| name | build-test-branch |
| description | Test every commit in the current branch, fix failures, create fixup commits, and repeat until all commits pass. |
Build Test Branch
Test each commit in the current branch against the upstream branch. Keep the main thread focused on the loop; use a subagent for failure fixing only when the user explicitly asks for subagents.
Setup
Create a temp work directory:
WORKDIR=$(mktemp -d)
Create ${WORKDIR}/report.md and record the start timestamp. Use iteration = 0 and max_iterations = 10.
Run Test Branch
Increment the iteration and run:
just --global-justfile test-branch > "${WORKDIR}/build.log" 2>&1; echo $?
Use a 30-minute timeout.
- Exit code 0: all commits pass.
- Non-zero: inspect the failure and continue.
Fix Failure
Check for auto-formatted changes:
git status --porcelain
If files changed, continue to test-fix. If there are no changes, identify the failing commit with:
git log --oneline -1
If the same commit failed in the previous iteration, stop and show the report.
Fix the error near the end of ${WORKDIR}/build.log using the build-fix skill. Do not create commits manually.
Run Test Fix
Run:
just --global-justfile test-fix > "${WORKDIR}/test-fix.log" 2>&1; echo $?
Use a 30-minute timeout.
- Exit code 0: all commits pass.
- Non-zero: loop back to test-branch.
Never run git test forget-results. Never rebase manually; test-fix owns the rebase/autosquash step.
Report
Show the failing commit, error summary, fix summary, changed files, and final result.
