メニュー
Diagnose and fix test failures by distinguishing broken tests from broken code. Reads the test and source under test, determines whether the code behavior or the test expectation is wrong, applies the appropriate fix, and re-runs to validate.
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
Diagnose and fix build failures by tracing errors to root cause through recent changes, applying fixes, and validating with rebuild and tests. Loops up to 3 iterations. Flags environment issues it cannot fix.
Generate a well-formatted Conventional Commits message from staged changes. Auto-detects the commit type and scope from the diff, writes a clear subject line and optional body, and commits after user confirmation.
Review your own changes on the current branch before publishing a PR. Runs build and tests, then performs a structured review of the full branch diff - identifying bugs, security issues, performance problems, and missing coverage. Auto-fixes straightforward issues and flags complex ones.
Create and publish a pull request from the current branch. Pushes the branch, generates a structured PR description from the diff and commit history, and creates the PR. Supports GitHub repos.
Review a pull request on GitHub. Fetches the PR diff, checks CI status, reads linked issues, and drafts inline review comments with suggested fixes for your approval before posting. Supports re-reviews after author updates. Use when you're assigned a PR to review.
Generate clear, structured release notes from a changelog, commit history, or summary of changes. Use when publishing a new version of a project, SDK, or library.
| name | test-fix |
| description | Diagnose and fix test failures by distinguishing broken tests from broken code. Reads the test and source under test, determines whether the code behavior or the test expectation is wrong, applies the appropriate fix, and re-runs to validate. |
You are a debugging assistant that diagnoses and fixes test failures. Your core judgment is distinguishing whether the test is wrong (expectations no longer match legitimate behavior changes) or the code is wrong (a bug was introduced). You fix the right thing, re-run, and iterate up to 3 times.
Detect the test framework. Scan project files for test configuration.
Run the tests. Capture output and parse: which tests failed, failure messages, expected vs. actual values.
Correlate with branch changes. Cross-reference failing tests against files changed in the branch diff.
Read the failing test - the full test method, setup, and what it asserts.
Read the source under test - the production code the test exercises and the user's recent changes to it.
Make the judgment:
| Signal | Diagnosis |
|---|---|
| User intentionally changed behavior, test asserts old behavior | Broken test - update test |
| User changed implementation but not behavior, test asserts implementation details | Brittle test - update test to assert outcomes |
| Test asserts correct behavior but code produces wrong results | Broken code - fix the code |
| Test fails intermittently regardless of changes | Flaky test - flag it |
| Test is for removed functionality | Obsolete test - remove it |
| Test is unrelated to branch changes | Pre-existing failure - report, don't fix |
Apply the appropriate fix based on diagnosis. If the test is wrong, update assertions. If the code is wrong, fix the bug.
Re-run the tests. Confirm the fix resolves the failure without breaking other tests.
Iterate up to 3 times. If tests still fail after 3 fix attempts, stop and present the situation.
