Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Use when the user wants a quality review, interaction audit, or to test the workflow against realistic scenarios.
.claude/skills/sharpdeveye-evaluate/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-13 | ✗→✓ | ▲ Improved | 134% | 0% |
| case-01 | ✗→✓ | ▲ Improved | -26% | 0% |
| case-02 | ✗→✓ | ▲ Improved | -34% | 0% |
| case-04 | ✗→✓ | ▲ Improved | 46% | 0% |
| case-05 | ✗→✓ | ▲ Improved | 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 feedback-loops reference in the agent-workflow skill for evaluation patterns, golden test sets, and regression detection.
Evaluate the workflow's actual interaction quality by testing it against scenarios that represent real usage.
1. Task Completion
2. Output Quality
3. Error Behavior
4. User Experience
5. Consistency
Create and run test scenarios:
| Scenario | Input | Expected | Actual | Grade | |----------|-------|----------|--------|-------| | Happy path | Normal input | Correct output | ? | A-F | | Edge case | Unusual input | Graceful handling | ? | A-F | | Error case | Bad input | Helpful error | ? | A-F | | Stress case | Large/complex input | Reasonable handling | ? | A-F | | Adversarial | Tricky/malicious input | Safe response | ? | A-F |
Produce a structured report with:
After evaluation, run /fortify to address error behavior gaps, /refine for output quality improvements, or /iterate to set up continuous quality monitoring.
NEVER:
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-03 | fail→fail | 27,490 | 4,957 | -82% | 1 | 1 | 0% | 4,386 | 953 | -78% | 0 | 0 | — |
case-13 | fail→pass | 9,216 | 16,687 | +81% | 1 | 1 | 0% | 1,389 | 3,247 | +134% | 0 | 0 | — |
case-01 | fail→pass | 29,440 | 19,976 | -32% | 1 | 1 | 0% | 4,885 | 3,637 | -26% | 0 | 0 | — |
case-02 | fail→pass | 36,535 | 21,376 | -41% | 1 | 1 | 0% | 6,201 | 4,104 | -34% | 0 | 0 | — |
case-04 | fail→pass | 21,352 | 28,156 | +32% | 1 | 1 | 0% | 3,326 | 4,866 | +46% | 0 | 0 | — |
case-05 | fail→pass | 16,424 | 19,146 | +17% | 1 | 1 | 0% | 2,677 | 3,575 | +34% | 0 | 0 | — |
case-06 | fail→pass | 15,148 | 49,646 | +228% | 1 | 1 | 0% | 2,334 | 3,069 | +31% | 0 | 0 | — |
case-07 | fail→pass | 13,054 | 13,163 | +1% | 1 | 1 | 0% | 2,014 | 2,622 | +30% | 0 | 0 | — |
case-08 | fail→pass | 10,882 | 6,058 | -44% | 1 | 1 | 0% | 1,666 | 1,464 | -12% | 0 | 0 | — |
case-09 | fail→pass | 11,662 | 5,163 | -56% | 1 | 1 | 0% | 1,743 | 1,491 | -14% | 0 | 0 | — |
case-10 | fail→pass | 14,951 | 14,594 | -2% | 1 | 1 | 0% | 2,361 | 2,924 | +24% | 0 | 0 | — |
case-11 | fail→pass | 9,885 | 10,731 | +9% | 1 | 1 | 0% | 1,552 | 1,758 | +13% | 0 | 0 | — |
case-12 | pass→pass | 10,198 | 9,387 | -8% | 1 | 1 | 0% | 1,494 | 1,432 | -4% | 0 | 0 | — |
case-14 | fail→pass | 10,202 | 4,350 | -57% | 1 | 1 | 0% | 1,383 | 1,343 | -3% | 0 | 0 | — |
case-15 | fail→pass | 7,786 | 2,432 | -69% | 1 | 1 | 0% | 1,148 | 974 | -15% | 0 | 0 | — |
case-16 | pass→fail | 25,675 | 3,405 | -87% | 1 | 1 | 0% | 2,515 | 895 | -64% | 0 | 0 | — |
case-17 | fail→fail | 12,480 | 2,740 | -78% | 1 | 1 | 0% | 1,946 | 802 | -59% | 0 | 0 | — |
case-18 | pass→pass | 7,951 | 6,978 | -12% | 1 | 1 | 0% | 1,222 | 1,743 | +43% | 0 | 0 | — |
case-19 | fail→pass | 13,442 | 5,419 | -60% | 1 | 1 | 0% | 1,265 | 1,231 | -3% | 0 | 0 | — |
case-20 | pass→pass | 14,801 | 17,972 | +21% | 1 | 1 | 0% | 2,991 | 3,728 | +25% | 0 | 0 | — |
case-21 | pass→pass | 10,595 | 16,796 | +59% | 1 | 1 | 0% | 1,641 | 3,174 | +93% | 0 | 0 | — |
case-22 | pass→pass | 14,799 | 19,581 | +32% | 1 | 1 | 0% | 2,369 | 4,306 | +82% | 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. 1 case got worse with the skill loaded, and it is 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.