Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Route tasks via the 3-tier model selector and learned patterns; emits a routing rationale via hooks_explain
.claude/skills/ruvnet-intelligence-route/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-07 | ✗→✓ | ▲ Improved | -30% | 0% |
| case-08 | ✗→✓ | ▲ Improved | -5% | 0% |
| case-09 | ✗→✓ | ▲ Improved | -40% | 0% |
| case-11 | ✗→✓ | ▲ Improved | 135% | 0% |
| case-12 | ✗→✓ | ▲ Improved | 21% | 0% |
Pick the optimal agent + model tier for a task using learned patterns + the 3-tier router. Emits a hooks_explain rationale so the choice is auditable.
Before starting any non-trivial task. Replaces manual agent selection with data-driven decisions.
mcp__plugin_ruflo-core_ruflo__hooks_route with the task description. Returns { recommended, confidence, reasoning }.mcp__plugin_ruflo-core_ruflo__hooks_model-route for Haiku/Sonnet/Opus selection.mcp__plugin_ruflo-core_ruflo__hooks_intelligence_pattern-search to find prior successes.mcp__plugin_ruflo-core_ruflo__neural_predict with the task description for a confidence-scored prediction.--why was passed) — call mcp__plugin_ruflo-core_ruflo__hooks_explain to surface the routing rationale to the user.mcp__plugin_ruflo-core_ruflo__hooks_model-outcome with success: true|false to train the router.| Tier | Handler | Latency | Cost | When | |------|---------|---------|------|------| | 1 | Deterministic codemod (TS compiler) | ~1ms | $0 | Structural transforms with no LLM: var-to-const, remove-console, add-logging | | 2 | Haiku | ~500ms | ~$0.0002 | Low complexity (<30%), bug fixes, quick patches | | 3 | Sonnet/Opus | 2–5s | $0.003–$0.015 | Complex reasoning, architecture, security, multi-file refactors |
When hooks_route returns [CODEMOD_AVAILABLE] for a deterministic intent (var-to-const, remove-console, add-logging), call mcp__plugin_ruflo-core_ruflo__hooks_codemod with the intent + file — it applies the transform via the TypeScript compiler at $0, no LLM. Note: add-types, add-error-handling, async-await require judgement and route to a model (Tier 2/3) per ADR-143; they are NOT $0 codemods. Agent Booster is a fast-apply merge engine for LLM-produced edits, not the Tier-1 path.
Closing the routing loop is mandatory:
bash# Success mcp tool call hooks_model-outcome --json -- '{"taskId": "T123", "success": true, "model": "haiku"}' # Failure with reason mcp tool call hooks_model-outcome --json -- '{"taskId": "T123", "success": false, "model": "haiku", "reason": "complexity-misjudged"}'
The router learns from these calls. Skipping them = no learning.
bashnpx @claude-flow/cli@latest hooks route --task "description" npx @claude-flow/cli@latest hooks pre-task --description "description" npx @claude-flow/cli@latest hooks explain --topic "routing decision"
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-07 | fail→pass | 12,349 | 3,561 | -71% | 1 | 1 | 0% | 2,110 | 1,483 | -30% | 0 | 0 | — |
case-01 | fail→fail | 6,450 | 9,809 | +52% | 1 | 1 | 0% | 836 | 2,218 | +165% | 0 | 0 | — |
case-02 | fail→fail | 11,152 | 5,769 | -48% | 1 | 1 | 0% | 2,036 | 1,373 | -33% | 0 | 0 | — |
case-08 | fail→pass | 9,967 | 3,410 | -66% | 1 | 1 | 0% | 1,528 | 1,444 | -5% | 0 | 0 | — |
case-03 | fail→fail | 14,668 | 12,812 | -13% | 1 | 1 | 0% | 2,636 | 1,374 | -48% | 0 | 0 | — |
case-04 | pass→pass | 13,395 | 14,804 | +11% | 1 | 1 | 0% | 2,708 | 3,551 | +31% | 0 | 0 | — |
case-05 | pass→pass | 8,593 | 7,366 | -14% | 1 | 1 | 0% | 1,736 | 2,381 | +37% | 0 | 0 | — |
case-06 | pass→pass | 9,912 | 8,767 | -12% | 1 | 1 | 0% | 2,161 | 2,867 | +33% | 0 | 0 | — |
case-09 | fail→pass | 13,167 | 2,664 | -80% | 1 | 1 | 0% | 2,133 | 1,285 | -40% | 0 | 0 | — |
case-10 | pass→pass | 9,015 | 6,759 | -25% | 1 | 1 | 0% | 1,486 | 2,137 | +44% | 0 | 0 | — |
case-11 | fail→pass | 3,627 | 3,663 | +1% | 1 | 1 | 0% | 645 | 1,516 | +135% | 0 | 0 | — |
case-12 | fail→pass | 9,256 | 9,322 | +1% | 1 | 1 | 0% | 1,631 | 1,969 | +21% | 0 | 0 | — |
case-18 | fail→pass | 13,879 | 2,067 | -85% | 1 | 1 | 0% | 2,253 | 1,215 | -46% | 0 | 0 | — |
case-13 | fail→pass | 11,422 | 2,440 | -79% | 1 | 1 | 0% | 2,073 | 1,286 | -38% | 0 | 0 | — |
case-14 | fail→fail | 9,635 | 2,515 | -74% | 1 | 1 | 0% | 1,734 | 1,294 | -25% | 0 | 0 | — |
case-15 | fail→pass | 10,242 | 3,248 | -68% | 1 | 1 | 0% | 1,781 | 1,283 | -28% | 0 | 0 | — |
case-16 | fail→pass | 23,379 | 1,524 | -93% | 1 | 1 | 0% | 4,294 | 1,091 | -75% | 0 | 0 | — |
case-17 | fail→pass | 9,981 | 3,711 | -63% | 1 | 1 | 0% | 1,609 | 1,157 | -28% | 0 | 0 | — |
case-19 | pass→pass | 9,989 | 2,824 | -72% | 1 | 1 | 0% | 1,627 | 1,334 | -18% | 0 | 0 | — |
case-20 | fail→pass | 10,662 | 2,687 | -75% | 1 | 1 | 0% | 1,808 | 1,316 | -27% | 0 | 0 | — |
case-21 | fail→pass | 9,341 | 2,408 | -74% | 1 | 1 | 0% | 1,757 | 1,274 | -27% | 0 | 0 | — |
case-22 | fail→pass | 9,587 | 4,159 | -57% | 1 | 1 | 0% | 1,724 | 1,798 | +4% | 0 | 0 | — |
case-23 | fail→pass | 13,914 | 1,835 | -87% | 1 | 1 | 0% | 2,419 | 1,144 | -53% | 0 | 0 | — |
case-24 | fail→pass | 12,008 | 1,691 | -86% | 1 | 1 | 0% | 2,170 | 1,086 | -50% | 0 | 0 | — |
case-25 | fail→pass | 11,734 | 4,325 | -63% | 1 | 1 | 0% | 2,154 | 1,108 | -49% | 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. 25 cases were attempted, and 23 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 +64 percentage points is the difference between those two pass rates over the 23 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.