tao-finetune-clip

CLIP vision-language model for image-text retrieval, zero-shot classification, embedding extraction, ONNX export, and TensorRT deployment. Use when fine-tuning or training CLIP, running zero-shot classification, computing image embeddings, or deploying CLIP to ONNX/TensorRT.

Quality

85%

Does it follow best practices?

Run evals on this skill

Adds up to 20 points to the overall score

View guide

Securityby

Low

Low-risk findings worth noting

Quality

Content

70%

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

A highly actionable, domain-rich body with concrete spec blocks and commands, but it loses points on the missing validation/feedback loops for risky checkpoint and engine operations and on progressive disclosure due to several references pointing at files that are not in the bundle.

Suggestions

Add explicit validation checkpoints and error-recovery loops for risky flows — e.g. after checkpoint loading and after gen_trt_engine build, state how to detect and recover from failure before proceeding.

Fix broken references: the body cites defaults.json, config.json, references/model_info.yaml, and schemas/train.schema.json, none of which are in the bundle; either add those files or correct the paths to the bundled references that exist.

Tighten the long multi-clause paragraphs (opening checkpoint-behavior block and AutoML policy) into shorter sentences or bulleted rules to improve token efficiency and readability.

Dimension	Reasoning	Score
Conciseness	Mostly efficient and dense with domain specifics, but several dense paragraphs (e.g. the opening checkpoint behavior block, the AutoML policy paragraph) pack multiple caveats into long run-on sentences that could be tightened, and a few lines repeat guidance already stated elsewhere (checkpoint resolver usage appears twice).	2 / 3
Actionability	Provides concrete, copy-paste-ready spec override blocks for train/evaluate/inference/export/gen_trt_engine, exact checkpoint filenames, in-container commands, and precise env-var guidance — fully executable and specific.	3 / 3
Workflow Clarity	Action sequencing and per-action dataset requirements are clearly tabulated, but risky operations (checkpoint loading, TensorRT engine pairing, separate-encoder export) lack explicit validation checkpoints or feedback loops — the rubric caps workflow clarity at 2 when validation is missing for destructive/batch operations.	2 / 3
Progressive Disclosure	Structure is present with signaled one-level references (error-patterns.md, spec-param-inference.md, tao-deploy-clip.md) that all exist, but the body also references nonexistent bundle files (defaults.json, config.json, model_info.yaml, references/model_info.yaml, schemas/train.schema.json), which breaks the navigation contract and signals incomplete disclosure structure.	2 / 3
	Total	9 / 12 Passed

Description

100%

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, concise description: it enumerates concrete capabilities, includes an explicit "Use when" trigger with natural terms, and occupies a clear CLIP-specific niche. It is third-person and free of fluff, with no meaningful weaknesses.

Dimension	Reasoning	Score
Specificity	Lists multiple concrete actions — "image-text retrieval, zero-shot classification, embedding extraction, ONNX export, and TensorRT deployment" — matching the anchor for enumerating several specific capabilities.	3 / 3
Completeness	It states what the skill does (CLIP capabilities) and gives an explicit "Use when..." trigger clause covering training, zero-shot classification, embedding extraction, and ONNX/TensorRT deployment.	3 / 3
Trigger Term Quality	Natural user-facing terms are well covered ("fine-tuning or training CLIP", "zero-shot classification", "computing image embeddings", "deploying CLIP to ONNX/TensorRT"), giving good coverage of phrases a user would actually say.	3 / 3
Distinctiveness Conflict Risk	The CLIP vision-language niche with ONNX/TensorRT deployment triggers is distinct and unlikely to collide with other skills; it is specific to CLIP rather than generic ML.	3 / 3
	Total	12 / 12 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
frontmatter_unknown_keys	Unknown frontmatter key(s) found; consider removing or moving to metadata	Warning
referenced_paths_exist	Referenced path issues: 1 missing	Warning

	Total	14 / 16 Passed

Repository: NVIDIA/skills
Path: skills/tao-finetune-clip/SKILL.md
Commit: 2528d5b

Reviewed: 1 day 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.