Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →MiCA compliance evidence and stablecoin risk scoring. Use when the user asks about stablecoin compliance, MiCA status, peg stability, or needs verifiable evidence for audit workflows.
.claude/skills/davepoon-feedoracle-compliance/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-04 | ✗→✓ | ▲ Improved | 11% | 0% |
| case-05 | ✗→✓ | ▲ Improved | -66% | 0% |
| case-07 | ✗→✓ | ▲ Improved | -32% | 0% |
| case-08 | ✗→✓ | ▲ Improved | -58% | 0% |
| case-12 | ✗→✓ | ▲ Improved | -57% | 0% |
You have access to FeedOracle's 27 MCP tools for compliance evidence in regulated tokenized markets. Every response is ES256K-signed and JWKS-verifiable.
MiCA compliance: mica_status for single token, mica_full_pack for full evidence, mica_market_overview for market-wide
Stablecoin risk: peg_deviation for current peg, peg_history for 30-day trend, compliance_preflight for PASS/WARN/BLOCK
Reserve and custody: reserve_quality for composition, custody_risk for custodian analysis
Audit trail: audit_verify to check chain integrity, audit_query to list decisions, audit_log to log (user-initiated only)
Natural language: ai_query routes to the right evidence API automatically
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 22,033 | 4,925 | -78% | 1 | 1 | 0% | 3,139 | 584 | -81% | 0 | 0 | — |
case-02 | fail→fail | 11,526 | 5,838 | -49% | 1 | 1 | 0% | 1,975 | 703 | -64% | 0 | 0 | — |
case-03 | fail→fail | 15,859 | 5,706 | -64% | 1 | 1 | 0% | 2,692 | 599 | -78% | 0 | 0 | — |
case-04 | fail→pass | 4,091 | 6,474 | +58% | 1 | 1 | 0% | 687 | 766 | +11% | 0 | 0 | — |
case-05 | fail→pass | 10,796 | 2,401 | -78% | 1 | 1 | 0% | 1,844 | 621 | -66% | 0 | 0 | — |
case-06 | fail→fail | 5,704 | 4,619 | -19% | 1 | 1 | 0% | 367 | 590 | +61% | 0 | 0 | — |
case-11 | fail→fail | 14,181 | 6,518 | -54% | 1 | 1 | 0% | 2,396 | 601 | -75% | 0 | 0 | — |
case-07 | fail→pass | 9,715 | 4,974 | -49% | 1 | 1 | 0% | 1,643 | 1,124 | -32% | 0 | 0 | — |
case-08 | fail→pass | 8,050 | 2,345 | -71% | 1 | 1 | 0% | 1,517 | 637 | -58% | 0 | 0 | — |
case-09 | pass→fail | 10,890 | 6,481 | -40% | 1 | 1 | 0% | 1,750 | 695 | -60% | 0 | 0 | — |
case-10 | pass→pass | 6,306 | 4,047 | -36% | 1 | 1 | 0% | 1,126 | 884 | -21% | 0 | 0 | — |
case-12 | fail→pass | 9,265 | 2,453 | -74% | 1 | 1 | 0% | 1,514 | 644 | -57% | 0 | 0 | — |
case-13 | pass→pass | 5,077 | 2,387 | -53% | 1 | 1 | 0% | 868 | 609 | -30% | 0 | 0 | — |
case-14 | fail→pass | 10,844 | 2,503 | -77% | 1 | 1 | 0% | 2,056 | 698 | -66% | 0 | 0 | — |
case-15 | fail→pass | 5,908 | 1,944 | -67% | 1 | 1 | 0% | 955 | 534 | -44% | 0 | 0 | — |
case-16 | fail→pass | 11,641 | 4,791 | -59% | 1 | 1 | 0% | 1,906 | 1,050 | -45% | 0 | 0 | — |
case-17 | fail→pass | 9,397 | 4,040 | -57% | 1 | 1 | 0% | 1,604 | 895 | -44% | 0 | 0 | — |
case-18 | fail→pass | 10,709 | 3,086 | -71% | 1 | 1 | 0% | 1,914 | 793 | -59% | 0 | 0 | — |
case-19 | fail→pass | 11,679 | 3,549 | -70% | 1 | 1 | 0% | 1,962 | 661 | -66% | 0 | 0 | — |
case-20 | pass→pass | 17,021 | 10,542 | -38% | 1 | 1 | 0% | 3,396 | 2,191 | -35% | 0 | 0 | — |
case-21 | pass→pass | 9,868 | 9,119 | -8% | 1 | 1 | 0% | 2,132 | 2,119 | -1% | 0 | 0 | — |
case-22 | pass→pass | 10,861 | 8,251 | -24% | 1 | 1 | 0% | 2,013 | 1,680 | -17% | 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 16 counted toward the lift figure. The other 6 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 16 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.