Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Use when a single agent demonstrably cannot handle the task and multi-agent coordination is justified.
.claude/skills/sharpdeveye-compose/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-04 | ✗→✓ | ▲ Improved | 10% | 0% |
| case-07 | ✗→✓ | ▲ Improved | 22% | 0% |
| case-08 | ✗→✓ | ▲ Improved | 0% | 0% |
| case-09 | ✗→✓ | ▲ Improved | 20% | 0% |
| case-11 | ✗→✓ | ▲ Improved | 23% | 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.
Design a multi-agent system. But first — are you sure you need one?
Answer these before proceeding:
If you can't articulate a specific limitation, use /amplify on the single agent instead.
Choose the right architecture pattern (consult the agent-architecture reference in the agent-workflow skill):
For each agent in the system, define:
markdown## Agent: [Name] Role: [One sentence] Responsibilities: [What it does] Boundaries: [What it does NOT do] Tools: [List of tools this agent has access to] Input: [What it receives] Output: [What it produces]
For each agent-to-agent connection:
markdown## Handoff: [Agent A] → [Agent B] Trigger: [When does A hand off to B?] Payload: [What data is passed?] Expected response: [What does A expect back?] Timeout: [How long to wait?] Failure handling: [What if B fails?]
Every multi-agent system needs a supervisor:
After composition, run /fortify to add error handling at every handoff, then /evaluate to test the multi-agent system end-to-end.
NEVER:
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-03 | pass→pass | 36,079 | 24,381 | -32% | 1 | 1 | 0% | 6,211 | 4,791 | -23% | 0 | 0 | — |
case-04 | fail→pass | 13,463 | 10,360 | -23% | 1 | 1 | 0% | 2,129 | 2,338 | +10% | 0 | 0 | — |
case-05 | pass→pass | 16,157 | 10,672 | -34% | 1 | 1 | 0% | 2,406 | 2,337 | -3% | 0 | 0 | — |
case-06 | pass→pass | 12,040 | 10,218 | -15% | 1 | 1 | 0% | 1,945 | 2,240 | +15% | 0 | 0 | — |
case-07 | fail→pass | 18,453 | 19,333 | +5% | 1 | 1 | 0% | 3,211 | 3,932 | +22% | 0 | 0 | — |
case-01 | fail→fail | 32,655 | 2,270 | -93% | 1 | 1 | 0% | 6,209 | 807 | -87% | 0 | 0 | — |
case-02 | pass→pass | 26,290 | 19,419 | -26% | 1 | 1 | 0% | 4,701 | 4,018 | -15% | 0 | 0 | — |
case-08 | fail→pass | 21,445 | 18,774 | -12% | 1 | 1 | 0% | 3,660 | 3,674 | +0% | 0 | 0 | — |
case-09 | fail→pass | 19,139 | 18,633 | -3% | 1 | 1 | 0% | 3,074 | 3,680 | +20% | 0 | 0 | — |
case-10 | pass→fail | 14,902 | 2,820 | -81% | 1 | 1 | 0% | 2,615 | 1,082 | -59% | 0 | 0 | — |
case-11 | fail→pass | 15,682 | 15,812 | +1% | 1 | 1 | 0% | 2,674 | 3,283 | +23% | 0 | 0 | — |
case-12 | fail→pass | 9,585 | 14,260 | +49% | 1 | 1 | 0% | 1,662 | 2,716 | +63% | 0 | 0 | — |
case-13 | fail→fail | 4,420 | 3,703 | -16% | 1 | 1 | 0% | 726 | 906 | +25% | 0 | 0 | — |
case-14 | fail→pass | 20,444 | 14,181 | -31% | 1 | 1 | 0% | 3,043 | 2,773 | -9% | 0 | 0 | — |
case-15 | fail→pass | 22,709 | 17,956 | -21% | 1 | 1 | 0% | 4,033 | 3,749 | -7% | 0 | 0 | — |
case-16 | fail→pass | 17,107 | 18,469 | +8% | 1 | 1 | 0% | 3,032 | 3,617 | +19% | 0 | 0 | — |
case-17 | pass→pass | 15,448 | 8,700 | -44% | 1 | 1 | 0% | 2,312 | 2,024 | -12% | 0 | 0 | — |
case-18 | fail→pass | 16,976 | 19,797 | +17% | 1 | 1 | 0% | 2,706 | 3,773 | +39% | 0 | 0 | — |
case-19 | pass→pass | 13,714 | 16,383 | +19% | 1 | 1 | 0% | 2,063 | 3,082 | +49% | 0 | 0 | — |
case-20 | fail→pass | 20,484 | 14,354 | -30% | 1 | 1 | 0% | 3,516 | 2,994 | -15% | 0 | 0 | — |
case-21 | pass→pass | 13,177 | 12,746 | -3% | 1 | 1 | 0% | 2,133 | 2,496 | +17% | 0 | 0 | — |
case-22 | pass→pass | 15,076 | 17,802 | +18% | 1 | 1 | 0% | 2,434 | 3,326 | +37% | 0 | 0 | — |
case-23 | fail→pass | 16,245 | 18,879 | +16% | 1 | 1 | 0% | 2,654 | 3,465 | +31% | 0 | 0 | — |
case-24 | pass→pass | 15,532 | 13,587 | -13% | 1 | 1 | 0% | 2,378 | 2,786 | +17% | 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. 24 cases were attempted, and 22 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 +46 percentage points is the difference between those two pass rates over the 22 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.