Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Start a Ralph Loop for iterative self-referential development. Use when the user asks to run a ralph loop, start an iterative loop, or wants repeated autonomous iteration on a task until completion.
.claude/skills/sediman-agent-ralph-loop/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-08 | ✗→✓ | ▲ Improved | -34% | 0% |
| case-10 | ✗→✓ | ▲ Improved | -61% | 0% |
| case-11 | ✗→✓ | ▲ Improved | -45% | 0% |
| case-12 | ✗→✓ | ▲ Improved | -27% | 0% |
| case-17 | ✗→✓ | ▲ Improved | -78% | 0% |
The user wants to start a Ralph loop. An iterative development loop where the same prompt is fed back after every turn, and the agent sees its own previous work each iteration.
max_iterations (number, default 0 for unlimited)completion_promise (text, or "null" if not set).cursor/ralph/ if it doesn't exist, then write the state file at .cursor/ralph/scratchpad.md with this exact format:markdown --- iteration: 1 max_iterations: <N or 0> completion_promise: "<TEXT>" or null ---
<the user's task prompt goes here>
Example: markdown --- iteration: 1 max_iterations: 20 completion_promise: "COMPLETE" ---
Build a REST API for todos with CRUD operations, input validation, and tests.
[Ralph loop iteration N.].<promise>TEXT</promise> when the statement is completely and genuinely true.max_iterations as a safety net.completion_promise value in the YAML frontmatter if it contains special characters.Confirm the loop is active (prompt, iteration limit, promise if set), then start working on the task immediately.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 11,234 | 5,989 | -47% | 1 | 1 | 0% | 2,354 | 860 | -63% | 0 | 0 | — |
case-02 | fail→fail | 6,096 | 5,232 | -14% | 1 | 1 | 0% | 926 | 766 | -17% | 0 | 0 | — |
case-03 | fail→fail | 8,863 | 6,515 | -26% | 1 | 1 | 0% | 1,436 | 809 | -44% | 0 | 0 | — |
case-04 | fail→fail | 11,831 | 8,865 | -25% | 1 | 1 | 0% | 1,823 | 843 | -54% | 0 | 0 | — |
case-05 | fail→fail | 13,075 | 7,139 | -45% | 1 | 1 | 0% | 2,233 | 785 | -65% | 0 | 0 | — |
case-06 | fail→fail | 5,083 | 5,706 | +12% | 1 | 1 | 0% | 742 | 741 | -0% | 0 | 0 | — |
case-07 | pass→pass | 4,098 | 2,137 | -48% | 1 | 1 | 0% | 607 | 761 | +25% | 0 | 0 | — |
case-08 | fail→pass | 8,827 | 2,651 | -70% | 1 | 1 | 0% | 1,273 | 836 | -34% | 0 | 0 | — |
case-09 | pass→pass | 9,371 | 3,290 | -65% | 1 | 1 | 0% | 1,393 | 928 | -33% | 0 | 0 | — |
case-10 | fail→pass | 22,605 | 1,551 | -93% | 1 | 1 | 0% | 1,612 | 624 | -61% | 0 | 0 | — |
case-11 | fail→pass | 19,704 | 1,426 | -93% | 1 | 1 | 0% | 1,093 | 606 | -45% | 0 | 0 | — |
case-12 | fail→pass | 7,458 | 3,888 | -48% | 1 | 1 | 0% | 1,218 | 895 | -27% | 0 | 0 | — |
case-13 | pass→pass | 5,308 | 2,990 | -44% | 1 | 1 | 0% | 938 | 911 | -3% | 0 | 0 | — |
case-14 | fail→fail | 7,873 | 11,532 | +46% | 1 | 1 | 0% | 1,262 | 773 | -39% | 0 | 0 | — |
case-15 | pass→pass | 3,462 | 4,499 | +30% | 1 | 1 | 0% | 529 | 1,243 | +135% | 0 | 0 | — |
case-16 | pass→pass | 10,520 | 2,716 | -74% | 1 | 1 | 0% | 1,492 | 799 | -46% | 0 | 0 | — |
case-17 | fail→pass | 17,314 | 1,495 | -91% | 1 | 1 | 0% | 2,595 | 573 | -78% | 0 | 0 | — |
case-18 | pass→pass | 6,302 | 9,153 | +45% | 1 | 1 | 0% | 900 | 1,273 | +41% | 0 | 0 | — |
case-19 | fail→pass | 18,821 | 1,405 | -93% | 1 | 1 | 0% | 1,206 | 580 | -52% | 0 | 0 | — |
case-20 | pass→pass | 13,980 | 15,695 | +12% | 1 | 1 | 0% | 2,584 | 2,389 | -8% | 0 | 0 | — |
case-21 | pass→pass | 13,090 | 6,368 | -51% | 1 | 1 | 0% | 2,144 | 1,399 | -35% | 0 | 0 | — |
case-22 | pass→pass | 8,277 | 12,131 | +47% | 1 | 1 | 0% | 1,488 | 2,484 | +67% | 0 | 0 | — |
case-23 | pass→fail | 5,343 | 4,731 | -11% | 1 | 1 | 0% | 965 | 1,002 | +4% | 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 18 counted toward the lift figure. The other 5 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 +22 percentage points is the difference between those two pass rates over the 18 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.
Other measured skills in the registry, with their headline benchmark lift.