メニュー
Use for ANY bug, test failure, crash, or unexpected behavior. Forces reproduce-then-isolate before proposing a fix. Stops the agent from guessing.
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
SOC 職業分類に基づく
Review a diff against the goal spec assuming the code is BROKEN. The reviewer that lives in the maker's head always agrees with itself — this pulls review into a hostile, separate pass. Invoke after every code change before marking work done.
Verify that an endpoint checks ownership, not just authentication. Use on any handler that reads or mutates user data.
Find the exact commit that introduced a bug. Use when something worked before and broke, and you don't know which change did it.
Turn a set of commits or a diff into a clean, user-facing changelog entry. Use before a release or PR description.
Turn messy WIP into clean, atomic commits with messages that explain why. Use before opening a PR.
Keep the agent's context lean so accuracy doesn't collapse. Use on long sessions, big files, or when the agent starts hallucinating.
| name | systematic-debugging |
| description | Use for ANY bug, test failure, crash, or unexpected behavior. Forces reproduce-then-isolate before proposing a fix. Stops the agent from guessing. |
| when_to_use | a test fails, a crash, wrong output, "it worked yesterday", a flaky failure |
The single most expensive agent failure: seeing an error and immediately generating a "fix" based on the error type, without reading what actually happened. Don't.
Read the whole error. The entire message and stack trace. A TypeError can mean a hundred different things — the trace tells you which one. Quote the exact line that throws.
Reproduce it first. If you can't reproduce it, you can't verify a fix. Write the smallest input that triggers it. "I think this fixes it" is gambling, not debugging.
Form one hypothesis, name it. "I think the value is null because the upstream call returns 204 with no body." State it before you touch anything.
Change one thing. Test. Repeat. If you change three things and it works, you don't know which one fixed it — and the other two may have added new bugs.
Fix the root cause, not the symptom. A null check that hides a null is not a fix. Find why it's null. The underlying bug will resurface differently if you only patch the crash.
End with: the root cause (one sentence), the minimal fix, and the test that now proves it.
