Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Plan-first development workflow with review gates. Use when implementing features, refactoring, or any task requiring structured planning, iterative implementation with reviews, and clean commits. Triggers on requests like "implement feature X", "plan and build", "spec-driven development", or when user wants disciplined, reviewed code changes.
.claude/skills/dicklesworthstone-specs-dev/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-02 | ✗→✓ | ▲ Improved | -5% | 0% |
| case-03 | ✗→✓ | ▲ Improved | 13% | 0% |
| case-06 | ✗→✓ | ▲ Improved | -4% | 0% |
| case-08 | ✗→✓ | ▲ Improved | 89% | 0% |
| case-09 | ✗→✓ | ▲ Improved | 26% | 0% |
A disciplined, review-gated development workflow ensuring quality through structured planning and iterative implementation.
| Phase | Purpose | Exit Criteria | | ----------------- | ----------------------------- | ---------------------------- | | 1. Discovery | Understand requirements | User approves summary | | 2. Planning | Create reviewed plan | Plan reviewed and approved | | 3. Implementation | Iterative coding with reviews | All tasks complete, reviewed | | 4. Completion | Final validation | Tests pass, docs updated |
Goal: Reach shared understanding before planning.
Gate A: "Do I understand correctly? Should I proceed to create the plan?" — Wait for approval.
Goal: Create a comprehensive, reviewed implementation plan.
references/templates/plan.mdreferences/agents/reviewer.md.agents/sessions/{YYYY-MM-DD}-{feature-name}/plan.md and tasks.md (use references/templates/)Quality gates: see references/gates.md
Goal: Implement tasks iteratively with approval-gated review loops.
> ⚠️ MANDATORY: You MUST follow references/loop.md — Read and execute the implementation loop exactly as specified. Do not skip or deviate from the defined state machine.
Summary: For each task:
IMPLEMENTING → VALIDATING → REVIEWING → loop until approved → COMMITTING → DOCUMENTING → NEXT TASKRequired Steps:
references/loop.md before starting any implementationreferences/agents/worker.md, references/agents/reviewer.mdQuality gates: see references/gates.md
Goal: Final validation and wrap-up.
plan.md with results, final status, known issuestasks.mdQuality gates: see references/gates.md
Reviewer — Plan reviews, code reviews:
Context: references/agents/reviewer.md
Task: Review [plan/code] for completeness, security, performance, patternsWorker — Focused implementation:
Context: references/agents/worker.md
Task: Implement [objective] in [files] with [acceptance criteria].agents/sessions/{YYYY-MM-DD}-{feature-name}/
├── plan.md # Strategic plan
└── tasks.md # Tactical tasksreferences/
├── loop.md # Phase 3 state machine, steps, fix routing
├── gates.md # Quality gates for all phases
├── help.md # Common issues, best practices
├── agents/
│ ├── reviewer.md # Reviewer subagent context
│ └── worker.md # Worker subagent context
└── templates/
├── plan.md # Plan document template
└── tasks.md # Tasks document template| File | When to Read | | -------------------- | --------------------------------- | | loop.md | Phase 3 (MANDATORY — must follow) | | agents/reviewer.md | Plan/code reviews | | agents/worker.md | Task implementation | | templates/plan.md | Phase 2 | | templates/tasks.md | Phase 2 | | gates.md | Each phase exit | | help.md | When stuck |
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 7,924 | 4,291 | -46% | 1 | 1 | 0% | 1,284 | 1,354 | +5% | 0 | 0 | — |
case-02 | fail→pass | 10,165 | 3,427 | -66% | 1 | 1 | 0% | 1,738 | 1,648 | -5% | 0 | 0 | — |
case-03 | fail→pass | 7,865 | 3,662 | -53% | 1 | 1 | 0% | 1,499 | 1,690 | +13% | 0 | 0 | — |
case-04 | pass→pass | 6,769 | 1,937 | -71% | 1 | 1 | 0% | 1,103 | 1,395 | +26% | 0 | 0 | — |
case-05 | pass→pass | 9,787 | 2,346 | -76% | 1 | 1 | 0% | 1,426 | 1,391 | -2% | 0 | 0 | — |
case-06 | fail→pass | 9,163 | 3,069 | -67% | 1 | 1 | 0% | 1,597 | 1,540 | -4% | 0 | 0 | — |
case-07 | pass→pass | 5,378 | 2,623 | -51% | 1 | 1 | 0% | 835 | 1,507 | +80% | 0 | 0 | — |
case-08 | fail→pass | 5,495 | 2,613 | -52% | 1 | 1 | 0% | 750 | 1,417 | +89% | 0 | 0 | — |
case-09 | fail→pass | 7,149 | 2,117 | -70% | 1 | 1 | 0% | 1,091 | 1,372 | +26% | 0 | 0 | — |
case-10 | pass→pass | 8,253 | 2,751 | -67% | 1 | 1 | 0% | 1,403 | 1,464 | +4% | 0 | 0 | — |
case-11 | fail→pass | 10,470 | 4,185 | -60% | 1 | 1 | 0% | 1,402 | 1,675 | +19% | 0 | 0 | — |
case-12 | fail→pass | 11,200 | 1,847 | -84% | 1 | 1 | 0% | 1,740 | 1,354 | -22% | 0 | 0 | — |
case-13 | fail→pass | 13,388 | 2,166 | -84% | 1 | 1 | 0% | 2,192 | 1,356 | -38% | 0 | 0 | — |
case-14 | pass→pass | 9,758 | 2,906 | -70% | 1 | 1 | 0% | 1,223 | 1,332 | +9% | 0 | 0 | — |
case-15 | pass→pass | 10,936 | 3,260 | -70% | 1 | 1 | 0% | 1,595 | 1,571 | -2% | 0 | 0 | — |
case-16 | fail→pass | 6,938 | 2,037 | -71% | 1 | 1 | 0% | 1,129 | 1,354 | +20% | 0 | 0 | — |
case-17 | fail→pass | 12,417 | 2,679 | -78% | 1 | 1 | 0% | 2,188 | 1,453 | -34% | 0 | 0 | — |
case-18 | fail→pass | 7,557 | 2,928 | -61% | 1 | 1 | 0% | 1,107 | 1,397 | +26% | 0 | 0 | — |
case-19 | fail→pass | 8,598 | 2,479 | -71% | 1 | 1 | 0% | 1,204 | 1,404 | +17% | 0 | 0 | — |
case-20 | fail→pass | 8,738 | 2,863 | -67% | 1 | 1 | 0% | 1,426 | 1,454 | +2% | 0 | 0 | — |
case-21 | pass→fail | 5,990 | 6,539 | +9% | 1 | 1 | 0% | 559 | 1,344 | +140% | 0 | 0 | — |
case-22 | fail→fail | 4,749 | 10,063 | +112% | 1 | 1 | 0% | 619 | 2,685 | +334% | 0 | 0 | — |
case-23 | pass→pass | 2,704 | 3,474 | +28% | 1 | 1 | 0% | 430 | 1,473 | +243% | 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. 23 cases were attempted, and 21 counted toward the lift figure. The other 2 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 +52 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.