Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →This skill should be used only when the user explicitly asks to use `$ralph-specum-implement`, or explicitly asks Ralph Specum in Codex to run implementation for approved tasks, quick mode, or an explicit continue request.
.claude/skills/tzachbon-ralph-specum-implement/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-16 | ✗→✓ | ▲ Improved | 7% | 0% |
| case-11 | ✗→✓ | ▲ Improved | -22% | 0% |
| case-13 | ✗→✓ | ▲ Improved | 28% | 0% |
| case-17 | ✗→✓ | ▲ Improved | 106% | 0% |
| case-15 | ✓→✗ | ▼ Worse | -20% | 0% |
You are a coordinator, not an executor -- delegate each task to a spec-executor sub-agent.
.current-spectasks.md.ralph-state.json only when all tasks are complete and verifiedtasks.md. Read .progress.md, current state, and current task markers.total, completed, and next_index.phase: "execution"awaitingApproval: falsetotalTasks: totaltaskIndex: next_indextaskIteration, maxTaskIterations, globalIteration, maxGlobalIterations, commitSpec, and relatedSpecsspec-executor sub-agent. Pass the task description, file targets, success criteria, and context from .progress.md. The sub-agent implements the task and outputs TASK_COMPLETE. Do NOT implement tasks yourself. Execute tasks in order until complete or blocked.[P] tasks may batch only when file sets do not overlap and verification is independent.[VERIFY] tasks stay in the same run and must produce explicit verification evidence.tasks.md. If markers are absent, treat tasks as non-batchable by default..progress.mdCommit line unless commits were explicitly disabled.ralph-state.json and report completion.tasks.md for completion and repair state to match.| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-16 | fail→pass | 8,779 | 4,217 | -52% | 1 | 1 | 0% | 1,237 | 1,318 | +7% | 0 | 0 | — |
case-01 | fail→fail | 4,088 | 5,905 | +44% | 1 | 1 | 0% | 227 | 937 | +313% | 0 | 0 | — |
case-02 | fail→fail | 3,000 | 5,300 | +77% | 1 | 1 | 0% | 297 | 822 | +177% | 0 | 0 | — |
case-03 | fail→fail | 3,468 | 4,892 | +41% | 1 | 1 | 0% | 145 | 704 | +386% | 0 | 0 | — |
case-04 | fail→fail | 13,413 | 6,195 | -54% | 1 | 1 | 0% | 2,000 | 1,037 | -48% | 0 | 0 | — |
case-05 | fail→fail | 9,026 | 5,448 | -40% | 1 | 1 | 0% | 1,510 | 934 | -38% | 0 | 0 | — |
case-06 | fail→fail | 5,435 | 8,037 | +48% | 1 | 1 | 0% | 783 | 1,428 | +82% | 0 | 0 | — |
case-07 | pass→pass | 7,228 | 2,297 | -68% | 1 | 1 | 0% | 1,051 | 948 | -10% | 0 | 0 | — |
case-08 | pass→pass | 7,791 | 5,290 | -32% | 1 | 1 | 0% | 1,166 | 1,387 | +19% | 0 | 0 | — |
case-09 | fail→fail | 6,919 | 5,907 | -15% | 1 | 1 | 0% | 466 | 864 | +85% | 0 | 0 | — |
case-10 | pass→pass | 8,500 | 2,854 | -66% | 1 | 1 | 0% | 1,258 | 947 | -25% | 0 | 0 | — |
case-15 | pass→fail | 6,159 | 5,772 | -6% | 1 | 1 | 0% | 1,026 | 825 | -20% | 0 | 0 | — |
case-11 | fail→pass | 10,269 | 4,216 | -59% | 1 | 1 | 0% | 1,638 | 1,271 | -22% | 0 | 0 | — |
case-12 | fail→fail | 2,026 | 3,003 | +48% | 1 | 1 | 0% | 229 | 802 | +250% | 0 | 0 | — |
case-13 | fail→pass | 7,486 | 6,052 | -19% | 1 | 1 | 0% | 1,272 | 1,627 | +28% | 0 | 0 | — |
case-14 | pass→fail | 6,872 | 6,558 | -5% | 1 | 1 | 0% | 686 | 929 | +35% | 0 | 0 | — |
case-17 | fail→pass | 4,505 | 5,097 | +13% | 1 | 1 | 0% | 721 | 1,482 | +106% | 0 | 0 | — |
case-18 | pass→pass | 1,725 | 7,860 | +356% | 1 | 1 | 0% | 251 | 1,327 | +429% | 0 | 0 | — |
case-19 | pass→pass | 11,218 | 3,542 | -68% | 1 | 1 | 0% | 1,695 | 1,082 | -36% | 0 | 0 | — |
case-20 | pass→fail | 31,808 | 4,897 | -85% | 1 | 1 | 0% | 5,073 | 835 | -84% | 0 | 0 | — |
case-21 | fail→fail | 23,614 | 8,625 | -63% | 1 | 1 | 0% | 3,048 | 1,283 | -58% | 0 | 0 | — |
case-22 | fail→fail | 2,490 | 15,158 | +509% | 1 | 1 | 0% | 337 | 2,229 | +561% | 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.