Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Shifts Layer 3 model selection to cost-optimized alternatives when economy mode is active.
.claude/skills/github-economy-mode/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-04 | ✗→✓ | ▲ Improved | -58% | 0% |
| case-06 | ✗→✓ | ▲ Improved | 802% | 0% |
| case-08 | ✗→✓ | ▲ Improved | -12% | 0% |
| case-09 | ✗→✓ | ▲ Improved | 53% | 0% |
| case-10 | ✗→✓ | ▲ Improved | 1% | 0% |
✅ THIS SKILL PRODUCES:
economyMode: true written to .squad/config.json when activated persistently💰 indicator when economy mode is active❌ THIS SKILL DOES NOT PRODUCE:
Economy mode shifts Layer 3 (Task-Aware Auto-Selection) to lower-cost alternatives. It does NOT override persistent config (defaultModel, agentModelOverrides) or per-agent charter preferences — those represent explicit user intent and always take priority.
Use this skill when the user wants to reduce costs across an entire session or permanently, without manually specifying models for each agent.
| Method | How | |--------|-----| | Session phrase | "use economy mode", "save costs", "go cheap", "reduce costs" | | Persistent config | "economyMode": true in .squad/config.json | | CLI flag | squad --economy |
Deactivation: "turn off economy mode", "disable economy mode", or remove economyMode from config.json.
When economy mode is active, Layer 3 auto-selection uses this table instead of the normal defaults:
| Task Output | Normal Mode | Economy Mode | |-------------|-------------|--------------| | Writing code (implementation, refactoring, bug fixes) | claude-sonnet-4.5 | gpt-5-mini | | Writing prompts or agent designs | claude-sonnet-4.5 | gpt-5-mini | | Docs, planning, triage, changelogs, mechanical ops | claude-haiku-4.5 | gpt-5-mini | | Architecture, code review, security audits | claude-opus-4.6 | claude-sonnet-4.5 | | Scribe / logger / mechanical file ops | claude-haiku-4.5 | gpt-5-mini |
Prefer gpt-5-mini for all economy-mode tasks where cost is the priority.
.squad/config.jsoneconomyMode: true — if present, activate economy mode for the sessionSession-only (no config change): "use economy mode", "save costs", "go cheap"
✅ Economy mode active — using cost-optimized models this session. (Layer 0 and Layer 2 preferences still apply)Persistent: "always use economy mode", "save economy mode"
economyMode: true to .squad/config.json (merge, don't overwrite other fields)✅ Economy mode saved — cost-optimized models will be used until disabled.💰 in spawn acknowledgment: 🔧 {Name} ({model} · 💰 economy) — {task}Trigger phrases: "turn off economy mode", "disable economy mode", "use normal models"
economyMode from .squad/config.json (if it was persisted)✅ Economy mode disabled — returning to standard model selection.After updating economy mode state and including the 💰 indicator in spawn acknowledgments, this skill is done. Do NOT:
.squad/config.json economy-related fields:
json{ "version": 1, "economyMode": true }
economyMode — when true, Layer 3 uses the economy table. Optional; absent = economy mode off.defaultModel and agentModelOverrides — Layer 0 always wins.defaultModel: "claude-opus-4.6", they want quality. Economy mode only affects Layer 3 auto-selection.config.json persist it.| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 9,130 | 3,536 | -61% | 1 | 1 | 0% | 1,038 | 1,675 | +61% | 0 | 0 | — |
case-02 | fail→fail | 7,657 | 3,056 | -60% | 1 | 1 | 0% | 1,008 | 1,647 | +63% | 0 | 0 | — |
case-03 | fail→fail | 5,099 | 2,705 | -47% | 1 | 1 | 0% | 627 | 1,560 | +149% | 0 | 0 | — |
case-04 | fail→pass | 31,615 | 4,693 | -85% | 1 | 1 | 0% | 4,335 | 1,838 | -58% | 0 | 0 | — |
case-05 | fail→fail | 26,965 | 8,420 | -69% | 1 | 1 | 0% | 5,207 | 2,818 | -46% | 0 | 0 | — |
case-06 | fail→pass | 5,661 | 4,856 | -14% | 1 | 1 | 0% | 226 | 2,038 | +802% | 0 | 0 | — |
case-07 | pass→pass | 9,072 | 6,129 | -32% | 1 | 1 | 0% | 1,439 | 2,246 | +56% | 0 | 0 | — |
case-08 | fail→pass | 13,988 | 3,943 | -72% | 1 | 1 | 0% | 2,053 | 1,816 | -12% | 0 | 0 | — |
case-09 | fail→pass | 11,643 | 6,095 | -48% | 1 | 1 | 0% | 1,257 | 1,920 | +53% | 0 | 0 | — |
case-10 | fail→pass | 11,400 | 3,907 | -66% | 1 | 1 | 0% | 1,748 | 1,770 | +1% | 0 | 0 | — |
case-11 | fail→pass | 10,956 | 3,594 | -67% | 1 | 1 | 0% | 1,792 | 1,741 | -3% | 0 | 0 | — |
case-12 | fail→pass | 4,609 | 2,710 | -41% | 1 | 1 | 0% | 527 | 1,526 | +190% | 0 | 0 | — |
case-13 | fail→pass | 18,203 | 4,615 | -75% | 1 | 1 | 0% | 2,741 | 1,952 | -29% | 0 | 0 | — |
case-14 | pass→pass | 4,628 | 4,185 | -10% | 1 | 1 | 0% | 624 | 1,880 | +201% | 0 | 0 | — |
case-15 | fail→pass | 20,204 | 4,170 | -79% | 1 | 1 | 0% | 3,284 | 1,900 | -42% | 0 | 0 | — |
case-16 | pass→pass | 13,044 | 6,366 | -51% | 1 | 1 | 0% | 1,888 | 2,243 | +19% | 0 | 0 | — |
case-17 | fail→pass | 10,118 | 3,047 | -70% | 1 | 1 | 0% | 1,406 | 1,596 | +14% | 0 | 0 | — |
case-18 | pass→pass | 4,105 | 3,394 | -17% | 1 | 1 | 0% | 466 | 1,648 | +254% | 0 | 0 | — |
case-19 | fail→pass | 12,697 | 5,996 | -53% | 1 | 1 | 0% | 1,771 | 2,198 | +24% | 0 | 0 | — |
case-20 | pass→pass | 13,918 | 7,374 | -47% | 1 | 1 | 0% | 2,146 | 2,338 | +9% | 0 | 0 | — |
case-21 | pass→pass | 9,246 | 5,465 | -41% | 1 | 1 | 0% | 1,316 | 2,096 | +59% | 0 | 0 | — |
case-22 | fail→pass | 13,591 | 14,105 | +4% | 1 | 1 | 0% | 1,933 | 2,066 | +7% | 0 | 0 | — |
case-23 | pass→pass | 24,059 | 7,737 | -68% | 1 | 1 | 0% | 2,751 | 2,562 | -7% | 0 | 0 | — |
case-24 | pass→pass | 10,351 | 5,051 | -51% | 1 | 1 | 0% | 1,467 | 1,903 | +30% | 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 23 counted toward the lift figure. The other 1 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 +50 percentage points is the difference between those two pass rates over the 23 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.