Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Enterprise-review-grade threat model from `harness threat-model <path>`. Categorizes MCP-surface threats; emits `worst: 'clean'|'low'|'medium'|'high'` + per-threat findings. Pure-read.
.claude/skills/ruvnet-harness-threat-model/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-10 | ✗→✓ | ▲ Improved | -66% | 0% |
| case-04 | ✗→✓ | ▲ Improved | -26% | 0% |
| case-05 | ✗→✓ | ▲ Improved | -61% | 0% |
| case-06 | ✗→✓ | ▲ Improved | -62% | 0% |
| case-07 | ✗→✓ | ▲ Improved | -30% | 0% |
The companion to harness-mcp-scan for enterprise security reviews. Where mcp-scan is a per-server static lint, threat-model produces a categorized report suitable for sharing with an InfoSec team.
Implementation: scripts/threat-model.mjs.
harness binary (metaharness@~0.3.0, resolved from alocal install or the one-time ~/.ruflo/metaharness-cache-<pin> cache — never @latest): harness threat-model <path> --json.
{ worst, findings[] }.--fail-on <severity>: exit 1 when worst >= fail-on. Default high.| Severity | Rank | |---|---:| | clean | 0 | | low | 1 | | medium | 2 | | high | 3 |
packet sent to security.
oia-audit backgroundworker (ADR-150 Phase 2) to detect MCP-surface drift.
Same pattern as the other skills: when harness is absent, emit { degraded: true } and exit 0. ADR-150 architectural constraint.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-10 | fail→pass | 11,631 | 1,877 | -84% | 1 | 1 | 0% | 1,850 | 635 | -66% | 0 | 0 | — |
case-01 | fail→fail | 19,367 | 10,847 | -44% | 1 | 1 | 0% | 2,481 | 1,659 | -33% | 0 | 0 | — |
case-02 | fail→fail | 6,682 | 8,145 | +22% | 1 | 1 | 0% | 712 | 1,188 | +67% | 0 | 0 | — |
case-03 | fail→fail | 9,728 | 11,816 | +21% | 1 | 1 | 0% | 1,808 | 2,813 | +56% | 0 | 0 | — |
case-04 | fail→pass | 21,236 | 8,952 | -58% | 1 | 1 | 0% | 1,087 | 805 | -26% | 0 | 0 | — |
case-05 | fail→pass | 9,102 | 1,759 | -81% | 1 | 1 | 0% | 1,583 | 624 | -61% | 0 | 0 | — |
case-06 | fail→pass | 9,511 | 2,161 | -77% | 1 | 1 | 0% | 1,878 | 708 | -62% | 0 | 0 | — |
case-07 | fail→pass | 6,240 | 5,323 | -15% | 1 | 1 | 0% | 1,107 | 770 | -30% | 0 | 0 | — |
case-08 | pass→pass | 6,754 | 3,065 | -55% | 1 | 1 | 0% | 1,473 | 808 | -45% | 0 | 0 | — |
case-09 | pass→pass | 8,461 | 3,079 | -64% | 1 | 1 | 0% | 1,708 | 972 | -43% | 0 | 0 | — |
case-11 | pass→pass | 6,652 | 1,091 | -84% | 1 | 1 | 0% | 1,095 | 478 | -56% | 0 | 0 | — |
case-12 | pass→pass | 10,118 | 3,141 | -69% | 1 | 1 | 0% | 1,899 | 527 | -72% | 0 | 0 | — |
case-13 | pass→pass | 9,047 | 1,308 | -86% | 1 | 1 | 0% | 1,687 | 443 | -74% | 0 | 0 | — |
case-14 | pass→pass | 9,229 | 1,144 | -88% | 1 | 1 | 0% | 1,615 | 497 | -69% | 0 | 0 | — |
case-15 | pass→pass | 5,250 | 1,966 | -63% | 1 | 1 | 0% | 886 | 707 | -20% | 0 | 0 | — |
case-16 | fail→pass | 8,082 | 1,663 | -79% | 1 | 1 | 0% | 1,283 | 600 | -53% | 0 | 0 | — |
case-17 | fail→pass | 9,064 | 1,698 | -81% | 1 | 1 | 0% | 1,492 | 610 | -59% | 0 | 0 | — |
case-18 | fail→pass | 3,826 | 2,707 | -29% | 1 | 1 | 0% | 569 | 574 | +1% | 0 | 0 | — |
case-19 | fail→pass | 6,800 | 1,297 | -81% | 1 | 1 | 0% | 1,236 | 497 | -60% | 0 | 0 | — |
case-20 | pass→pass | 12,675 | 5,524 | -56% | 1 | 1 | 0% | 2,749 | 1,354 | -51% | 0 | 0 | — |
case-21 | pass→pass | 6,168 | 7,688 | +25% | 1 | 1 | 0% | 1,228 | 1,758 | +43% | 0 | 0 | — |
case-22 | pass→pass | 9,759 | 6,082 | -38% | 1 | 1 | 0% | 1,998 | 1,608 | -20% | 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. The headline lift of +41 percentage points is the difference between those two pass rates over the 22 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.