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.

Quality

58%

Does it follow best practices?

Run evals on this skill

Adds up to 20 points to the overall score

View guide

Securityby

Passed

No findings from the security scan

Fix and improve this skill with Tessl

tessl review fix ./backend/cli/skills/llm-tools/nemo-guardrails/SKILL.md

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 body is well-structured with concrete, mostly-executable Colang examples and clear issue-remediation guidance, but it is over-long for an overview, contains undefined placeholder functions in the custom-action examples, and its advanced references point to files that are not present in the bundle.

Suggestions

Provide the missing reference files (references/colang-guide.md, integrations.md, performance.md) or remove the dangling links so progressive disclosure navigation is not broken.

Complete the custom-action examples by defining or stub-importing helper functions (toxicity_detector, verify_facts, extract_facts) so the code is copy-paste executable.

Tighten the body by moving the five full worked workflows into a reference file, keeping only the quick-start example inline to reduce token overhead.

Dimension	Reasoning	Score
Conciseness	The body is action-oriented with minimal concept explanation, but ~290 lines with five full worked workflows plus repeated "Safety mechanisms"/"When to use" lists could be tightened, fitting "mostly efficient but could be tightened".	2 / 3
Actionability	Most Colang config blocks and pip/Python snippets are executable, but the custom-action examples call undefined functions (toxicity_detector, verify_facts, extract_facts, fact_check_action), which is incomplete guidance rather than fully copy-paste-ready code.	2 / 3
Workflow Clarity	Workflows are numbered and sequenced with code, but they are independent feature demos without explicit validate->fix->retry checkpoints; validation is only implicit inside the flows, matching "steps listed but checkpoints missing or implicit".	2 / 3
Progressive Disclosure	An "Advanced topics" section cleanly signals one-level-deep references (colang-guide.md, integrations.md, performance.md), but no references/ directory exists so the links are dangling, and substantial worked-example content remains inline in the body.	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, naming a clear niche and multiple concrete safety capabilities, but it omits any explicit "Use when..." trigger guidance, which caps completeness and weakens trigger-term quality.

Suggestions

Add a "Use when ..." clause with natural trigger phrases, e.g. "Use when adding runtime safety rails, jailbreak/PII/hallucination filtering, or when the user mentions guardrails for LLM apps."

Include common conversational trigger variations (e.g. "prompt injection", "safety filters", "content moderation") alongside the technical feature names.

Drop unverifiable marketing claims like "Production-ready, runs on T4 GPU" unless they serve as concrete triggers, to keep the description tight.

Dimension	Reasoning	Score
Specificity	Enumerates multiple concrete capabilities — "jailbreak detection, input/output validation, fact-checking, hallucination detection, PII filtering, toxicity detection" — plus the Colang 2.0 DSL mechanism, matching the "lists multiple specific concrete actions" anchor.	3 / 3
Completeness	The "what" is clearly stated ("NVIDIA's runtime safety framework... Features ...") but the "when" is entirely absent with no explicit trigger guidance, which the guidelines cap at 2.	2 / 3
Trigger Term Quality	Terms like "jailbreak detection", "PII filtering", and "hallucination detection" are natural, but there are no "Use when..." trigger phrases and common conversational variations are missing, so it lands at "some relevant keywords but missing common variations".	2 / 3
Distinctiveness Conflict Risk	The NeMo Guardrails / Colang 2.0 runtime-safety niche with specific guardrail features is distinct and unlikely to trigger for unrelated skills, matching the "clear niche with distinct triggers" anchor.	3 / 3
	Total	10 / 12 Passed

Validation

81%

Warnings & errors only

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

Validation — 13 / 16 Passed

Validation for skill structure

Criteria	Description	Result
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: synthetic-sciences/openscience
Path: backend/cli/skills/llm-tools/nemo-guardrails/SKILL.md
Commit: 52845c3

Reviewed: about 14 hours ago

Table of Contents

Discovery Implementation Validation

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.