CtrlK
BlogDocsLog inGet started
Tessl Logo
SkillPublic

nemotron-transcribe

Reference and routing skill for NVIDIA Nemotron 3.5 Transcribe. Use for model facts, checkpoint selection, inference constraints, and locating the canonical NeMo Speech main-stage training, fine-tuning, DFlash2, timestamp-serving, and quantization workflows; do not use as a substitute for those operational skills.

Quality

78%

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/nemotron-transcribe/SKILL.md
SKILL.md
Quality
Evals
Security

Quality

Content

67%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 an unusually disciplined routing document: lean, model-specific, and rich with exact upstream links and explicit gap statements. Its one serious defect is that the entire "Start here" navigation chain references files absent from the bundle, which breaks progressive disclosure and makes the entry sequence unexecutable as shipped.

Suggestions

Ship the referenced files (INDEX.md, context/quick-reference.md, model-card.md, recipes/overview.md) in the skill bundle, or remove/inline those pointers — currently the whole "Start here" chain is unresolvable.

Link the "Nemotron Transcribe vLLM cookbook" rows in the routing table and name the repository for usage-cookbook/Nemotron-3.5-Transcribe/README.md so every routing target is directly reachable.

Consolidate the precision guidance duplicated between "Source priority" and "Precision selection", and move the time-sensitive "only after it is publicly accessible" citation condition into a clearly marked note.

DimensionReasoningScore

Conciseness

The body is dense and model-specific ("supports recordings up to the one-hour training limit; internal 30-second chunks are not streaming") with no explanations of concepts Claude already knows. It is not anchor 5 because precision guidance is duplicated between "Source priority" and "Precision selection", and the time-sensitive instruction "Cite the stable report URL above only after it is publicly accessible" is a conditional that could be trimmed or rehomed.

4 / 5

Actionability

Guidance is concrete and executable: exact upstream skill links (train-speech-llm, finetune-speech-llm, train-dflash, quantize-speech-llm), config and middleware links (salm_nemotron.yaml, ctc_serving.py), specific settings ("reasoning disabled and temperature 0"), and exact architecture identifiers. Not anchor 5 because the "Nemotron Transcribe vLLM cookbook" routing rows carry no link and "from the repository root" never names which repository.

4 / 5

Workflow Clarity

A numbered "Start here" entry sequence plus an intent-to-source routing table with stage-order guidance ("give the relevant stage order and prerequisites first, then point to the upstream skill") forms a clear, coherent workflow; validation checkpoints are not applicable since this routing skill performs no destructive or batch operations. Not anchor 5 because the first four entry steps point to files that are not in the bundle, so the sequence cannot actually be followed as shipped.

4 / 5

Progressive Disclosure

On paper the body is an overview pointing one level deep (INDEX.md, context/quick-reference.md, model-card.md, recipes/overview.md, cookbook README), but the actual bundle contains none of these files — every internal reference is dangling, so the disclosure chain does not resolve. The SKILL.md itself is well-structured, keeping it above anchor 1, but the broken reference structure matches anchor 2's failure of content organization across files.

2 / 5

Total

14

/

20

Passed

Description

78%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.

A strong, specific description that clearly delimits a reference/routing role for a uniquely named model and includes an explicit "Use for" clause plus a negative boundary. The main improvement opportunity is framing triggers as user situations and adding natural speech-domain synonyms (transcription, ASR, audio).

DimensionReasoningScore

Specificity

The description lists several specific capability areas — "model facts, checkpoint selection, inference constraints, and locating the canonical NeMo Speech main-stage training, fine-tuning, DFlash2, timestamp-serving, and quantization workflows" — with broad coverage of the router's scope. It falls short of anchor 5 because these are mostly topic noun-phrases rather than multiple concrete action verbs (only "locating" is an action), and above anchor 3 because it names far more than 1-2 concrete items.

4 / 5

Completeness

Both parts are present: a clear "what" ("Reference and routing skill for NVIDIA Nemotron 3.5 Transcribe") and an explicit "Use for..." trigger clause, plus a negative scope statement ("do not use as a substitute for those operational skills"). It is not anchor 5 because the "when" is topic-based rather than framed as concrete user situations (e.g., "when the user asks to fine-tune or serve Nemotron Transcribe"), matching anchor 4's "'when' could be more explicit or specific".

4 / 5

Trigger Term Quality

Good keyword coverage: "Nemotron 3.5 Transcribe", "checkpoint selection", "fine-tuning", "DFlash2", "timestamp-serving", "quantization" are terms a user of this model would naturally say. Not anchor 5 because core synonyms for a speech model — "transcription", "ASR", "speech recognition", "audio", "vLLM/serving" — are absent; not anchor 3 because the primary product name plus the workflow terms cover most natural phrasing.

4 / 5

Distinctiveness Conflict Risk

"NVIDIA Nemotron 3.5 Transcribe" is a unique product niche, and the explicit disclaimer "do not use as a substitute for those operational skills" prevents overlap with the NeMo Speech skills it routes to. This matches anchor 5: clear niche with distinct triggers and minimal conflict risk.

5 / 5

Total

17

/

20

Passed

Validation

100%

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

Validation — 16 / 16 Passed

Validation for skill structure

No warnings or errors.

Repository

NVIDIA-NeMo/Nemotron

Path

skills/nemotron-transcribe/SKILL.md

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.