the root cause is not obvious, the scope is narrowed down using a code explorer before writing the red test.
These are control gates that pause execution. Gate 1 validates the generated plan, and Gate 2 confirms the commit before applying it to the repository.
It does not require a specific MCP. It works with Claude's orchestration engine and the project's testing tools (the repository's testing framework).
Yes, as long as the project has a testing system. It's ideal for regression testing, calculation errors, crashes, and documented behaviors that no longer work.
Bug Fixing Orchestrator — Reproducible bugs as tests with Claude AI

¿Prefieres escuchar el contenido? Genera la narración de audio con un clic.