Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Check token budget and run-log spend before and after a loop run. Enforces early exit when over budget or when there is no actionable work.
.claude/skills/cobusgreyling-loop-budget/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-04 | ✗→✓ | ▲ Improved | -26% | 0% |
| case-05 | ✗→✓ | ▲ Improved | 11% | 0% |
| case-06 | ✗→✓ | ▲ Improved | 27% | 0% |
| case-07 | ✗→✓ | ▲ Improved | -1% | 0% |
| case-10 | ✗→✓ | ▲ Improved | -51% | 0% |
Run at the start and end of every loop iteration.
loop-budget.md for daily caps and kill-switch flags.loop-run-log.md (last 24h).tokens_estimate for the active pattern today.STATE.md, yield to the budget-negotiator skill (if installed). Otherwise, if spend ≥ 100% or loop-pause-all is set → exit immediately with a one-line note in STATE.md.Append one JSON object to loop-run-log.md:
json{ "run_id": "<ISO8601>", "pattern": "<pattern-id>", "duration_s": <number>, "items_found": <number>, "actions_taken": <number>, "escalations": <number>, "tokens_estimate": <number>, "outcome": "no-op | report-only | fix-proposed | escalated" }
max sub-agent spawns/run from loop-budget.md.loop-budget.md under Alerts This Period.| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 5,232 | 2,993 | -43% | 1 | 1 | 0% | 946 | 947 | +0% | 0 | 0 | — |
case-02 | fail→fail | 11,154 | 3,382 | -70% | 1 | 1 | 0% | 1,782 | 757 | -58% | 0 | 0 | — |
case-03 | fail→fail | 7,517 | 4,215 | -44% | 1 | 1 | 0% | 1,007 | 779 | -23% | 0 | 0 | — |
case-04 | fail→pass | 9,429 | 3,560 | -62% | 1 | 1 | 0% | 1,631 | 1,199 | -26% | 0 | 0 | — |
case-05 | fail→pass | 5,698 | 3,072 | -46% | 1 | 1 | 0% | 935 | 1,034 | +11% | 0 | 0 | — |
case-06 | fail→pass | 4,529 | 2,681 | -41% | 1 | 1 | 0% | 777 | 985 | +27% | 0 | 0 | — |
case-07 | fail→pass | 5,324 | 2,247 | -58% | 1 | 1 | 0% | 835 | 823 | -1% | 0 | 0 | — |
case-08 | pass→pass | 12,055 | 3,089 | -74% | 1 | 1 | 0% | 1,939 | 878 | -55% | 0 | 0 | — |
case-09 | fail→fail | 9,588 | 2,055 | -79% | 1 | 1 | 0% | 1,676 | 708 | -58% | 0 | 0 | — |
case-10 | fail→pass | 9,138 | 1,821 | -80% | 1 | 1 | 0% | 1,605 | 780 | -51% | 0 | 0 | — |
case-11 | pass→fail | 5,388 | 1,928 | -64% | 1 | 1 | 0% | 982 | 775 | -21% | 0 | 0 | — |
case-12 | pass→pass | 6,826 | 1,715 | -75% | 1 | 1 | 0% | 1,309 | 659 | -50% | 0 | 0 | — |
case-13 | fail→pass | 10,859 | 2,007 | -82% | 1 | 1 | 0% | 1,833 | 772 | -58% | 0 | 0 | — |
case-14 | pass→pass | 6,813 | 1,950 | -71% | 1 | 1 | 0% | 1,063 | 760 | -29% | 0 | 0 | — |
case-15 | pass→pass | 8,126 | 2,737 | -66% | 1 | 1 | 0% | 1,286 | 933 | -27% | 0 | 0 | — |
case-16 | fail→fail | 8,027 | 1,943 | -76% | 1 | 1 | 0% | 1,332 | 732 | -45% | 0 | 0 | — |
case-17 | fail→fail | 10,743 | 2,058 | -81% | 1 | 1 | 0% | 2,071 | 749 | -64% | 0 | 0 | — |
case-18 | fail→pass | 15,093 | 3,415 | -77% | 1 | 1 | 0% | 2,622 | 1,007 | -62% | 0 | 0 | — |
case-19 | fail→fail | 8,467 | 1,512 | -82% | 1 | 1 | 0% | 1,440 | 729 | -49% | 0 | 0 | — |
case-20 | pass→pass | 15,721 | 13,357 | -15% | 1 | 1 | 0% | 2,710 | 2,682 | -1% | 0 | 0 | — |
case-21 | pass→pass | 4,935 | 4,378 | -11% | 1 | 1 | 0% | 980 | 1,249 | +27% | 0 | 0 | — |
case-22 | pass→pass | 3,655 | 2,952 | -19% | 1 | 1 | 0% | 862 | 1,053 | +22% | 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 20 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 +27 percentage points is the difference between those two pass rates over the 20 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.