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

60%

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

57%Weight 40%Scale 1-5

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

The content is actionable with many executable examples and a useful troubleshooting section, but it is a monolithic single file with mild redundancy, an empty section, time-sensitive version info, and no progressive disclosure to supporting files. Validation feedback loops are absent despite the safety-critical context.

Suggestions

Split the per-mechanism workflows (PII/Presidio, LlamaGuard, fact-checking) into separate reference files and link to them from a concise overview, improving progressive disclosure.

Remove the empty 'Advanced topics' heading and relocate version numbers and star counts into a clearly labeled 'Versions / status' or deprecated section to stop time-sensitive noise from aging the skill.

Add explicit validation checkpoints (e.g. test rails against a known jailbreak/PII corpus, iterate on threshold tuning) so workflows include a feedback loop for the safety-critical operations.

Dimension	Reasoning	Score
Conciseness	The body is mostly code and reasonably efficient, but the redundant 'Safety mechanisms' list restates the workflows, the 'Advanced topics' heading is empty, and time-sensitive items ('v0.9.0+ (v0.12.0 expected)', '⭐ 4,300+') appear outside a deprecated/old-patterns section.	3 / 5
Actionability	Provides copy-paste-ready Colang configs and Python snippets across five workflows, with only minor gaps from undefined helper functions like toxicity_detector(), extract_facts(), and verify_facts().	4 / 5
Workflow Clarity	Each workflow shows a clear define-config-then-generate sequence and a troubleshooting section, but there are no validate->fix->retry checkpoints or feedback loops for the operations shown.	3 / 5
Progressive Disclosure	Section headers are clear and well-organized, but ~280 lines of workflows, common issues, hardware, and resources are all inlined in one file with no external references, exceeding the 50-line simple-skill threshold that would allow a higher score.	3 / 5
	Total	13 / 20 Passed

Description

63%Weight 40%Scale 1-5

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, enumerating concrete safety capabilities, but it omits an explicit 'when to use' trigger clause and relies on specialized jargon over natural user phrasing. It is in appropriate third-person voice.

Suggestions

Append an explicit trigger clause, e.g. 'Use when adding runtime safety checks, jailbreak/PII/toxicity filtering, or output validation to LLM applications.'

Soften jargon by pairing technical terms with natural synonyms users would actually say (e.g. 'guardrails', 'safety filters', 'content moderation').

Drop time-sensitive or environment-specific details like 'runs on T4 GPU' from the description; move them to the body's hardware section.

Dimension	Reasoning	Score
Specificity	Lists multiple concrete capabilities ('jailbreak detection, input/output validation, fact-checking, hallucination detection, PII filtering, toxicity detection') plus 'Colang 2.0 DSL' and 'T4 GPU', giving comprehensive coverage rather than abstract language.	5 / 5
Completeness	Clearly states what the skill does ('runtime safety framework... Features...') but lacks any explicit 'Use when...' or equivalent trigger guidance, capping completeness at 3 per the rubric guideline.	3 / 5
Trigger Term Quality	Contains relevant keywords like 'jailbreak detection' and 'PII filtering' but leans on technical jargon ('Colang 2.0 DSL', 'programmable rails', 'T4 GPU') and omits common synonyms or natural variations a user would say.	3 / 5
Distinctiveness Conflict Risk	Targets a clear NVIDIA/NeMo niche with distinct safety-rail triggers, with only minor overlap risk against adjacent moderation skills (LlamaGuard, Moderation API).	4 / 5
	Total	15 / 20 Passed

Validation

93%

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 — 15 / 16 Passed

Validation for skill structure

Criteria	Description	Result
frontmatter_unknown_keys	Unknown frontmatter key(s) found; consider removing or moving to metadata	Warning

	Total	15 / 16 Passed

Repository: synthetic-sciences/openscience
Path: backend/cli/skills/llm-tools/nemo-guardrails/SKILL.md
Commit: 7dabbf0

Reviewed: 2 days 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.