Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Route tasks through hooks_route, partition by Agent Booster availability, and report Tier 1 bypass utilization with $0 cost
.claude/skills/ruvnet-cost-booster-route/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-18 | ✗→✓ | ▲ Improved | -7% | 0% |
| case-08 | ✗→✓ | ▲ Improved | 31% | 0% |
| case-09 | ✗→✓ | ▲ Improved | 26% | 0% |
| case-10 | ✗→✓ | ▲ Improved | -21% | 0% |
| case-11 | ✗→✓ | ▲ Improved | -44% | 0% |
Wraps mcp__plugin_ruflo-core_ruflo__hooks_route and reports how many tasks the 3-tier router classified as Agent Booster (Tier 1) eligible. Tier 1 bypasses run as WASM transforms — no LLM call, structurally $0 cost.
Before a batch of similar tasks, or when cost-report shows Sonnet/Opus spend on descriptions that look like simple transforms (var-to-const, add-types, add-error-handling, async-await, add-logging, remove-console).
cost-tracking via memory_search. Cap batch at 50.hooks_route with the description; capture the full response string.[AGENT_BOOSTER_AVAILABLE] (Tier 1) vs. not (Tier 2/3). Extract [TASK_MODEL_RECOMMENDATION] Use model="X" when present. === Booster bypass report === Tasks analyzed: 50 Tier 1 (booster): 18 (36%) — $0.00 Tier 2/3 (LLM): 32 (64%) — $X.XX (upper-bound) Booster intents: var-to-const (8), add-types (5), remove-console (5)
memory_store --namespace cost-tracking --key "booster-route-$(date +%Y%m%d-%H%M%S)" --value '{"tier1": N, "tier2_or_3": M, ...}' so cost-report picks up the tier signal.[AGENT_BOOSTER_AVAILABLE] fires only when the upstream router populates routeResult.agentBoosterIntent.type (v3/@claude-flow/cli/src/mcp-tools/hooks-tools.ts:1228). The published CLI's semantic-VectorDb path does not always trigger the classifier — treat the partition as a lower bound on Tier 1 eligibility.<1ms latency and 352× faster than LLM. <1ms and $0 are structural; 352× is claimed upstream, not yet verified here. Report what the router actually returns.docs/benchmarks/0002-baseline.md for the full upstream-claims-vs-measured table.ADR-0002 Decision #1 · ruflo-intelligence ADR-0001 §"Neutral" (closes the routing-outcomes loop via cost-optimize step 8) · CLAUDE.md root §"3-Tier Model Routing (ADR-026)".
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-18 | fail→pass | 12,915 | 7,048 | -45% | 1 | 1 | 0% | 2,524 | 2,350 | -7% | 0 | 0 | — |
case-01 | fail→fail | 3,661 | 5,236 | +43% | 1 | 1 | 0% | 306 | 1,195 | +291% | 0 | 0 | — |
case-02 | fail→fail | 11,953 | 6,011 | -50% | 1 | 1 | 0% | 2,585 | 1,264 | -51% | 0 | 0 | — |
case-03 | fail→fail | 9,606 | 7,462 | -22% | 1 | 1 | 0% | 2,345 | 1,124 | -52% | 0 | 0 | — |
case-04 | pass→fail | 10,515 | 5,121 | -51% | 1 | 1 | 0% | 2,454 | 1,059 | -57% | 0 | 0 | — |
case-05 | fail→fail | 3,800 | 8,795 | +131% | 1 | 1 | 0% | 707 | 2,377 | +236% | 0 | 0 | — |
case-06 | pass→pass | 10,354 | 13,998 | +35% | 1 | 1 | 0% | 2,209 | 3,734 | +69% | 0 | 0 | — |
case-07 | fail→fail | 7,491 | 14,840 | +98% | 1 | 1 | 0% | 695 | 1,957 | +182% | 0 | 0 | — |
case-08 | fail→pass | 30,963 | 7,008 | -77% | 1 | 1 | 0% | 910 | 1,191 | +31% | 0 | 0 | — |
case-09 | fail→pass | 4,945 | 1,975 | -60% | 1 | 1 | 0% | 914 | 1,155 | +26% | 0 | 0 | — |
case-10 | fail→pass | 9,614 | 3,915 | -59% | 1 | 1 | 0% | 1,806 | 1,435 | -21% | 0 | 0 | — |
case-11 | fail→pass | 10,526 | 1,628 | -85% | 1 | 1 | 0% | 1,891 | 1,056 | -44% | 0 | 0 | — |
case-12 | fail→pass | 8,330 | 5,272 | -37% | 1 | 1 | 0% | 1,590 | 1,184 | -26% | 0 | 0 | — |
case-13 | fail→pass | 10,551 | 8,017 | -24% | 1 | 1 | 0% | 1,989 | 2,347 | +18% | 0 | 0 | — |
case-14 | fail→pass | 11,470 | 7,403 | -35% | 1 | 1 | 0% | 2,009 | 2,171 | +8% | 0 | 0 | — |
case-15 | pass→pass | 11,548 | 4,186 | -64% | 1 | 1 | 0% | 2,114 | 1,535 | -27% | 0 | 0 | — |
case-16 | pass→pass | 13,262 | 6,952 | -48% | 1 | 1 | 0% | 2,488 | 2,361 | -5% | 0 | 0 | — |
case-17 | fail→pass | 10,944 | 6,748 | -38% | 1 | 1 | 0% | 1,902 | 1,885 | -1% | 0 | 0 | — |
case-19 | fail→pass | 11,442 | 3,880 | -66% | 1 | 1 | 0% | 2,605 | 1,583 | -39% | 0 | 0 | — |
case-20 | fail→fail | 11,000 | 2,718 | -75% | 1 | 1 | 0% | 1,641 | 1,245 | -24% | 0 | 0 | — |
case-21 | fail→fail | 9,667 | 4,374 | -55% | 1 | 1 | 0% | 1,855 | 957 | -48% | 0 | 0 | — |
case-22 | fail→pass | 4,435 | 1,986 | -55% | 1 | 1 | 0% | 707 | 1,086 | +54% | 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 16 counted toward the lift figure. The other 6 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 +45 percentage points is the difference between those two pass rates over the 16 comparable cases. 3 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.