Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →NVIDIA NeMo Guardrails — programmable guardrails for LLM applications. Colang-based dialog management, topical rails (fact-checking, moderation), safety rails, and security rails for production AI.
.claude/skills/mkurman-nvidia-nemo-guardrails/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-11 | ✗→✓ | ▲ Improved | -46% | 0% |
| case-17 | ✓→✓ | ＝ Same ✓ | 48% | 0% |
| case-02 | ✓→✓ | ＝ Same ✓ | 41% | 0% |
| case-03 | ✓→✓ | ＝ Same ✓ | -18% | 0% |
| case-04 | ✓→✓ | ＝ Same ✓ | -18% | 0% |
NVIDIA NeMo Guardrails provides programmable guardrails for LLM applications. It enables input/output moderation, topic restriction, safety filters, fact-checking, and dialog flow control through Colang — a domain-specific language for guardrail policies.
bashuv pip install nemoguardrails
pythonfrom nemoguardrails import RailsConfig, LLMRails config = RailsConfig.from_path("config") rails = LLMRails(config) response = rails.generate(messages=[{"role": "user", "content": "How do I hack a system?"}]) print(response["content"]) # Blocked or safe response
yaml# config/config.yml rails: input: flows: - self check input output: flows: - self check output # config/prompts.yml define user said inappropriate "I want to hack" define bot refuse to respond "I cannot help with that request." define flow user said inappropriate bot refuse to respond
pythonfrom nemoguardrails import LLMRails rails = LLMRails(config) rails.register_topic("politics", danger_level=3) rails.register_topic("medical_advice", danger_level=2) response = rails.generate("What is the best treatment for covid?") # Guardrails can restrict to general info or block entirely
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-17 | pass→pass | 2,910 | 2,522 | -13% | 1 | 1 | 0% | 518 | 766 | +48% | 0 | 0 | — |
case-01 | fail→fail | 14,630 | 8,781 | -40% | 1 | 1 | 0% | 2,600 | 1,959 | -25% | 0 | 0 | — |
case-02 | pass→pass | 2,982 | 1,749 | -41% | 1 | 1 | 0% | 480 | 675 | +41% | 0 | 0 | — |
case-03 | pass→pass | 5,932 | 3,134 | -47% | 1 | 1 | 0% | 1,120 | 913 | -18% | 0 | 0 | — |
case-04 | pass→pass | 9,343 | 5,910 | -37% | 1 | 1 | 0% | 1,670 | 1,372 | -18% | 0 | 0 | — |
case-05 | pass→pass | 10,470 | 3,635 | -65% | 1 | 1 | 0% | 1,304 | 989 | -24% | 0 | 0 | — |
case-06 | pass→pass | 15,843 | 3,592 | -77% | 1 | 1 | 0% | 1,218 | 970 | -20% | 0 | 0 | — |
case-07 | pass→pass | 5,922 | 2,717 | -54% | 1 | 1 | 0% | 1,017 | 792 | -22% | 0 | 0 | — |
case-08 | fail→fail | 13,820 | 6,846 | -50% | 1 | 1 | 0% | 2,290 | 1,639 | -28% | 0 | 0 | — |
case-09 | pass→pass | 12,927 | 6,619 | -49% | 1 | 1 | 0% | 2,171 | 1,567 | -28% | 0 | 0 | — |
case-10 | fail→fail | 9,318 | 5,119 | -45% | 1 | 1 | 0% | 1,626 | 1,307 | -20% | 0 | 0 | — |
case-11 | fail→pass | 9,528 | 2,838 | -70% | 1 | 1 | 0% | 1,572 | 856 | -46% | 0 | 0 | — |
case-12 | pass→pass | 16,000 | 3,209 | -80% | 1 | 1 | 0% | 2,591 | 1,026 | -60% | 0 | 0 | — |
case-18 | pass→pass | 8,571 | 4,150 | -52% | 1 | 1 | 0% | 1,638 | 1,079 | -34% | 0 | 0 | — |
case-13 | pass→pass | 8,174 | 2,814 | -66% | 1 | 1 | 0% | 1,354 | 818 | -40% | 0 | 0 | — |
case-14 | pass→pass | 6,611 | 6,861 | +4% | 1 | 1 | 0% | 1,092 | 1,589 | +46% | 0 | 0 | — |
case-15 | pass→pass | 4,313 | 3,889 | -10% | 1 | 1 | 0% | 525 | 1,025 | +95% | 0 | 0 | — |
case-16 | pass→pass | 8,043 | 4,683 | -42% | 1 | 1 | 0% | 1,462 | 1,238 | -15% | 0 | 0 | — |
case-19 | fail→fail | 9,188 | 5,235 | -43% | 1 | 1 | 0% | 1,614 | 1,246 | -23% | 0 | 0 | — |
case-20 | pass→pass | 8,183 | 5,974 | -27% | 1 | 1 | 0% | 1,560 | 1,526 | -2% | 0 | 0 | — |
case-21 | fail→fail | 13,522 | 8,644 | -36% | 1 | 1 | 0% | 2,536 | 2,113 | -17% | 0 | 0 | — |
case-22 | pass→pass | 11,975 | 10,709 | -11% | 1 | 1 | 0% | 1,951 | 2,305 | +18% | 0 | 0 | — |
case-23 | pass→pass | 13,564 | 8,454 | -38% | 1 | 1 | 0% | 2,403 | 1,870 | -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. 23 cases were attempted. The headline lift of +4 percentage points is the difference between those two pass rates over the 23 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.