CtrlK
BlogDocsLog inGet started
Tessl Logo

nemo-guardrails

NVIDIA's runtime safety framework for LLM applications. Features jailbreak detection, input/output validation, fact-checking, hallucination detection, PII filtering, toxicity detection. Uses Colang 2.0 DSL for programmable rails. Production-ready, runs on T4 GPU.

52

Quality

58%

Does it follow best practices?

Run evals on this skill

Adds up to 20 points to the overall score

View guide

SecuritybySnyk

Passed

No findings from the security scan

Fix and improve this skill with Tessl

tessl review fix ./skills/safety-alignment/nemo-guardrails/SKILL.md
SKILL.md
Quality
Evals
Security

Quality

Content

50%

Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.

The content is concrete and example-rich but uniformly lands at the mid level: code examples contain non-executable placeholders, workflows lack validation feedback loops, and the advanced-topic references point to files that do not exist. Tightening, real bundle files, and executable helpers would raise every dimension.

Suggestions

Make code examples fully executable by defining or sourcing the placeholder helpers (toxicity_detector, extract_facts, verify_facts, detect/mask pii) or by noting the integration they come from, so examples are copy-paste ready.

Actually provide references/colang-guide.md, references/integrations.md, and references/performance.md (or remove the dead links) and move the five inline workflows plus common-issues section into them to slim the main body.

Add explicit validation/feedback-loop steps to workflows (e.g., confirm a check returned a score, retry or fall back on failure) so multi-step processes have clear checkpoints.

DimensionReasoningScore

Conciseness

The body runs ~290 lines with five full workflow code blocks plus common-issues, hardware, and resource sections that could be trimmed or moved to references. Matches anchor 2 (mostly efficient but could be tightened); not 3 due to verbosity, not 1 because it avoids explaining concepts Claude already knows.

2 / 3

Actionability

Mostly concrete Python/Colang code, but workflows invoke undefined placeholders like "toxicity_detector(user_message)", "extract_facts", "verify_facts", and "detect pii"/"mask pii entities". Matches anchor 2 (some concrete guidance but incomplete/pseudocode); not 3 because several examples are not copy-paste executable.

2 / 3

Workflow Clarity

Workflows are standalone code recipes rather than sequenced multi-step processes, and aside from Workflow 2's "if not $safe ... stop" there are no validation checkpoints or feedback loops. Matches anchor 2 (sequence present but checkpoints missing/implicit); not 3 due to absent explicit validation feedback loops.

2 / 3

Progressive Disclosure

"See [references/colang-guide.md]" and two sibling links are clearly signaled and one-level-deep, but no references/ directory exists so the links are dead, and substantial content (5 workflows, common issues) is inline rather than split out. Matches anchor 2 (some structure, content that should be separate is inline); not 3 due to broken references and inline bulk.

2 / 3

Total

8

/

12

Passed

Description

67%

Based on the skill's description, can an agent find and select it at the right time? Clear, specific descriptions lead to better discovery.

The description is specific and distinctive with concrete safety capabilities, but lacks an explicit "Use when..." trigger clause and leans on technical jargon over natural user keywords. Completeness and trigger-term quality are the weak points.

Suggestions

Add an explicit 'Use when...' trigger clause naming common user phrasings, e.g. 'Use when adding runtime safety guardrails to an LLM app, or when the user mentions jailbreaks, prompt injection, content moderation, or PII/toxicity filtering.'

Pair jargon terms ('Colang 2.0 DSL', 'programmable rails', 'T4 GPU') with natural keywords like 'safety guardrails', 'content moderation', and 'prompt injection' so trigger matching hits everyday phrasings.

DimensionReasoningScore

Specificity

Quotes "jailbreak detection, input/output validation, fact-checking, hallucination detection, PII filtering, toxicity detection" — six concrete capabilities listed, matching anchor 3 (multiple specific concrete actions). Not anchor 2 because the enumeration is comprehensive rather than "some actions".

3 / 3

Completeness

It answers "what" in detail but contains no "Use when..." trigger clause, and the guideline caps completeness at 2 when explicit trigger guidance is missing. Not 1 because "what" is strong; not 3 because "when" is absent.

2 / 3

Trigger Term Quality

Quotes "jailbreak detection", "PII filtering", "toxicity detection" are natural terms, but "Colang 2.0 DSL", "programmable rails", and "T4 GPU" are jargon and common variations like "content moderation" or "prompt injection" are absent. Matches anchor 2 (some relevant keywords, missing common variations); not 3 due to the jargon weight and gaps.

2 / 3

Distinctiveness Conflict Risk

Quotes "NVIDIA's runtime safety framework for LLM applications" plus "Colang 2.0 DSL" — a clear, distinctive niche unlikely to trigger the wrong skill. Matches anchor 3; not 2 because the niche is specific rather than generic.

3 / 3

Total

10

/

12

Passed

Validation

81%

Checks the skill against the spec for correct structure and formatting. All validation checks must pass before discovery and implementation can be scored.

Validation13 / 16 Passed

Validation for skill structure

CriteriaDescriptionResult

frontmatter_unknown_keys

Unknown frontmatter key(s) found; consider removing or moving to metadata

Warning

relative_links

Relative link issues: 3 missing

Warning

referenced_paths_exist

Referenced path issues: 6 missing

Warning

Total

13

/

16

Passed

Repository
OpenLAIR/dr-claw
Reviewed

Table of Contents

Is this your skill?

If you maintain this skill, you can claim it as your own. Once claimed, you can manage eval scenarios, bundle related skills, attach documentation or rules, and ensure cross-agent compatibility.