Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Feature specification creation from codebase research. Produces requirements, acceptance criteria, architecture decisions, implementation plans, and risk analysis.
.claude/skills/a5c-ai-spec-creation/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | -36% | 0% |
| case-03 | ✗→✓ | ▲ Improved | -11% | 0% |
| case-07 | ✗→✓ | ▲ Improved | 10% | 0% |
| case-15 | ✗→✓ | ▲ Improved | 7% | 0% |
| case-18 | ✗→✓ | ▲ Improved | -36% | 0% |
Clear boundaries on what the feature does and does not include.
Detailed requirements with unique identifiers for tracking.
Testable, measurable criteria for each requirement.
Decision records with rationale and alternatives considered.
Phased approach ordered by dependency, not priority.
Identified risks with probability, impact, and mitigation strategies.
Interface definitions and data model schemas.
Mapping of unit, integration, and E2E tests to requirements.
Specifications are saved to docs/specs/{feature}.md for reference by the execution workflow.
/spec:create [feature] slash commandclaudekit-spec-workflow (create mode)| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→pass | 35,581 | 20,233 | -43% | 1 | 1 | 0% | 6,189 | 3,943 | -36% | 0 | 0 | — |
case-02 | fail→fail | 23,827 | 3,777 | -84% | 1 | 1 | 0% | 4,798 | 502 | -90% | 0 | 0 | — |
case-03 | fail→pass | 33,699 | 27,453 | -19% | 1 | 1 | 0% | 6,184 | 5,492 | -11% | 0 | 0 | — |
case-04 | fail→fail | 16,764 | 18,831 | +12% | 1 | 1 | 0% | 3,896 | 4,631 | +19% | 0 | 0 | — |
case-05 | fail→fail | 3,988 | 2,429 | -39% | 1 | 1 | 0% | 589 | 476 | -19% | 0 | 0 | — |
case-06 | fail→fail | 19,309 | 24,815 | +29% | 1 | 1 | 0% | 3,220 | 4,315 | +34% | 0 | 0 | — |
case-07 | fail→pass | 13,579 | 13,537 | -0% | 1 | 1 | 0% | 2,212 | 2,436 | +10% | 0 | 0 | — |
case-08 | fail→fail | 17,727 | 17,687 | -0% | 1 | 1 | 0% | 2,779 | 3,413 | +23% | 0 | 0 | — |
case-09 | fail→fail | 14,452 | 11,646 | -19% | 1 | 1 | 0% | 2,604 | 2,318 | -11% | 0 | 0 | — |
case-10 | fail→fail | 14,470 | 16,166 | +12% | 1 | 1 | 0% | 2,413 | 2,850 | +18% | 0 | 0 | — |
case-11 | fail→fail | 19,367 | 15,141 | -22% | 1 | 1 | 0% | 2,862 | 2,854 | -0% | 0 | 0 | — |
case-12 | fail→fail | 27,736 | 20,509 | -26% | 1 | 1 | 0% | 5,125 | 3,799 | -26% | 0 | 0 | — |
case-13 | fail→fail | 10,345 | 7,053 | -32% | 1 | 1 | 0% | 1,778 | 1,320 | -26% | 0 | 0 | — |
case-14 | fail→fail | 12,240 | 9,361 | -24% | 1 | 1 | 0% | 2,261 | 1,921 | -15% | 0 | 0 | — |
case-15 | fail→pass | 15,772 | 15,100 | -4% | 1 | 1 | 0% | 2,479 | 2,649 | +7% | 0 | 0 | — |
case-16 | fail→fail | 15,209 | 12,554 | -17% | 1 | 1 | 0% | 2,824 | 2,719 | -4% | 0 | 0 | — |
case-17 | fail→fail | 18,243 | 26,098 | +43% | 1 | 1 | 0% | 3,200 | 5,045 | +58% | 0 | 0 | — |
case-18 | fail→pass | 14,397 | 8,111 | -44% | 1 | 1 | 0% | 2,307 | 1,468 | -36% | 0 | 0 | — |
case-19 | fail→fail | 9,646 | 2,329 | -76% | 1 | 1 | 0% | 1,572 | 623 | -60% | 0 | 0 | — |
case-20 | fail→fail | 18,553 | 11,946 | -36% | 1 | 1 | 0% | 2,646 | 2,166 | -18% | 0 | 0 | — |
case-21 | fail→pass | 13,403 | 11,889 | -11% | 1 | 1 | 0% | 2,599 | 2,730 | +5% | 0 | 0 | — |
case-22 | fail→fail | 22,926 | 24,525 | +7% | 1 | 1 | 0% | 4,541 | 4,213 | -7% | 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 21 counted toward the lift figure. The other 1 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 +27 percentage points is the difference between those two pass rates over the 21 comparable cases. 2 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.