Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Create a governed session override for a policy that would otherwise deny — mandatory justification, server-clamped TTL, blocked for critical-risk policies
.claude/skills/hashgraph-online-create-override/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-04 | ✗→✓ | ▲ Improved | 197% | 0% |
| case-06 | ✗→✓ | ▲ Improved | 34% | 0% |
| case-10 | ✗→✓ | ▲ Improved | 389% | 0% |
| case-09 | ✓→✗ | ▼ Worse | -33% | 0% |
| case-15 | ✓→✗ | ▼ Worse | -1% | 0% |
Use this skill when a user has been blocked by a non-critical policy and wants to bypass it for the rest of their session with a documented justification.
Before calling this skill, the user MUST provide:
policy_id — the ID of the policy to override (typically surfaced by explain-decision)policy_type — "static" (system / dynamic registry policy) or "dynamic"override_reason — free-text justification (1-500 chars). This is mandatory and recorded in the audit trail.Optional:
tool_signature — restrict the override to a specific tool name (recommended; narrower scope)ttl_seconds — requested duration. Server clamps to [60, 86400]; default 3600 (1h).Call the create_override MCP tool. The platform will reject the request (HTTP 403) when:
allow_override: falseSurface the response's id, expires_at, and any clamped flag (set when the requested TTL was clamped down). Tell the user: "Override <id> active until <expires_at> — the next attempt at the blocked tool will succeed within that window."
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-18 | fail→fail | 24,520 | 13,191 | -46% | 1 | 1 | 0% | 1,559 | 853 | -45% | 0 | 0 | — |
case-19 | fail→fail | 5,226 | 15,544 | +197% | 1 | 1 | 0% | 468 | 539 | +15% | 0 | 0 | — |
case-01 | fail→fail | 13,471 | 18,344 | +36% | 1 | 1 | 0% | 695 | 650 | -6% | 0 | 0 | — |
case-02 | fail→fail | 15,698 | 64,601 | +312% | 1 | 1 | 0% | 1,118 | 3,290 | +194% | 0 | 0 | — |
case-03 | fail→fail | 17,296 | 10,719 | -38% | 1 | 1 | 0% | 1,700 | 585 | -66% | 0 | 0 | — |
case-04 | fail→pass | 10,314 | 43,031 | +317% | 1 | 1 | 0% | 840 | 2,491 | +197% | 0 | 0 | — |
case-05 | fail→fail | 9,738 | 16,290 | +67% | 1 | 1 | 0% | 848 | 706 | -17% | 0 | 0 | — |
case-06 | fail→pass | 17,175 | 23,606 | +37% | 1 | 1 | 0% | 1,003 | 1,340 | +34% | 0 | 0 | — |
case-07 | fail→fail | 11,766 | 23,621 | +101% | 1 | 1 | 0% | 787 | 1,000 | +27% | 0 | 0 | — |
case-08 | pass→pass | 20,921 | 14,920 | -29% | 1 | 1 | 0% | 1,928 | 1,403 | -27% | 0 | 0 | — |
case-09 | pass→fail | 12,995 | 10,442 | -20% | 1 | 1 | 0% | 1,059 | 706 | -33% | 0 | 0 | — |
case-10 | fail→pass | 14,676 | 33,770 | +130% | 1 | 1 | 0% | 683 | 3,342 | +389% | 0 | 0 | — |
case-11 | fail→fail | 9,057 | 16,933 | +87% | 1 | 1 | 0% | 1,386 | 1,321 | -5% | 0 | 0 | — |
case-17 | fail→fail | 19,639 | 12,054 | -39% | 1 | 1 | 0% | 1,024 | 884 | -14% | 0 | 0 | — |
case-12 | fail→fail | 15,946 | 16,372 | +3% | 1 | 1 | 0% | 2,009 | 725 | -64% | 0 | 0 | — |
case-13 | fail→fail | 16,681 | 5,358 | -68% | 1 | 1 | 0% | 416 | 618 | +49% | 0 | 0 | — |
case-14 | fail→fail | 21,475 | 18,415 | -14% | 1 | 1 | 0% | 1,160 | 608 | -48% | 0 | 0 | — |
case-15 | pass→fail | 12,276 | 28,609 | +133% | 1 | 1 | 0% | 2,024 | 2,004 | -1% | 0 | 0 | — |
case-16 | fail→fail | 20,425 | 22,597 | +11% | 1 | 1 | 0% | 2,212 | 824 | -63% | 0 | 0 | — |
case-20 | pass→pass | 15,519 | 19,565 | +26% | 1 | 1 | 0% | 2,096 | 2,351 | +12% | 0 | 0 | — |
case-21 | pass→fail | 12,081 | 6,178 | -49% | 1 | 1 | 0% | 891 | 1,369 | +54% | 0 | 0 | — |
case-22 | pass→fail | 10,412 | 26,310 | +153% | 1 | 1 | 0% | 1,026 | 2,324 | +127% | 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 6 counted toward the lift figure. The other 16 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 -5 percentage points is the difference between those two pass rates over the 6 comparable cases. 4 cases got worse with the skill loaded, and they are 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.