llava

Large Language and Vision Assistant. Enables visual instruction tuning and image-based conversations. Combines CLIP vision encoder with Vicuna/LLaMA language models. Supports multi-turn image chat, visual question answering, and instruction following. Use for vision-language chatbots or image understanding tasks. Best for conversational image analysis.

Quality

82%

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

65%

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

The body is highly actionable with executable code and commands, but it is somewhat verbose and lacks validation checkpoints for risky operations. It also fails to link the bundled training reference, leaving progressive disclosure underdeveloped.

Suggestions

Trim marketing/padding (the "Metrics" star counts and "GPT-4V level capabilities" claims) and consolidate redundant sections to tighten the token budget.

Add explicit validation/recovery checkpoints to the training, quantization, and batch-processing workflows (e.g., verify loss/gradient health after each stage, confirm output validity before proceeding).

Replace the inline "Training custom model" section with a signaled link to references/training.md (e.g., "See [training.md](references/training.md) for the full training guide") so the bundle is discoverable.

Dimension	Reasoning	Score
Conciseness	Mostly code-forward and avoids long conceptual teaching, but the ~295-line body includes padding (a "Metrics" block with "23,000+ GitHub stars," "GPT-4V level capabilities (targeted)") and redundant sections that could be trimmed, so it is not fully lean. It is better than a 1 (no 'PDF is a common format...' style concept explanation) but not tight enough for a 3.	2 / 3
Actionability	Provides executable, copy-paste-ready code for loading the model, processing images, and generating responses, plus concrete CLI and Gradio launch commands with specific flags. This matches the fully-executable score-3 anchor rather than the pseudocode score-2 anchor.	3 / 3
Workflow Clarity	Multi-step processes (training stages, multi-turn conversation) are sequenced, but risky operations like training, quantization, and batch processing lack explicit validation checkpoints or error-recovery feedback loops, which caps workflow clarity at 2 per the rubric's feedback-loop guideline.	2 / 3
Progressive Disclosure	A bundle file exists at references/training.md but the body's "Training custom model" section is inline and never signals or links to it, so a one-level-deep reference is present but not clearly signaled and content that could be separate stays inline.	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.

The description is strong: it concisely states concrete capabilities, includes natural trigger terms, explicitly addresses both what and when, and carves out a distinct niche. It avoids vagueness and verbosity.

Dimension	Reasoning	Score
Specificity	Lists multiple specific concrete actions ("visual instruction tuning," "image-based conversations," "multi-turn image chat," "visual question answering," "instruction following"), matching the score-3 anchor rather than the domain-only score-2 anchor.	3 / 3
Completeness	Explicitly answers both what (the capability list) and when ("Use for vision-language chatbots or image understanding tasks. Best for conversational image analysis"), with explicit trigger guidance rather than implied-only.	3 / 3
Trigger Term Quality	Includes natural terms a user would say ("image chat," "visual question answering," "image understanding tasks," "vision-language chatbots," "image-based conversations") with good coverage of common variations, not jargon-only.	3 / 3
Distinctiveness Conflict Risk	Occupies a clear vision-language niche with distinct triggers unlikely to fire for other skills; it is not a generic "helps with models" description that would conflict.	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: 2 missing, 2 deeper-than-1-level	Warning

	Total	14 / 16 Passed

Repository: Orchestra-Research/AI-Research-SKILLs
Path: 18-multimodal/llava/SKILL.md
Commit: 773a529

Reviewed: about 11 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.