Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Archives or quarantines current Clean Room artifacts and restarts from the scope gate with a fresh task id.
.claude/skills/hashgraph-online-start-over/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-08 | ✗→✓ | ▲ Improved | 86% | 0% |
| case-09 | ✗→✓ | ▲ Improved | 716% | 0% |
| case-10 | ✗→✓ | ▲ Improved | 107% | 0% |
| case-11 | ✗→✓ | ▲ Improved | 510% | 0% |
| case-12 | ✗→✓ | ▲ Improved | 171% | 0% |
Start a clean-room run over without overwriting or deleting existing artifacts in place.
Use the canonical clean-room skill workflow and references in this plugin. Preserve the same clean-room boundary, role separation, artifact schemas, leakage rules, implementation-root rules, and hook expectations.
Require explicit user confirmation before changing artifact locations. Stop if confirmation is absent, authorization is unclear, or the archive target overlaps a source root in a way that would make provenance unclear.
Archive or quarantine previous artifacts before creating new ones:
task-manifest.json artifact_paths.quarantine path.clean-room-project.json, or the shared project-level implementation/ root; sibling tasks depend on them, and implementation content is destination state, not per-task artifact state.preflight-goal.json, task-manifest.json, ledgers, handoff packages, behavior specs, skeleton manifests, implementation plans, implementation reports, QC reports, incident records, and open delta tickets.If safe archive targets cannot be proven from task-manifest.json, root environment, or explicit user input, stop before moving anything.
Start from the preflight gate, not from prior QC:
preflight-goal.json or explicitly reuse a reviewed goal contract before recreating active artifacts.task_id by default. Use task- plus 8 lowercase hex characters unless the user provides an explicitly approved neutral ID. Do not derive the new ID or output directory names from source folder names.proj- plus 8 lowercase hex, never source-derived).run_state.generation, run_state.started_at, optional run_state.previous_generation_ref, and run_state.restart_reason.clean-run-context.json from the new effective preflight and initialization choices; do not carry forward an old clean context by default.source-index.json, or visual-index.json for visual fallback runs, unless the user explicitly says the source or visual scope is unchanged and a recorded old index hash can still be validated.Do not carry forward prior chat history or unapproved clean artifacts. Treat archived artifacts as audit history, not active task state.
Return a concise restart summary:
task_id or generation identifier.preflight-goal.json is active.source-index.json or visual-index.json must be rebuilt.Do not carry forward unapproved implementation code as active task state.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | pass→fail | 13,115 | 21,546 | +64% | 1 | 1 | 0% | 1,420 | 1,622 | +14% | 0 | 0 | — |
case-02 | fail→fail | 13,696 | 17,593 | +28% | 1 | 1 | 0% | 1,758 | 1,306 | -26% | 0 | 0 | — |
case-03 | fail→fail | 12,613 | 14,679 | +16% | 1 | 1 | 0% | 1,302 | 1,345 | +3% | 0 | 0 | — |
case-04 | fail→fail | 6,201 | 22,722 | +266% | 1 | 1 | 0% | 973 | 3,698 | +280% | 0 | 0 | — |
case-05 | pass→pass | 17,259 | 26,920 | +56% | 1 | 1 | 0% | 2,289 | 4,855 | +112% | 0 | 0 | — |
case-06 | pass→fail | 10,607 | 20,005 | +89% | 1 | 1 | 0% | 1,524 | 3,162 | +107% | 0 | 0 | — |
case-07 | pass→pass | 28,608 | 27,648 | -3% | 1 | 1 | 0% | 1,727 | 2,953 | +71% | 0 | 0 | — |
case-08 | fail→pass | 20,969 | 9,944 | -53% | 1 | 1 | 0% | 1,382 | 2,564 | +86% | 0 | 0 | — |
case-09 | fail→pass | 2,952 | 14,169 | +380% | 1 | 1 | 0% | 280 | 2,285 | +716% | 0 | 0 | — |
case-10 | fail→pass | 16,249 | 16,521 | +2% | 1 | 1 | 0% | 1,318 | 2,732 | +107% | 0 | 0 | — |
case-11 | fail→pass | 8,099 | 15,806 | +95% | 1 | 1 | 0% | 423 | 2,580 | +510% | 0 | 0 | — |
case-12 | fail→pass | 11,566 | 10,079 | -13% | 1 | 1 | 0% | 1,066 | 2,886 | +171% | 0 | 0 | — |
case-13 | fail→pass | 9,302 | 14,468 | +56% | 1 | 1 | 0% | 728 | 2,308 | +217% | 0 | 0 | — |
case-14 | fail→pass | 10,062 | 15,125 | +50% | 1 | 1 | 0% | 428 | 2,618 | +512% | 0 | 0 | — |
case-15 | fail→fail | 6,493 | 14,529 | +124% | 1 | 1 | 0% | 1,107 | 2,300 | +108% | 0 | 0 | — |
case-16 | fail→pass | 15,169 | 5,615 | -63% | 1 | 1 | 0% | 1,821 | 1,734 | -5% | 0 | 0 | — |
case-17 | fail→pass | 13,030 | 12,774 | -2% | 1 | 1 | 0% | 1,185 | 2,137 | +80% | 0 | 0 | — |
case-18 | pass→pass | 18,705 | 24,318 | +30% | 1 | 1 | 0% | 1,591 | 1,775 | +12% | 0 | 0 | — |
case-19 | fail→pass | 16,040 | 15,323 | -4% | 1 | 1 | 0% | 1,161 | 2,141 | +84% | 0 | 0 | — |
case-20 | fail→pass | 395,205 | 258,686 | -35% | 1 | 1 | 0% | 655 | 2,076 | +217% | 0 | 0 | — |
case-21 | fail→pass | 23,745 | 75,023 | +216% | 1 | 1 | 0% | 1,499 | 1,558 | +4% | 0 | 0 | — |
case-22 | fail→pass | 298,485 | 82,012 | -73% | 1 | 1 | 0% | 1,360 | 1,446 | +6% | 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 19 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 +50 percentage points is the difference between those two pass rates over the 19 comparable cases. 4 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.