Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Runs the Ralph autonomous loop. Executes stories from prds/*.json using git worktrees.
.claude/skills/majiayu000-ralph-json-start-loop/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-04 | ✗→✓ | ▲ Improved | -32% | 0% |
| case-07 | ✗→✓ | ▲ Improved | -54% | 0% |
| case-08 | ✗→✓ | ▲ Improved | -77% | 0% |
| case-11 | ✗→✓ | ▲ Improved | -42% | 0% |
| case-12 | ✗→✓ | ▲ Improved | -54% | 0% |
Run the autonomous loop to execute features from prds/ directory.
/ralph # Run next available project (respects dependencies)
/ralph 25 # Run with 25 iterations
/ralph auth-flow # Run specific projectRun the loop script in background mode:
bash~/.claude/skills/ralph/ralph.sh [iterations] [project-name]
Use run_in_background: true to prevent timeout. After starting, tell the user to check progress with tail -f <worktree>/.ralph-progress.txt.
../{repo}-{feature}/passes: falsefeat: [id] - [title]<promise>COMPLETE</promise>Ralph reads dependsOn from each PRD and enforces ordering:
json{ "projectName": "Dashboard", "dependsOn": ["auth-flow", "user-profile"] }
Projects with incomplete dependencies are blocked. Ralph picks the first ready project alphabetically.
.json PRD file in .claude/plans/, plans/, or prds//ralph-json-create-issues to convert it| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 14,237 | 7,637 | -46% | 1 | 1 | 0% | 967 | 826 | -15% | 0 | 0 | — |
case-02 | fail→fail | 10,992 | 12,457 | +13% | 1 | 1 | 0% | 1,116 | 971 | -13% | 0 | 0 | — |
case-03 | fail→fail | 7,861 | 4,845 | -38% | 1 | 1 | 0% | 1,283 | 579 | -55% | 0 | 0 | — |
case-04 | fail→pass | 15,312 | 9,583 | -37% | 1 | 1 | 0% | 1,158 | 785 | -32% | 0 | 0 | — |
case-05 | pass→pass | 9,360 | 2,289 | -76% | 1 | 1 | 0% | 1,611 | 742 | -54% | 0 | 0 | — |
case-06 | pass→pass | 5,104 | 2,284 | -55% | 1 | 1 | 0% | 791 | 751 | -5% | 0 | 0 | — |
case-07 | fail→pass | 15,176 | 11,376 | -25% | 1 | 1 | 0% | 1,739 | 793 | -54% | 0 | 0 | — |
case-08 | fail→pass | 16,499 | 6,622 | -60% | 1 | 1 | 0% | 2,568 | 593 | -77% | 0 | 0 | — |
case-09 | pass→pass | 9,112 | 2,817 | -69% | 1 | 1 | 0% | 686 | 867 | +26% | 0 | 0 | — |
case-10 | pass→pass | 14,805 | 9,281 | -37% | 1 | 1 | 0% | 1,366 | 1,123 | -18% | 0 | 0 | — |
case-11 | fail→pass | 29,652 | 10,234 | -65% | 1 | 1 | 0% | 2,199 | 1,285 | -42% | 0 | 0 | — |
case-12 | fail→pass | 18,357 | 7,096 | -61% | 1 | 1 | 0% | 1,528 | 698 | -54% | 0 | 0 | — |
case-13 | fail→pass | 17,086 | 6,542 | -62% | 1 | 1 | 0% | 1,499 | 595 | -60% | 0 | 0 | — |
case-14 | fail→pass | 18,084 | 7,045 | -61% | 1 | 1 | 0% | 1,904 | 598 | -69% | 0 | 0 | — |
case-15 | pass→pass | 19,770 | 6,782 | -66% | 1 | 1 | 0% | 2,303 | 634 | -72% | 0 | 0 | — |
case-16 | fail→pass | 17,417 | 1,741 | -90% | 1 | 1 | 0% | 2,988 | 653 | -78% | 0 | 0 | — |
case-17 | pass→pass | 3,410 | 8,982 | +163% | 1 | 1 | 0% | 516 | 1,046 | +103% | 0 | 0 | — |
case-18 | pass→pass | 13,358 | 2,993 | -78% | 1 | 1 | 0% | 1,166 | 720 | -38% | 0 | 0 | — |
case-19 | fail→pass | 24,717 | 8,007 | -68% | 1 | 1 | 0% | 821 | 833 | +1% | 0 | 0 | — |
case-20 | fail→pass | 14,323 | 9,665 | -33% | 1 | 1 | 0% | 2,285 | 1,256 | -45% | 0 | 0 | — |
case-21 | pass→pass | 12,899 | 4,895 | -62% | 1 | 1 | 0% | 1,236 | 914 | -26% | 0 | 0 | — |
case-22 | pass→pass | 18,297 | 4,023 | -78% | 1 | 1 | 0% | 1,384 | 1,080 | -22% | 0 | 0 | — |
case-23 | fail→pass | 5,861 | 6,986 | +19% | 1 | 1 | 0% | 896 | 718 | -20% | 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 20 counted toward the lift figure. The other 3 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 +48 percentage points is the difference between those two pass rates over the 20 comparable cases.
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.