Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Execute an approved GRACE 4 GraceChangePlan in sequential or parallel-safe mode with recovery-aware preflight and centralized durable apply.
.claude/skills/osovv-grace-execute/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-07 | ✗→✓ | ▲ Improved | 65% | 0% |
| case-01 | ✗→✓ | ▲ Improved | 400% | 0% |
| case-08 | ✗→✓ | ▲ Improved | 337% | 0% |
| case-11 | ✗→✓ | ▲ Improved | -49% | 0% |
| case-12 | ✗→✓ | ▲ Improved | -9% | 0% |
<skill> <preflight> Require one active bundle with approved, identity-matched spec.xml and plan.xml. Approved plans are immutable. Read context, projections, assertions, scopes, task dependencies, and verification before editing. Reject phase-incompatible plans before writes: MustPassCommand must be leaf project evidence, and neither target assertions nor post-write task verification may invoke --assertions current or nest GRACE lifecycle commands. Supersede and replan instead of editing an approved conflict in place. </preflight>
<assertion_commands>
grace lint --path PROJECT --assertions currentgrace lint --path PROJECT --change C-ID --assertions baseline (add --run-commands when the baseline declares MustPassCommand)grace lint --path PROJECT --change C-ID --assertions targetgrace lint --path PROJECT --change C-ID --assertions target --run-commandsgrace lint --path PROJECT --change C-ID --assertions final (add --run-commands when the target declares MustPassCommand)grace lint --path PROJECT --parallel-preflight</assertion_commands>
<mode_selection> Wait for explicit sequential or parallel-safe choice. Parallel-safe requires the explicit preflight to pass. Workers never mutate approved plans; durable .grace changes are applied centrally after observed work verifies. </mode_selection>
<recovery_decision_table> | state | required action | | clean-to-start | Run selected baseline, then execute tasks. | | partial-observed-writes | Inspect the declared observed scope and ask whether to resume or revert. | | durable-state-changed | Hard stop; supersede and replan. Approved assertions are immutable. | | target-already-satisfied | Run final end-state validation, opted-in command evidence when declared, durable reconciliation, and ask for explicit apply confirmation. | | unsafe-unknown-drift | Hard stop and report unexplained files. | </recovery_decision_table>
<execution_rules>
--run-commands when its assertions declare MustPassCommand.--assertions final as the outermost lifecycle gate, including --run-commands when MustPassCommand is declared. Final mode performs full project lint, evaluates the selected target, keeps unrelated approved baselines active, and does not re-evaluate the selected plan's superseded baseline.applied and archive the complete bundle.</execution_rules> </skill>
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-07 | fail→pass | 4,734 | 11,968 | +153% | 1 | 1 | 0% | 755 | 1,247 | +65% | 0 | 0 | — |
case-01 | fail→pass | 2,852 | 14,272 | +400% | 1 | 1 | 0% | 405 | 2,024 | +400% | 0 | 0 | — |
case-02 | fail→fail | 2,055 | 5,276 | +157% | 1 | 1 | 0% | 226 | 1,019 | +351% | 0 | 0 | — |
case-03 | fail→fail | 4,850 | 6,233 | +29% | 1 | 1 | 0% | 222 | 966 | +335% | 0 | 0 | — |
case-04 | fail→fail | 5,260 | 3,788 | -28% | 1 | 1 | 0% | 807 | 1,423 | +76% | 0 | 0 | — |
case-05 | pass→pass | 8,361 | 5,026 | -40% | 1 | 1 | 0% | 1,399 | 1,492 | +7% | 0 | 0 | — |
case-06 | pass→pass | 9,515 | 2,488 | -74% | 1 | 1 | 0% | 1,527 | 1,155 | -24% | 0 | 0 | — |
case-08 | fail→pass | 14,648 | 11,164 | -24% | 1 | 1 | 0% | 236 | 1,032 | +337% | 0 | 0 | — |
case-09 | fail→fail | 14,488 | 2,081 | -86% | 1 | 1 | 0% | 2,635 | 1,119 | -58% | 0 | 0 | — |
case-10 | fail→fail | 20,617 | 1,866 | -91% | 1 | 1 | 0% | 2,002 | 1,095 | -45% | 0 | 0 | — |
case-11 | fail→pass | 9,338 | 1,101 | -88% | 1 | 1 | 0% | 1,787 | 918 | -49% | 0 | 0 | — |
case-16 | pass→pass | 4,759 | 1,930 | -59% | 1 | 1 | 0% | 786 | 1,044 | +33% | 0 | 0 | — |
case-12 | fail→pass | 9,217 | 3,583 | -61% | 1 | 1 | 0% | 1,592 | 1,447 | -9% | 0 | 0 | — |
case-13 | pass→pass | 8,765 | 3,427 | -61% | 1 | 1 | 0% | 1,379 | 1,338 | -3% | 0 | 0 | — |
case-14 | fail→pass | 7,199 | 3,508 | -51% | 1 | 1 | 0% | 1,182 | 1,341 | +13% | 0 | 0 | — |
case-15 | pass→pass | 7,871 | 2,423 | -69% | 1 | 1 | 0% | 1,310 | 1,142 | -13% | 0 | 0 | — |
case-17 | fail→pass | 8,582 | 1,991 | -77% | 1 | 1 | 0% | 1,335 | 1,032 | -23% | 0 | 0 | — |
case-18 | pass→pass | 2,485 | 3,424 | +38% | 1 | 1 | 0% | 418 | 1,362 | +226% | 0 | 0 | — |
case-19 | fail→pass | 8,289 | 3,762 | -55% | 1 | 1 | 0% | 1,356 | 1,370 | +1% | 0 | 0 | — |
case-20 | fail→pass | 8,848 | 3,646 | -59% | 1 | 1 | 0% | 1,481 | 1,127 | -24% | 0 | 0 | — |
case-21 | pass→pass | 6,874 | 3,207 | -53% | 1 | 1 | 0% | 1,113 | 1,312 | +18% | 0 | 0 | — |
case-22 | fail→fail | 8,395 | 13,372 | +59% | 1 | 1 | 0% | 1,803 | 3,635 | +102% | 0 | 0 | — |
case-23 | fail→fail | 7,800 | 8,078 | +4% | 1 | 1 | 0% | 1,565 | 2,366 | +51% | 0 | 0 | — |
case-24 | fail→fail | 8,032 | 6,917 | -14% | 1 | 1 | 0% | 1,546 | 2,156 | +39% | 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. 24 cases were attempted, and 21 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 +38 percentage points is the difference between those two pass rates over the 21 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.