Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →[BETA] Offload plan phase to Claude Code's ultraplan cloud — drafts remotely while terminal stays free, review in browser with inline comments, import back via /gsd:import. Claude Code only.
.claude/skills/davepoon-gsd-ultraplan-phase/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-04 | ✗→✓ | ▲ Improved | -67% | 0% |
| case-05 | ✗→✓ | ▲ Improved | -37% | 0% |
| case-06 | ✗→✓ | ▲ Improved | -69% | 0% |
| case-09 | ✗→✓ | ▲ Improved | -70% | 0% |
| case-10 | ✗→✓ | ▲ Improved | -32% | 0% |
<objective> Offload GSD's plan phase to Claude Code's ultraplan cloud infrastructure.
Ultraplan drafts the plan in a remote cloud session while your terminal stays free. Review and comment on the plan in your browser, then import it back via /gsd:import --from.
⚠ BETA: ultraplan is in research preview. Use /gsd:plan-phase for stable local planning. Requirements: Claude Code v2.1.91+, claude.ai account, GitHub repository. </objective>
<execution_context> @${CLAUDE_PLUGIN_ROOT}/workflows/ultraplan-phase.md @${CLAUDE_PLUGIN_ROOT}/references/ui-brand.md </execution_context>
<context> $ARGUMENTS </context>
<process> Execute the ultraplan-phase workflow end-to-end. </process>
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 4,584 | 5,507 | +20% | 1 | 1 | 0% | 713 | 556 | -22% | 0 | 0 | — |
case-02 | fail→fail | 6,089 | 6,064 | -0% | 1 | 1 | 0% | 985 | 457 | -54% | 0 | 0 | — |
case-03 | fail→fail | 4,306 | 4,559 | +6% | 1 | 1 | 0% | 731 | 365 | -50% | 0 | 0 | — |
case-04 | fail→pass | 12,069 | 3,181 | -74% | 1 | 1 | 0% | 2,018 | 666 | -67% | 0 | 0 | — |
case-05 | fail→pass | 6,206 | 2,645 | -57% | 1 | 1 | 0% | 963 | 602 | -37% | 0 | 0 | — |
case-06 | fail→pass | 6,200 | 1,182 | -81% | 1 | 1 | 0% | 1,103 | 344 | -69% | 0 | 0 | — |
case-07 | pass→pass | 7,873 | 1,289 | -84% | 1 | 1 | 0% | 1,270 | 367 | -71% | 0 | 0 | — |
case-08 | pass→pass | 8,819 | 2,679 | -70% | 1 | 1 | 0% | 1,466 | 555 | -62% | 0 | 0 | — |
case-09 | fail→pass | 11,794 | 2,057 | -83% | 1 | 1 | 0% | 1,860 | 553 | -70% | 0 | 0 | — |
case-10 | fail→pass | 7,740 | 3,833 | -50% | 1 | 1 | 0% | 1,286 | 878 | -32% | 0 | 0 | — |
case-11 | fail→pass | 9,344 | 4,230 | -55% | 1 | 1 | 0% | 1,533 | 950 | -38% | 0 | 0 | — |
case-12 | pass→pass | 6,247 | 1,927 | -69% | 1 | 1 | 0% | 986 | 468 | -53% | 0 | 0 | — |
case-13 | pass→pass | 10,144 | 2,475 | -76% | 1 | 1 | 0% | 1,672 | 651 | -61% | 0 | 0 | — |
case-14 | pass→pass | 9,829 | 3,048 | -69% | 1 | 1 | 0% | 1,572 | 667 | -58% | 0 | 0 | — |
case-15 | fail→pass | 10,267 | 2,124 | -79% | 1 | 1 | 0% | 1,680 | 501 | -70% | 0 | 0 | — |
case-16 | fail→pass | 11,418 | 4,488 | -61% | 1 | 1 | 0% | 1,991 | 1,005 | -50% | 0 | 0 | — |
case-17 | fail→pass | 10,822 | 2,315 | -79% | 1 | 1 | 0% | 1,792 | 567 | -68% | 0 | 0 | — |
case-18 | fail→pass | 6,933 | 2,133 | -69% | 1 | 1 | 0% | 1,174 | 634 | -46% | 0 | 0 | — |
case-19 | fail→pass | 4,805 | 1,539 | -68% | 1 | 1 | 0% | 742 | 406 | -45% | 0 | 0 | — |
case-20 | fail→pass | 5,256 | 3,149 | -40% | 1 | 1 | 0% | 912 | 713 | -22% | 0 | 0 | — |
case-21 | fail→fail | 5,960 | 3,172 | -47% | 1 | 1 | 0% | 1,106 | 794 | -28% | 0 | 0 | — |
case-22 | fail→pass | 7,580 | 3,238 | -57% | 1 | 1 | 0% | 1,266 | 742 | -41% | 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 +59 percentage points is the difference between those two pass rates over the 19 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.