nemotron-policy-generator

Generates BYO custom safety policies for NVIDIA Nemotron content-safety guardrails — Nemotron-Content-Safety-Reasoning-4B (text) and multimodal Nemotron-3-Content-Safety. Produces a Markdown policy, JSON taxonomy, and drop-in inference prompts. Maps rough words or an existing policy to V2 categories, adding custom categories or topic-following rules.

Quality

73%

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 ./skills/nemotron-policy-generator/SKILL.md

Quality

Content

71%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 body is a well-organized, actionable instruction skill with excellent progressive disclosure and a clear six-step workflow. Its main weakness is repetition of the severity/Sn-label distinction and a duplicated 'trained on these exact shapes' phrase that inflates token cost without adding clarity.

Suggestions

State the Sn-label-vs-severity distinction once (ideally in Step 3's `sn_label`/`severity` field notes) and remove the repeats in the Target Models and Step 5 sections.

Remove the duplicated 'both models were trained on these exact shapes and deviating reduces accuracy' sentence at lines ~165–167, keeping a single occurrence.

Consider adding one fully filled example category inline in Step 3 to make the field spec copy-paste ready, which would push actionability to 5.

Dimension	Reasoning	Score
Conciseness	Mostly efficient and domain-specific (it does not explain generic concepts Claude already knows), but the Sn-label-vs-severity distinction is restated three times (lines ~90, ~141, ~169) and 'both models were trained on these exact shapes and deviating reduces accuracy' appears twice in succession (lines ~165 and ~167), which could be tightened.	3 / 5
Actionability	Concrete, executable guidance throughout: a six-step workflow, an explicit per-category field list with examples, named template files in assets/, and exact output filenames — minor gaps only because no fully filled example category is shown inline.	4 / 5
Workflow Clarity	A clearly sequenced six-step process with a validation checkpoint ('Validate against it before saving' for the JSON schema) and ambiguity-handling guidance in Step 1; not a full validate→fix→retry feedback loop, so it stops short of 5.	4 / 5
Progressive Disclosure	Well-structured overview with one-level-deep references: a dedicated 'Reference Files' section lists each real references/ and assets/ file with a one-line purpose, and content is appropriately split (taxonomy and model specs in references/, templates in assets/).	5 / 5
	Total	16 / 20 Passed

Description

75%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, distinctive, and rich in natural trigger terms, but it omits any explicit 'Use when...' trigger guidance, capping its completeness. Adding a usage-trigger clause would lift it to a top-tier description.

Suggestions

Append an explicit 'Use when...' clause naming concrete trigger situations (e.g., 'Use when the user asks to build, draft, or extend a Nemotron/NCS safety policy, BYO taxonomy, or guardrail config, or describes moderation rules in rough words').

Add a few common-synonym trigger terms users actually say ('moderation policy', 'guardrail config', 'Aegis taxonomy') to round out keyword coverage.

Keep the third-person voice and concrete output list as-is — those are already strong.

Dimension	Reasoning	Score
Specificity	Lists multiple concrete actions — 'Generates BYO custom safety policies', 'Produces a Markdown policy, JSON taxonomy, and drop-in inference prompts', 'Maps rough words or an existing policy to V2 categories, adding custom categories or topic-following rules' — with comprehensive coverage of the skill's outputs.	5 / 5
Completeness	The 'what' is clear and concrete, but there is no 'Use when...' clause or equivalent explicit trigger guidance, which per the rubric caps completeness at 3.	3 / 5
Trigger Term Quality	Strong natural term coverage ('safety policies', 'content-safety guardrails', 'BYO custom safety policies', 'Markdown policy', 'JSON taxonomy', 'inference prompts', 'V2 categories', 'topic-following rules'), though a few common user phrases like 'moderation policy', 'guardrail config', or 'Aegis taxonomy' are absent.	4 / 5
Distinctiveness Conflict Risk	A clear niche (NVIDIA Nemotron content-safety guardrails) with named models and V2 taxonomy triggers, making conflict with unrelated skills minimal.	5 / 5
	Total	17 / 20 Passed

Validation

87%

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 — 14 / 16 Passed

Validation for skill structure

Criteria	Description	Result
metadata_field	'metadata' should map string keys to string values	Warning
frontmatter_unknown_keys	Unknown frontmatter key(s) found; consider removing or moving to metadata	Warning

	Total	14 / 16 Passed

Repository: NVIDIA-NeMo/Nemotron
Path: skills/nemotron-policy-generator/SKILL.md
Commit: 907d408

Reviewed: 6 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.