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

66%

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

Fix and improve this skill with Tessl

tessl review fix ./backend/cli/skills/llm-tools/llava/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 content is well-sectioned and highly actionable with executable code across inference, CLI, web UI, and quantization, but it is held back by verbosity from repeated stub snippets and generic advice, missing validation checkpoints in batch/training workflows, and poor progressive disclosure where training content is inlined instead of pointing to the existing reference file.

Suggestions

Replace the inlined "Training custom model" section with a one-line pointer to references/training.md (e.g., "See [training.md](references/training.md) for training and fine-tuning") so the existing reference is actually used and the body stays an overview.

Define or remove the undefined helper functions (ask(), generate()) used in the Common tasks and Multi-turn sections, or note they are pseudocode, so the examples are truly executable.

Add explicit validation checkpoints to batch inference and training workflows (e.g., verify decoded output is non-empty before continuing; confirm checkpoint saved) to satisfy the workflow-clarity feedback-loop requirement.

Trim the repetitive "Common tasks" snippets and generic "Best practices"/"Limitations" lists to reduce token cost and remove content Claude already knows.

Dimension	Reasoning	Score
Conciseness	The body is mostly efficient with sectioned, executable code, but it pads several sections ("Common tasks" repeats near-identical ask(model, image, question) snippets, "Best practices" and "Limitations" list generic advice Claude already knows) and includes redundant sections (training, resources) that could be trimmed, placing it at the 'mostly efficient with some unnecessary content' anchor.	3 / 5
Actionability	Provides concrete, copy-paste-ready code for loading models, inference, CLI, Gradio, and quantization with specific model paths and arguments; the only gaps are a few stub functions (ask(), generate()) that are referenced but never defined, keeping it just below fully executable.	4 / 5
Workflow Clarity	Multi-turn and training flows are sequenced, but batch/inference and training workflows lack explicit validation checkpoints (no "verify output" or "validate before proceeding" steps), so per the rubric's batch/destructive cap and missing-feedback-loop guidance it cannot exceed 3 despite the clear sequencing.	3 / 5
Progressive Disclosure	A references/training.md bundle file exists and holds the full training guide, yet the body inlines a "Training custom model" section that duplicates it and never links to training.md or any bundle file, leaving the reference orphaned and inlining content that clearly belongs in the separate file — the 'content that should be separate is inline' anchor.	2 / 5
	Total	12 / 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.

The description is specific, third-person, and clearly articulates both what the skill does and when to use it, with strong distinctiveness in the vision-language niche. It falls just short of top marks on trigger-term coverage and completeness because a few natural synonyms and sharper user-phrase triggers are missing.

Suggestions

Add a few more natural user synonyms to the trigger phrasing (e.g., "image description", "captioning", "reading what's in an image") to broaden trigger coverage.

Tighten the "Use for..." clause to reference the exact phrases a user would say (e.g., "Use when the user asks to analyze, describe, or chat about an image").

Consider trimming the architecture detail (CLIP/Vicuna/LLaMA) if the goal is a leaner trigger-focused description, since it adds length without improving triggerability.

Dimension	Reasoning	Score
Specificity	Lists several concrete actions ("visual instruction tuning", "image-based conversations", "multi-turn image chat", "visual question answering", "instruction following") and names components (CLIP, Vicuna/LLaMA), with only minor gaps in coverage rather than comprehensive enumeration.	4 / 5
Completeness	Clearly answers "what" (enables visual instruction tuning, image-based conversations, VQA) and provides an explicit "Use for..." trigger for when to use it; the "when" is present and concrete but could be more sharply tied to specific user phrasings, so it sits just below the level-5 anchor.	4 / 5
Trigger Term Quality	Includes natural user-facing terms like "image chat", "visual question answering", "image understanding tasks", and "vision-language chatbots", but omits common synonyms/variations a user might say (e.g., "image description", "captioning", "read an image").	4 / 5
Distinctiveness Conflict Risk	Targets a clear niche (vision-language chatbots / conversational image analysis) with distinct multimodal triggers and named architecture (CLIP + Vicuna/LLaMA), giving minimal overlap risk with non-vision or non-conversational image skills.	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
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: synthetic-sciences/openscience
Path: backend/cli/skills/llm-tools/llava/SKILL.md
Commit: fcd2e11

Reviewed: 2 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.