Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Work on a single highest-priority feature from current_ documents, test it, update docs, and commit
.claude/skills/sterlingcrispin-ralph-once-single-feature-implementation-cycle/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-05 | ✗→✓ | ▲ Improved | 280% | 0% |
| case-08 | ✗→✓ | ▲ Improved | -31% | 0% |
| case-10 | ✗→✓ | ▲ Improved | 85% | 0% |
| case-15 | ✗→✓ | ▲ Improved | -22% | 0% |
| case-16 | ✗→✓ | ▲ Improved | -21% | 0% |
You are about to execute a focused, single-feature development cycle. Follow these steps precisely:
Find and read ALL files in the current project that start with current_ (e.g., current_PROGRESS.md, current_PRD.json, or for example current_SYSTEM-PLAN.md, or any other current_ files could be .txt , .json , .md or other). These contain the project roadmap, PRD, and progress tracking.
After reading all current_* documents, YOU decide which single feature has the highest priority. Consider:
Do NOT just pick the first item in a list. Use your judgment about what will provide the most value or unblock the most work.
Announce your choice and briefly explain WHY you chose it before proceeding.
Work on ONLY this single feature. Do not scope creep. Do not "also fix" other things. Stay focused.
Find the current_* document that serves as the PRD/plan (likely current_PRD.md or similar). Update it to reflect:
"passes": false to "passes": true] if using that or change "done" to true)Find current_PROGRESS.md (or the equivalent progress journal tracking file). Append a new entry to the Session Log with:
Stage and commit all changes with a clear commit message describing the feature implemented.
Do NOT push unless explicitly asked.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 3,272 | 2,919 | -11% | 1 | 1 | 0% | 222 | 749 | +237% | 0 | 0 | — |
case-02 | fail→fail | 4,234 | 2,719 | -36% | 1 | 1 | 0% | 131 | 805 | +515% | 0 | 0 | — |
case-03 | fail→fail | 3,904 | 2,992 | -23% | 1 | 1 | 0% | 174 | 828 | +376% | 0 | 0 | — |
case-04 | pass→fail | 8,383 | 3,787 | -55% | 1 | 1 | 0% | 1,355 | 854 | -37% | 0 | 0 | — |
case-05 | fail→pass | 4,818 | 5,678 | +18% | 1 | 1 | 0% | 265 | 1,007 | +280% | 0 | 0 | — |
case-06 | pass→fail | 7,162 | 4,572 | -36% | 1 | 1 | 0% | 1,169 | 892 | -24% | 0 | 0 | — |
case-07 | pass→pass | 3,910 | 4,830 | +24% | 1 | 1 | 0% | 641 | 1,492 | +133% | 0 | 0 | — |
case-08 | fail→pass | 11,065 | 3,876 | -65% | 1 | 1 | 0% | 1,716 | 1,192 | -31% | 0 | 0 | — |
case-09 | fail→fail | 6,585 | 4,187 | -36% | 1 | 1 | 0% | 1,158 | 1,317 | +14% | 0 | 0 | — |
case-10 | fail→pass | 4,660 | 4,591 | -1% | 1 | 1 | 0% | 778 | 1,437 | +85% | 0 | 0 | — |
case-11 | pass→pass | 10,335 | 2,696 | -74% | 1 | 1 | 0% | 1,512 | 1,015 | -33% | 0 | 0 | — |
case-12 | pass→pass | 11,271 | 3,801 | -66% | 1 | 1 | 0% | 1,772 | 1,201 | -32% | 0 | 0 | — |
case-13 | pass→pass | 7,371 | 4,765 | -35% | 1 | 1 | 0% | 1,128 | 1,373 | +22% | 0 | 0 | — |
case-14 | pass→pass | 10,474 | 2,331 | -78% | 1 | 1 | 0% | 1,578 | 937 | -41% | 0 | 0 | — |
case-15 | fail→pass | 11,583 | 4,535 | -61% | 1 | 1 | 0% | 1,755 | 1,372 | -22% | 0 | 0 | — |
case-16 | fail→pass | 11,300 | 4,950 | -56% | 1 | 1 | 0% | 1,591 | 1,259 | -21% | 0 | 0 | — |
case-17 | fail→fail | 7,953 | 4,584 | -42% | 1 | 1 | 0% | 1,169 | 924 | -21% | 0 | 0 | — |
case-18 | pass→pass | 5,223 | 6,893 | +32% | 1 | 1 | 0% | 893 | 1,225 | +37% | 0 | 0 | — |
case-19 | pass→fail | 5,781 | 2,517 | -56% | 1 | 1 | 0% | 834 | 954 | +14% | 0 | 0 | — |
case-20 | pass→fail | 29,074 | 2,410 | -92% | 1 | 1 | 0% | 5,992 | 816 | -86% | 0 | 0 | — |
case-21 | pass→fail | 13,403 | 4,275 | -68% | 1 | 1 | 0% | 2,480 | 845 | -66% | 0 | 0 | — |
case-22 | pass→fail | 13,165 | 4,890 | -63% | 1 | 1 | 0% | 2,711 | 772 | -72% | 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 13 counted toward the lift figure. The other 9 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 -5 percentage points is the difference between those two pass rates over the 13 comparable cases. 6 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.