Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Run Ralph autonomous build loop. Use when user asks to "ralph build", "run build loop", or needs to process subtasks autonomously. Executes iterations against a subtasks.json queue.
.claude/skills/majiayu000-ralph-build/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-07 | ✗→✓ | ▲ Improved | 49% | 0% |
| case-08 | ✗→✓ | ▲ Improved | -6% | 0% |
| case-09 | ✗→✓ | ▲ Improved | -72% | 0% |
| case-10 | ✗→✓ | ▲ Improved | -26% | 0% |
| case-13 | ✗→✓ | ▲ Improved | -84% | 0% |
Execute the Ralph autonomous build loop to process subtasks from a queue.
/ralph-build [options]| Option | Description | |--------|-------------| | --subtasks <path> | Path to subtasks.json file (will prompt if not provided) | | -i, --interactive | Pause between iterations for user review | | -p, --print | Output the prompt without executing (dry run) | | --validate-first | Run pre-build validation before starting the loop | | --max-iterations <n> | Maximum retry attempts per subtask (default: 3) |
If --subtasks is not provided, prompt the user:
> "Which subtasks.json file should I use? Provide the path or I'll look for docs/planning/milestones/*/subtasks.json"
If print mode is requested:
If validate-first is requested:
For each iteration, follow the ralph-iteration workflow:
@context/workflows/ralph/building/ralph-iteration.md
After each completed iteration:
If a subtask fails repeatedly:
max-iterations failures on the same subtaskThis skill provides the same functionality as:
bashaaa ralph build [options]
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 17,229 | 13,424 | -22% | 1 | 1 | 0% | 344 | 1,812 | +427% | 0 | 0 | — |
case-02 | fail→fail | 16,191 | 29,430 | +82% | 1 | 1 | 0% | 245 | 835 | +241% | 0 | 0 | — |
case-03 | fail→fail | 7,122 | 7,864 | +10% | 1 | 1 | 0% | 325 | 863 | +166% | 0 | 0 | — |
case-04 | fail→fail | 14,758 | 9,664 | -35% | 1 | 1 | 0% | 773 | 756 | -2% | 0 | 0 | — |
case-05 | fail→fail | 2,453 | 14,258 | +481% | 1 | 1 | 0% | 300 | 708 | +136% | 0 | 0 | — |
case-06 | pass→fail | 11,530 | 10,389 | -10% | 1 | 1 | 0% | 1,285 | 1,476 | +15% | 0 | 0 | — |
case-07 | fail→pass | 9,968 | 4,469 | -55% | 1 | 1 | 0% | 656 | 978 | +49% | 0 | 0 | — |
case-08 | fail→pass | 6,724 | 8,174 | +22% | 1 | 1 | 0% | 1,154 | 1,087 | -6% | 0 | 0 | — |
case-09 | fail→pass | 19,923 | 6,915 | -65% | 1 | 1 | 0% | 3,018 | 833 | -72% | 0 | 0 | — |
case-10 | fail→pass | 12,801 | 7,420 | -42% | 1 | 1 | 0% | 1,265 | 942 | -26% | 0 | 0 | — |
case-11 | fail→fail | 14,617 | 2,882 | -80% | 1 | 1 | 0% | 2,335 | 969 | -59% | 0 | 0 | — |
case-12 | fail→fail | 35,072 | 6,991 | -80% | 1 | 1 | 0% | 1,018 | 834 | -18% | 0 | 0 | — |
case-13 | fail→pass | 54,440 | 9,043 | -83% | 1 | 1 | 0% | 5,448 | 849 | -84% | 0 | 0 | — |
case-14 | fail→pass | 10,980 | 7,586 | -31% | 1 | 1 | 0% | 985 | 993 | +1% | 0 | 0 | — |
case-15 | pass→pass | 13,177 | 2,307 | -82% | 1 | 1 | 0% | 1,096 | 822 | -25% | 0 | 0 | — |
case-16 | fail→pass | 63,054 | 8,483 | -87% | 1 | 1 | 0% | 4,292 | 1,093 | -75% | 0 | 0 | — |
case-17 | fail→pass | 18,857 | 1,947 | -90% | 1 | 1 | 0% | 2,447 | 801 | -67% | 0 | 0 | — |
case-18 | fail→pass | 9,680 | 7,953 | -18% | 1 | 1 | 0% | 1,440 | 1,011 | -30% | 0 | 0 | — |
case-19 | pass→pass | 14,135 | 4,900 | -65% | 1 | 1 | 0% | 1,190 | 987 | -17% | 0 | 0 | — |
case-20 | fail→pass | 15,541 | 8,638 | -44% | 1 | 1 | 0% | 1,814 | 1,116 | -38% | 0 | 0 | — |
case-21 | fail→pass | 15,146 | 8,059 | -47% | 1 | 1 | 0% | 1,715 | 1,006 | -41% | 0 | 0 | — |
case-22 | fail→pass | 16,623 | 2,233 | -87% | 1 | 1 | 0% | 1,895 | 747 | -61% | 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. 22 cases were attempted, and 16 counted toward the lift figure. The other 6 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 +50 percentage points is the difference between those two pass rates over the 16 comparable cases. 3 cases got worse with the skill loaded, and they are 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.
Other measured skills in the registry, with their headline benchmark lift.