菜单
Build a signal that goes red on the bug before you theorize or fix it. Use when debugging something broken, slow, or failing, when confirming a bug report is real, or before concluding a bug cannot be reproduced. Triggers include "diagnose", "why is this happening", "can you reproduce this".
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
| name | repro-an-issue |
| description | Build a signal that goes red on the bug before you theorize or fix it. Use when debugging something broken, slow, or failing, when confirming a bug report is real, or before concluding a bug cannot be reproduced. Triggers include "diagnose", "why is this happening", "can you reproduce this". |
No hypothesis and no fix until you have one command that goes red on this exact bug and green once it is fixed. Build the signal first and the cause falls out of it. Reproducing is cheap next to fixing, so spend the cheap effort ruthlessly.
Name one command (a test, a curl, a CLI diff, a script) that you have already run once (paste it and its output) and that:
If you are reading code to build a theory before this command exists, stop. Jumping to a hypothesis without a signal is the exact failure this skill prevents.
Done when: you can paste one command and its output, and it can go red on this bug.
A green on the wrong path is not a green. If your repro runs against a different config, port, build, or entry point than the user's, a pass proves nothing: you tested a scenario the bug does not live in. Before you conclude "cannot reproduce", match the reported setup on every axis that could matter. A config artifact that silently dodges the bug (a non-default port, a stub, the wrong build) is how you wrongly clear a real bug.
Done when: the signal goes red on the user's failure, not a lookalike nearby.
Cut inputs, callers, config, and steps one at a time, re-running after each cut, until every remaining element is load-bearing (removing any one turns it green). The minimal repro shrinks the hypothesis space and becomes the regression test.
Done when: nothing left can be removed without turning the signal green.
List three to five ranked, falsifiable hypotheses before testing any ("if X is the cause, changing Y makes it disappear"). Test one variable at a time. Tag every temporary log with a unique prefix so cleanup is one grep.
Turn the minimised repro into a regression test at the correct seam, then hand it to prove-the-test: revert the fix, watch it go red, restore. Re-run the signal against the original scenario. Remove all temporary instrumentation.
Done when: the original repro no longer reproduces, and a committed test goes red without the fix.
Learn an unfamiliar codebase by shipping a real change, where you reconstruct and predict while the agent tutors instead of doing it for you. Use when onboarding to an unfamiliar repo, picking up a first issue, or the user asks to be walked through a codebase rather than handed a fix.
Choose what's worth fixing in someone else's issue backlog, verify it's real before investing, and ship it as a clean PR with credit. Use when landing in an unfamiliar or shared repo with a large open-issue list and needing to decide what to pick up, for example "what should I work on here" or "find a good issue to fix". Contributor-side selection (what to take FROM a repo), not maintainer-side tracker management.
Prove a test fails when the code is wrong before you trust it green. Use when writing or reviewing a test for a change or bug fix, when a change ships with passing tests you are not sure cover it, or when the user asks "does this test have teeth", "is this covered", or "add a test for this fix".
