Code
Find a regression
Bisect between a good and a bad commit until it names one.
Fill it in
Any git remote.
A ref where it worked.
A ref where it does not.
The command that fails on a bad commit.
Where to run the bisect.
Only for a private repository.
Your prompt
Find where this broke in [your repository].
Good: [your last good commit]. Bad: [your first bad commit].
Check with: `[your how to tell]`
Clone onto using `${credential:${credential:ACCESS_TOKEN}}` and run `git bisect` driven by that command.
Verify the endpoints FIRST: the check must pass at [your last good commit] and fail at
[your first bad commit]. If it does not, the bisect will confidently name an innocent commit,
and that is worse than no answer.
When it lands, show the commit, its message, and the part of its diff that
explains the failure. A hash on its own is not an answer.Continue and add your fileOpens with everything above already filled in.
Why this works
Bisect needs a command that answers yes or no, not a description of the symptom. Asking for it as a field is what turns a vague report into a search that terminates.