Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Use when the workflow feels over-engineered, has premature optimizations, unnecessary abstraction layers, or complexity beyond actual requirements.
.claude/skills/sharpdeveye-temper/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-10 | ✗→✓ | ▲ Improved | 12% | 0% |
| case-17 | ✗→✓ | ▲ Improved | 27% | 0% |
| case-18 | ✗→✓ | ▲ Improved | -15% | 0% |
| case-19 | ✓→✗ | ▼ Worse | -41% | 0% |
| case-06 | ✓→✓ | ＝ Same ✓ | 34% | 0% |
Invoke /agent-workflow — it contains workflow principles, anti-patterns, and the Context Gathering Protocol. Follow the protocol before proceeding — if no workflow context exists yet, you MUST run /teach-maestro first. Consult the agent-architecture reference in the agent-workflow skill for topology patterns and when multi-agent is justified.
Pull back from over-engineering. The most common mistake isn't building too little — it's building too much.
Signs you've over-engineered:
For each component:
Collapse Unnecessary Agents
textOVER-ENGINEERED: User → Classifier → Router → Specialist → Formatter → Checker (6 components) TEMPERED: User → Single Agent with good prompt (1 component, same quality)
Remove Premature Abstraction
textOVER-ENGINEERED: class AgentOrchestrator with 5 strategy interfaces TEMPERED: async function runWorkflow(input) — direct, readable
Simplify Configuration
textOVER-ENGINEERED: config.yaml (200 lines, 47 params, 3 inheritance levels) TEMPERED: config.yaml (20 lines, essential params only, sensible defaults)
After tempering, run /evaluate to confirm quality is preserved, or /diagnose for a full health check.
NEVER:
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-06 | pass→pass | 11,390 | 10,861 | -5% | 1 | 1 | 0% | 1,822 | 2,434 | +34% | 0 | 0 | — |
case-01 | fail→fail | 26,856 | 22,572 | -16% | 1 | 1 | 0% | 3,486 | 4,287 | +23% | 0 | 0 | — |
case-02 | fail→fail | 10,889 | 4,193 | -61% | 1 | 1 | 0% | 1,872 | 859 | -54% | 0 | 0 | — |
case-03 | fail→fail | 17,238 | 21,142 | +23% | 1 | 1 | 0% | 2,782 | 3,568 | +28% | 0 | 0 | — |
case-04 | fail→fail | 14,725 | 3,767 | -74% | 1 | 1 | 0% | 2,318 | 671 | -71% | 0 | 0 | — |
case-05 | pass→pass | 12,157 | 10,071 | -17% | 1 | 1 | 0% | 1,703 | 2,100 | +23% | 0 | 0 | — |
case-07 | pass→pass | 12,566 | 7,949 | -37% | 1 | 1 | 0% | 1,997 | 1,854 | -7% | 0 | 0 | — |
case-08 | pass→pass | 13,428 | 9,516 | -29% | 1 | 1 | 0% | 1,882 | 1,623 | -14% | 0 | 0 | — |
case-09 | pass→pass | 13,877 | 11,925 | -14% | 1 | 1 | 0% | 2,051 | 1,833 | -11% | 0 | 0 | — |
case-10 | fail→pass | 13,336 | 10,370 | -22% | 1 | 1 | 0% | 1,976 | 2,214 | +12% | 0 | 0 | — |
case-11 | pass→pass | 10,840 | 5,787 | -47% | 1 | 1 | 0% | 1,646 | 1,407 | -15% | 0 | 0 | — |
case-12 | pass→pass | 14,498 | 10,199 | -30% | 1 | 1 | 0% | 2,133 | 2,071 | -3% | 0 | 0 | — |
case-13 | pass→pass | 13,975 | 8,755 | -37% | 1 | 1 | 0% | 1,996 | 1,938 | -3% | 0 | 0 | — |
case-14 | pass→pass | 15,376 | 13,459 | -12% | 1 | 1 | 0% | 2,246 | 2,693 | +20% | 0 | 0 | — |
case-15 | pass→pass | 12,023 | 10,198 | -15% | 1 | 1 | 0% | 1,894 | 2,112 | +12% | 0 | 0 | — |
case-16 | pass→pass | 12,877 | 9,920 | -23% | 1 | 1 | 0% | 1,926 | 2,007 | +4% | 0 | 0 | — |
case-17 | fail→pass | 11,113 | 12,777 | +15% | 1 | 1 | 0% | 1,744 | 2,215 | +27% | 0 | 0 | — |
case-18 | fail→pass | 6,393 | 2,552 | -60% | 1 | 1 | 0% | 1,012 | 857 | -15% | 0 | 0 | — |
case-19 | pass→fail | 11,844 | 8,750 | -26% | 1 | 1 | 0% | 1,805 | 1,062 | -41% | 0 | 0 | — |
case-20 | pass→pass | 14,807 | 10,893 | -26% | 1 | 1 | 0% | 3,015 | 2,519 | -16% | 0 | 0 | — |
case-21 | pass→pass | 10,824 | 7,944 | -27% | 1 | 1 | 0% | 2,007 | 1,972 | -2% | 0 | 0 | — |
case-22 | pass→pass | 18,173 | 10,559 | -42% | 1 | 1 | 0% | 2,756 | 2,505 | -9% | 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 +9 percentage points is the difference between those two pass rates over the 19 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.