blip-2-vision-language

Vision-language pre-training framework bridging frozen image encoders and LLMs. Use when you need image captioning, visual question answering, image-text retrieval, or multimodal chat with state-of-the-art zero-shot performance.

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 complete executable code, but it is over-long for a SKILL.md overview and buries core-concept and performance detail inline instead of offloading it to the existing reference files. Batch workflows also lack validation checkpoints.

Suggestions

Move the architecture overview, Q-Former component tables, and performance/memory tables into references/advanced-usage.md (or a new concepts file) and keep SKILL.md as a lean quick-start + pointers.

Add explicit validation/checkpoint steps to the batch and retrieval workflows (e.g., verify inputs loaded, check feature shapes before indexing, handle decode failures) to lift workflow clarity above 2.

Trim repeated model-loading boilerplate across examples by defining the load step once and reusing it, cutting tokens that Claude can already infer.

Dimension	Reasoning	Score
Conciseness	At ~565 lines the body is padded with material Claude largely already knows (ASCII architecture diagrams, repeated model-loading boilerplate, performance tables), so it is mostly efficient but could be tightened rather than fully lean — not 3.	2 / 3
Actionability	Provides fully executable, copy-paste-ready code throughout (captioning, VQA, LAVIS usage, quantization, three complete workflow classes) with specific commands and concrete examples.	3 / 3
Workflow Clarity	Workflows are sequenced but lack explicit validation checkpoints and error-recovery feedback loops for batch operations, which per the rubric caps this dimension at 2.	2 / 3
Progressive Disclosure	Two references (advanced-usage.md, troubleshooting.md) are clearly signaled and one level deep, but large blocks of inline content (core concepts, architecture, performance tables) that belong in separate files keep it from the appropriately-split level at 3.	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 concise, specific, and includes an explicit 'Use when' trigger covering multiple concrete multimodal tasks. It clearly answers both what the skill does and when to invoke it with minimal risk of conflict.

Dimension	Reasoning	Score
Specificity	Lists multiple concrete actions — 'image captioning, visual question answering, image-text retrieval, or multimodal chat' — matching the multiple-specific-actions anchor rather than the partial-coverage anchor at 2.	3 / 3
Completeness	Explicitly answers both what ('Vision-language pre-training framework bridging frozen image encoders and LLMs') and when ('Use when you need...'), with an explicit 'Use when' trigger clause.	3 / 3
Trigger Term Quality	Uses natural user-facing terms ('image captioning', 'visual question answering', 'multimodal chat', 'image-text retrieval') that a user would actually say, giving good coverage rather than the some-keywords level at 2.	3 / 3
Distinctiveness Conflict Risk	Occupies a clear VLM niche with distinct triggers (captioning, VQA, image-text retrieval) unlikely to fire for unrelated skills, rather than the overlap-prone level at 2.	3 / 3
	Total	12 / 12 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
skill_md_line_count	SKILL.md is long (566 lines); consider splitting into references/ and linking	Warning
frontmatter_unknown_keys	Unknown frontmatter key(s) found; consider removing or moving to metadata	Warning
relative_links	Relative link issues: 7 missing	Warning

	Total	13 / 16 Passed

Repository: synthetic-sciences/openscience
Path: backend/cli/skills/llm-tools/blip-2/SKILL.md
Commit: 52845c3

Reviewed: about 9 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.