nemo-curator

Curate LLM training data: dedupe, filter, PII redaction.

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 ./optional-skills/mlops/nemo-curator/SKILL.md

Quality

Content

53%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 well-structured with real reference files and concrete code, but it is padded with redundant performance/benchmark sections, its main code examples are admittedly non-runnable 0.x pseudocode, and the batch pipeline lacks any validation checkpoints. Progressive disclosure is good but under-uses the references directory.

Suggestions

Collapse the duplicate performance content (Quick start Performance, GPU vs CPU table, Performance benchmarks, Cost comparison) into a single concise section to remove padding.

Add explicit validation/verification checkpoints to the curation pipeline (e.g., verify record counts before/after dedup, sample-check filtered output, confirm PII redaction coverage) to lift workflow clarity above the batch-operation cap.

Replace or clearly mark the 0.x conceptual snippets with runnable 1.x stage examples, or move the detailed 0.x API reference into a separate reference file and keep only the 1.x quickstart inline.

Dimension	Reasoning	Score
Conciseness	Code blocks are lean and assume Claude's competence, but performance figures (16x speedup, benchmark tables) are repeated across 'Performance', 'GPU vs CPU', 'Performance benchmarks', and 'Cost comparison' sections, creating noticeable padding; not 2 because no basic concepts are over-explained.	3 / 5
Actionability	Installation commands and the 1.x pipeline skeleton are executable, but the bulk of stage/multi-modal snippets are explicitly flagged as 'conceptual' 0.x code that is not runnable as written; the flexibility is justified, yet the majority of guidance is non-executable pseudocode, matching anchor 3.	3 / 5
Workflow Clarity	Stage 1-4 sequencing is clear, but this is a batch/destructive data pipeline with no validation or verification checkpoints (no output checks after dedup/filter/redaction), so per the rubric cap workflow clarity stays at 3; not 4 because checkpoints are absent rather than merely minor gaps.	3 / 5
Progressive Disclosure	Clear section headers and two real, one-level-deep reference links (references/filtering.md, references/deduplication.md, both present) give good navigation; not 5 because substantial API-detail and benchmark content is inlined rather than split into reference files, leaving only two bundle files for a very large body.	4 / 5
	Total	13 / 20 Passed

Description

66%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 concise and names a clear domain with three concrete capabilities, but it omits any explicit 'Use when...' trigger guidance, which caps completeness. Trigger-term and specificity coverage are good but lack synonyms and full action coverage.

Suggestions

Add an explicit 'Use when...' clause naming natural triggers (e.g., 'Use when preparing LLM training data from web scrapes, deduplicating corpora, filtering low-quality text, or redacting PII').

Expand trigger synonyms for broader recall (e.g., 'deduplication', 'data cleaning', 'personally identifiable information').

Mention at least one more concrete action (e.g., classifier-based quality filtering) to round out capability coverage.

Dimension	Reasoning	Score
Specificity	Names the domain ('Curate LLM training data') and three concrete actions ('dedupe, filter, PII redaction'), matching the anchor that lists several specific actions with minor coverage gaps; not 5 because classification, multi-modal, and scaling are absent.	4 / 5
Completeness	Provides a clear 'what' but no 'Use when...' clause or equivalent trigger guidance, so per the rubric completeness is capped at 3; not 4 because 'when' is entirely absent rather than weakly implied.	3 / 5
Trigger Term Quality	Includes natural working terms ('LLM training data', 'dedupe', 'filter', 'PII redaction') that practitioners would say, giving good keyword coverage; not 5 because synonyms like 'data cleaning', 'deduplication', and 'personally identifiable information' are missing.	4 / 5
Distinctiveness Conflict Risk	The LLM-training-data curation niche with dedupe/filter/PII is mostly distinct with only minor overlap risk against general data-processing skills; not 5 because 'filter' and 'dedupe' are generic verbs that several other skills could also claim.	4 / 5
	Total	15 / 20 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
metadata_version	'metadata.version' is missing	Warning
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	13 / 16 Passed

Repository: NousResearch/hermes-agent
Path: optional-skills/mlops/nemo-curator/SKILL.md
Commit: 6680afb

Reviewed: about 4 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.