Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Read loop-constraints.md at the start of every run and enforce every rule. This skill runs BEFORE triage or any action skill. Constraints are binding.
.claude/skills/cobusgreyling-loop-constraints/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-07 | ✗→✓ | ▲ Improved | -39% | 0% |
| case-08 | ✗→✓ | ▲ Improved | -55% | 0% |
| case-17 | ✗→✓ | ▲ Improved | -29% | 0% |
| case-19 | ✗→✓ | ▲ Improved | 6% | 0% |
| case-20 | ✗→✓ | ▲ Improved | -23% | 0% |
You are the guardrail. Before any other work begins, you MUST:
loop-constraints.md from the project root.loop-pause-all is active → exit immediately.Always begin with a one-line confirmation. Count each individual constraint/rule in loop-constraints.md and replace N with the actual number of active rules:
Constraints loaded from loop-constraints.md: N rules active.If no loop-constraints.md exists, say so and proceed with default safety rules from docs/safety.md.
loop-triage — constraints may override triage priority (e.g. "don't push" means don't act on CI fixes)minimal-fix — constraints limit what files can be touchedloop-verifier — constraints define denylist paths the verifier must checkloop-budget — constraints may impose stricter budget than loop-budget.mdIf loop-constraints.md is absent, enforce these minimums:
.env, .env.*, auth/, payments/, secrets/, credentials/| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 5,150 | 11,521 | +124% | 1 | 1 | 0% | 827 | 721 | -13% | 0 | 0 | — |
case-02 | fail→fail | 16,763 | 10,408 | -38% | 1 | 1 | 0% | 1,238 | 725 | -41% | 0 | 0 | — |
case-03 | fail→fail | 10,039 | 10,059 | +0% | 1 | 1 | 0% | 201 | 680 | +238% | 0 | 0 | — |
case-04 | pass→pass | 14,816 | 11,794 | -20% | 1 | 1 | 0% | 1,619 | 1,550 | -4% | 0 | 0 | — |
case-05 | pass→pass | 14,038 | 9,565 | -32% | 1 | 1 | 0% | 1,458 | 1,188 | -19% | 0 | 0 | — |
case-06 | pass→pass | 12,750 | 5,395 | -58% | 1 | 1 | 0% | 978 | 1,518 | +55% | 0 | 0 | — |
case-07 | fail→pass | 15,144 | 4,349 | -71% | 1 | 1 | 0% | 1,983 | 1,208 | -39% | 0 | 0 | — |
case-08 | fail→pass | 10,855 | 8,358 | -23% | 1 | 1 | 0% | 1,801 | 806 | -55% | 0 | 0 | — |
case-09 | pass→pass | 14,646 | 8,624 | -41% | 1 | 1 | 0% | 1,490 | 1,049 | -30% | 0 | 0 | — |
case-10 | pass→pass | 12,895 | 8,406 | -35% | 1 | 1 | 0% | 1,175 | 1,080 | -8% | 0 | 0 | — |
case-11 | pass→pass | 14,213 | 9,160 | -36% | 1 | 1 | 0% | 1,341 | 1,122 | -16% | 0 | 0 | — |
case-12 | pass→pass | 12,353 | 4,978 | -60% | 1 | 1 | 0% | 1,663 | 1,077 | -35% | 0 | 0 | — |
case-13 | pass→pass | 7,108 | 9,434 | +33% | 1 | 1 | 0% | 1,052 | 1,097 | +4% | 0 | 0 | — |
case-14 | pass→pass | 9,319 | 4,968 | -47% | 1 | 1 | 0% | 1,508 | 1,048 | -31% | 0 | 0 | — |
case-15 | pass→pass | 8,931 | 3,441 | -61% | 1 | 1 | 0% | 1,491 | 998 | -33% | 0 | 0 | — |
case-16 | pass→pass | 10,795 | 7,753 | -28% | 1 | 1 | 0% | 870 | 925 | +6% | 0 | 0 | — |
case-17 | fail→pass | 9,858 | 9,046 | -8% | 1 | 1 | 0% | 1,530 | 1,079 | -29% | 0 | 0 | — |
case-18 | pass→pass | 9,433 | 7,082 | -25% | 1 | 1 | 0% | 630 | 760 | +21% | 0 | 0 | — |
case-19 | fail→pass | 12,075 | 4,631 | -62% | 1 | 1 | 0% | 1,112 | 1,180 | +6% | 0 | 0 | — |
case-20 | fail→pass | 12,381 | 10,685 | -14% | 1 | 1 | 0% | 1,975 | 1,521 | -23% | 0 | 0 | — |
case-21 | fail→pass | 9,374 | 3,821 | -59% | 1 | 1 | 0% | 1,472 | 1,055 | -28% | 0 | 0 | — |
case-22 | pass→pass | 9,644 | 9,489 | -2% | 1 | 1 | 0% | 698 | 1,139 | +63% | 0 | 0 | — |
case-23 | pass→pass | 12,781 | 4,784 | -63% | 1 | 1 | 0% | 1,239 | 1,149 | -7% | 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. 23 cases were attempted, and 20 counted toward the lift figure. The other 3 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 +26 percentage points is the difference between those two pass rates over the 20 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.
| Model | Method | Date | Lift |
|---|---|---|---|
| gemini-3.6-flash | verified | 8/11/2026 | +23% |
Other measured skills in the registry, with their headline benchmark lift.