Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →3-tier budget enforcement for AI agent workspaces. Visibility dashboards (always on), soft alerts at 80% threshold, hard ceilings at 100% that auto-pause agents. Tracks per-agent, per-task, and per-project costs in both tokens and dollars. Triggers on: "budget", "cost", "spending", "token usage", "billing"
.claude/skills/miosa-osa-budget/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | 37% | 0% |
| case-02 | ✗→✓ | ▲ Improved | -6% | 0% |
| case-04 | ✗→✓ | ▲ Improved | -25% | 0% |
| case-06 | ✗→✓ | ▲ Improved | 0% | 0% |
| case-07 | ✗→✓ | ▲ Improved | -6% | 0% |
> 3-tier budget enforcement — visibility, alerts, hard ceilings.
Prevent runaway AI costs in multi-agent workspaces. Every token consumed is tracked at three levels: per-agent, per-task, and per-project. Three enforcement tiers ensure costs never surprise you: Tier 1 (visibility) shows real-time dashboards, Tier 2 (soft alerts) warns at 80%, Tier 3 (hard ceilings) auto-pauses agents at 100%.
bash# View budget dashboard /budget # View specific agent's budget /budget --agent researcher # Set project budget /budget set --project 500000 --dollars 25.00 # Set per-agent budget /budget set --agent researcher 50000 # View burn rate /budget burn-rate # Export cost report /budget report --format csv --period 2026-03 # Reset budget counters (new billing period) /budget reset --confirm
| Subcommand | Description | |------------|-------------| | (default) | Show budget dashboard | | set | Set budget limits | | burn-rate | Show token consumption velocity | | report | Export detailed cost report | | reset | Reset counters for new period |
| Flag | Type | Default | Description | |------|------|---------|-------------| | --agent | string | all | Filter to specific agent | | --project | int | — | Set project-level token budget | | --dollars | float | — | Set dollar budget (converted via model pricing) | | --format | enum | terminal | terminal, csv, json, markdown | | --period | string | current | Billing period (YYYY-MM or date range) | | --confirm | flag | false | Required for destructive operations (reset) | | --alert-threshold | float | 0.8 | Tier 2 soft alert threshold (0.0–1.0) | | --hard-ceiling | float | 1.0 | Tier 3 hard ceiling (0.0–1.0+, >1.0 allows overflow) |
/board, /heartbeat, and dedicated dashboard. No action taken.budget_warning event.budget_exceeded message to orchestrator. Agent cannot resume until budget is increased or reset.markdown## Budget Dashboard — 2026-03-20 ### Project Total | Metric | Value | |--------|-------| | Token budget | 500,000 | | Tokens used | 167,300 (33%) | | Dollar budget | $25.00 | | Dollars spent | $8.42 (34%) | | Burn rate | 12,400 tokens/hr | | Time to ceiling | ~27 hours at current rate | ### Per-Agent Breakdown | Agent | Budget | Used | % | $/hr | Status | |-------|--------|------|---|------|--------| | researcher | 50,000 | 23,400 | 47% | $1.20 | active | | writer | 30,000 | 11,200 | 37% | $0.85 | active | | analyst | 50,000 | 45,100 | 90% | $2.10 | ⚠ ALERT | | coder | 100,000 | 0 | 0% | $0.00 | idle | | unallocated | 270,000 | — | — | — | — | ### Top Tasks by Cost | Task | Agent | Tokens | Cost | |------|-------|--------|------| | Revenue analysis | analyst | 38,200 | $4.12 | | Competitor scan | researcher | 18,400 | $1.85 | | Pitch draft v1 | writer | 9,800 | $0.98 | ### Alerts - ⚠ analyst: 90% of budget consumed (soft alert threshold: 80%)
/heartbeat — Agent status for pause/resume/inbox — Alert delivery| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→pass | 12,262 | 12,810 | +4% | 1 | 1 | 0% | 2,375 | 3,246 | +37% | 0 | 0 | — |
case-02 | fail→pass | 12,417 | 5,304 | -57% | 1 | 1 | 0% | 2,159 | 2,020 | -6% | 0 | 0 | — |
case-03 | fail→fail | 8,159 | 8,904 | +9% | 1 | 1 | 0% | 1,189 | 3,069 | +158% | 0 | 0 | — |
case-04 | fail→pass | 15,869 | 5,106 | -68% | 1 | 1 | 0% | 2,674 | 2,018 | -25% | 0 | 0 | — |
case-05 | fail→fail | 23,607 | 2,068 | -91% | 1 | 1 | 0% | 3,296 | 1,606 | -51% | 0 | 0 | — |
case-06 | fail→pass | 12,873 | 4,570 | -64% | 1 | 1 | 0% | 1,877 | 1,872 | -0% | 0 | 0 | — |
case-07 | fail→pass | 16,238 | 8,232 | -49% | 1 | 1 | 0% | 3,275 | 3,091 | -6% | 0 | 0 | — |
case-08 | fail→pass | 28,148 | 2,761 | -90% | 1 | 1 | 0% | 2,019 | 1,639 | -19% | 0 | 0 | — |
case-09 | pass→pass | 27,592 | 3,064 | -89% | 1 | 1 | 0% | 4,515 | 1,761 | -61% | 0 | 0 | — |
case-10 | fail→fail | 12,960 | 2,773 | -79% | 1 | 1 | 0% | 2,028 | 1,683 | -17% | 0 | 0 | — |
case-11 | pass→pass | 8,572 | 7,713 | -10% | 1 | 1 | 0% | 1,641 | 2,590 | +58% | 0 | 0 | — |
case-12 | pass→pass | 6,798 | 3,356 | -51% | 1 | 1 | 0% | 1,360 | 1,978 | +45% | 0 | 0 | — |
case-13 | fail→pass | 8,175 | 4,230 | -48% | 1 | 1 | 0% | 1,150 | 1,959 | +70% | 0 | 0 | — |
case-14 | fail→pass | 6,565 | 2,670 | -59% | 1 | 1 | 0% | 951 | 1,727 | +82% | 0 | 0 | — |
case-15 | fail→pass | 13,656 | 3,544 | -74% | 1 | 1 | 0% | 2,030 | 1,689 | -17% | 0 | 0 | — |
case-16 | pass→pass | 13,744 | 12,464 | -9% | 1 | 1 | 0% | 2,109 | 3,324 | +58% | 0 | 0 | — |
case-17 | pass→pass | 14,463 | 1,722 | -88% | 1 | 1 | 0% | 2,572 | 1,624 | -37% | 0 | 0 | — |
case-18 | fail→pass | 17,023 | 5,274 | -69% | 1 | 1 | 0% | 3,468 | 2,375 | -32% | 0 | 0 | — |
case-19 | pass→pass | 5,466 | 1,585 | -71% | 1 | 1 | 0% | 749 | 1,513 | +102% | 0 | 0 | — |
case-20 | pass→pass | 13,023 | 13,367 | +3% | 1 | 1 | 0% | 2,604 | 4,008 | +54% | 0 | 0 | — |
case-21 | pass→pass | 19,239 | 16,784 | -13% | 1 | 1 | 0% | 2,908 | 3,772 | +30% | 0 | 0 | — |
case-22 | pass→pass | 15,063 | 15,190 | +1% | 1 | 1 | 0% | 2,392 | 3,733 | +56% | 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 21 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 +45 percentage points is the difference between those two pass rates over the 21 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.