Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Merge a base ref and safely regenerate compiled workflow lock-file conflicts.
.claude/skills/github-resolve-merge-conflicts/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-09 | ✗→✓ | ▲ Improved | 10% | 0% |
| case-07 | ✗→✓ | ▲ Improved | -6% | 0% |
| case-11 | ✗→✓ | ▲ Improved | -80% | 0% |
| case-23 | ✗→✓ | ▲ Improved | 7% | 0% |
| case-15 | ✗→✓ | ▲ Improved | -17% | 0% |
Use this skill when merging origin/main into a branch, especially when the only conflicts are generated .github/workflows/*.lock.yml files.
From the repository root, run:
bash./.github/skills/resolve-merge-conflicts/resolve.sh origin/main
The command works both before a merge and after another command has stopped on conflicts. It:
--no-commit, or resumes the current merge..lock.yml..github/workflows/*.md for leftover conflict-marker lines(<<<<<<<, |||||||, =======, >>>>>>>) and aborts before compiling if any are found — see "Why the marker scan matters" below.
make recompile once so generated files come from the merged Markdown.The script does not fetch, commit, push, abort, or edit workflow Markdown. Refresh origin/main first only when credentials are available. After success, review the staged merge, run the repository's final validation gate, then commit and push.
A source .md conflict resolved manually (by a human or an agent) can leave a stray conflict-marker line behind — most often the rarely-noticed ||||||| base (original) diff3 marker — inside a workflow's YAML frontmatter. git diff --check only inspects lines touched by the current diff/staged hunks, so a marker already committed in otherwise-unchanged file content passes silently. The gh-aw compiler then parses the marker text as a literal YAML header option (e.g. invalid header option: "|||||| base (original)"), which fails compilation later — often in an unrelated scheduled recompilation run, far from the original merge, making the root cause hard to trace back.
Run the standalone check any time you suspect a workflow .md file went through manual conflict resolution, even outside this script's merge flow:
bash./.github/skills/resolve-merge-conflicts/resolve.sh --verify-markers
It exits non-zero and lists the offending file(s)/line(s) if any marker is found, and does not modify the working tree.
ours or theirs for compiled lock files; regenerate them..lock.yml files..md, Go, JavaScript, or other mixed conflict.merits, stage them, and rerun the same command. It will regenerate the remaining lock conflicts.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 12,569 | 16,918 | +35% | 1 | 1 | 0% | 1,120 | 863 | -23% | 0 | 0 | — |
case-02 | fail→fail | 11,166 | 24,788 | +122% | 1 | 1 | 0% | 1,001 | 821 | -18% | 0 | 0 | — |
case-03 | fail→fail | 11,943 | 15,919 | +33% | 1 | 1 | 0% | 1,036 | 959 | -7% | 0 | 0 | — |
case-04 | pass→pass | 10,923 | 7,849 | -28% | 1 | 1 | 0% | 919 | 1,115 | +21% | 0 | 0 | — |
case-09 | fail→pass | 13,293 | 8,593 | -35% | 1 | 1 | 0% | 1,205 | 1,320 | +10% | 0 | 0 | — |
case-05 | pass→pass | 9,317 | 7,600 | -18% | 1 | 1 | 0% | 658 | 1,031 | +57% | 0 | 0 | — |
case-06 | pass→fail | 11,908 | 16,109 | +35% | 1 | 1 | 0% | 1,195 | 1,017 | -15% | 0 | 0 | — |
case-07 | fail→pass | 15,740 | 11,476 | -27% | 1 | 1 | 0% | 1,877 | 1,769 | -6% | 0 | 0 | — |
case-08 | pass→pass | 15,408 | 9,041 | -41% | 1 | 1 | 0% | 1,810 | 1,379 | -24% | 0 | 0 | — |
case-10 | pass→pass | 22,333 | 7,875 | -65% | 1 | 1 | 0% | 2,614 | 1,163 | -56% | 0 | 0 | — |
case-11 | fail→pass | 32,184 | 8,209 | -74% | 1 | 1 | 0% | 4,598 | 921 | -80% | 0 | 0 | — |
case-12 | pass→pass | 25,193 | 13,373 | -47% | 1 | 1 | 0% | 3,139 | 1,870 | -40% | 0 | 0 | — |
case-13 | pass→pass | 14,886 | 9,670 | -35% | 1 | 1 | 0% | 1,589 | 1,565 | -2% | 0 | 0 | — |
case-23 | fail→pass | 16,936 | 14,091 | -17% | 1 | 1 | 0% | 1,892 | 2,023 | +7% | 0 | 0 | — |
case-14 | pass→pass | 12,059 | 8,724 | -28% | 1 | 1 | 0% | 1,184 | 1,032 | -13% | 0 | 0 | — |
case-15 | fail→pass | 13,795 | 7,553 | -45% | 1 | 1 | 0% | 1,214 | 1,003 | -17% | 0 | 0 | — |
case-16 | fail→pass | 13,870 | 9,993 | -28% | 1 | 1 | 0% | 1,476 | 1,568 | +6% | 0 | 0 | — |
case-17 | pass→pass | 8,782 | 6,676 | -24% | 1 | 1 | 0% | 567 | 950 | +68% | 0 | 0 | — |
case-18 | pass→pass | 11,401 | 7,469 | -34% | 1 | 1 | 0% | 925 | 1,033 | +12% | 0 | 0 | — |
case-19 | fail→pass | 15,249 | 8,505 | -44% | 1 | 1 | 0% | 1,651 | 1,281 | -22% | 0 | 0 | — |
case-20 | fail→pass | 13,080 | 8,976 | -31% | 1 | 1 | 0% | 1,532 | 1,401 | -9% | 0 | 0 | — |
case-21 | pass→pass | 14,533 | 9,117 | -37% | 1 | 1 | 0% | 1,530 | 1,284 | -16% | 0 | 0 | — |
case-22 | fail→pass | 33,346 | 7,270 | -78% | 1 | 1 | 0% | 1,252 | 1,030 | -18% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 23 cases were attempted, and 19 counted toward the lift figure. The other 4 produced results that are not comparable between the two arms, so they are excluded from the headline rather than averaged into it. The headline lift of +35 percentage points is the difference between those two pass rates over the 19 comparable cases. 1 case got worse with the skill loaded, and it is included in that figure.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
| Model | Method | Date | Lift |
|---|---|---|---|
| gemini-3.6-flash | verified | 8/9/2026 | +41% |
Other measured skills in the registry, with their headline benchmark lift.