Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Per-conversation cost view — list every session in cost-tracking with started-at, message count, top model, and total cost
.claude/skills/ruvnet-cost-conversation/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-02 | ✗→✓ | ▲ Improved | -46% | 0% |
| case-03 | ✗→✓ | ▲ Improved | -76% | 0% |
| case-06 | ✗→✓ | ▲ Improved | -53% | 0% |
| case-07 | ✗→✓ | ▲ Improved | -68% | 0% |
| case-08 | ✗→✓ | ▲ Improved | -51% | 0% |
cost-report and cost-optimize aggregate by agent and model. This skill aggregates by conversation (session) — a different lens that surfaces which conversations cost the most. Useful for retrospectives ("which sessions ran long on Opus?") and for evaluating whether a given project's session pattern is sustainable.
CONV_NAMESPACE=cost-tracking-<project>.bash node plugins/ruflo-cost-tracker/scripts/conversation.mjs
Optional env:
CONV_FORMAT=json — emit JSON instead of markdownCONV_LIMIT=20 — show only the most recent N conversationsCONV_NAMESPACE=cost-tracking — override target namespacecost-track — the producer that populates cost-tracking:session-*cost-report — same data, per-agent / per-model lenscost-trend — drift across bench runs (different axis: corpus runs vs conversations)cost-budget-check — sums across conversations to evaluate the budget threshold| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 9,502 | 2,932 | -69% | 1 | 1 | 0% | 1,788 | 634 | -65% | 0 | 0 | — |
case-02 | fail→pass | 6,782 | 4,943 | -27% | 1 | 1 | 0% | 1,236 | 668 | -46% | 0 | 0 | — |
case-03 | fail→pass | 10,776 | 1,447 | -87% | 1 | 1 | 0% | 2,372 | 560 | -76% | 0 | 0 | — |
case-04 | pass→pass | 14,736 | 3,862 | -74% | 1 | 1 | 0% | 3,332 | 1,094 | -67% | 0 | 0 | — |
case-05 | pass→pass | 12,618 | 2,107 | -83% | 1 | 1 | 0% | 2,352 | 761 | -68% | 0 | 0 | — |
case-06 | fail→pass | 13,711 | 5,083 | -63% | 1 | 1 | 0% | 2,817 | 1,331 | -53% | 0 | 0 | — |
case-07 | fail→pass | 11,116 | 1,477 | -87% | 1 | 1 | 0% | 2,009 | 636 | -68% | 0 | 0 | — |
case-08 | fail→pass | 7,978 | 2,090 | -74% | 1 | 1 | 0% | 1,529 | 743 | -51% | 0 | 0 | — |
case-09 | fail→pass | 10,107 | 1,951 | -81% | 1 | 1 | 0% | 1,743 | 701 | -60% | 0 | 0 | — |
case-10 | fail→pass | 10,496 | 1,308 | -88% | 1 | 1 | 0% | 1,836 | 525 | -71% | 0 | 0 | — |
case-11 | pass→pass | 7,110 | 1,893 | -73% | 1 | 1 | 0% | 1,243 | 692 | -44% | 0 | 0 | — |
case-12 | fail→pass | 7,915 | 1,243 | -84% | 1 | 1 | 0% | 1,406 | 545 | -61% | 0 | 0 | — |
case-13 | pass→pass | 4,915 | 1,833 | -63% | 1 | 1 | 0% | 870 | 644 | -26% | 0 | 0 | — |
case-14 | pass→pass | 8,803 | 2,330 | -74% | 1 | 1 | 0% | 1,330 | 807 | -39% | 0 | 0 | — |
case-15 | fail→pass | 6,813 | 1,758 | -74% | 1 | 1 | 0% | 1,184 | 652 | -45% | 0 | 0 | — |
case-16 | fail→pass | 7,552 | 1,923 | -75% | 1 | 1 | 0% | 1,188 | 697 | -41% | 0 | 0 | — |
case-17 | pass→pass | 7,133 | 1,881 | -74% | 1 | 1 | 0% | 1,342 | 634 | -53% | 0 | 0 | — |
case-18 | fail→pass | 12,256 | 1,779 | -85% | 1 | 1 | 0% | 2,250 | 656 | -71% | 0 | 0 | — |
case-19 | pass→pass | 12,868 | 8,005 | -38% | 1 | 1 | 0% | 2,066 | 1,746 | -15% | 0 | 0 | — |
case-20 | fail→pass | 15,692 | 5,755 | -63% | 1 | 1 | 0% | 2,975 | 1,415 | -52% | 0 | 0 | — |
case-21 | fail→pass | 5,989 | 1,715 | -71% | 1 | 1 | 0% | 1,164 | 667 | -43% | 0 | 0 | — |
case-22 | fail→pass | 8,345 | 2,015 | -76% | 1 | 1 | 0% | 1,463 | 727 | -50% | 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. The headline lift of +64 percentage points is the difference between those two pass rates over the 22 comparable cases.
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.