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

70%

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/blip-2/SKILL.md

Quality

Content

57%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.

A thorough, highly actionable reference with copy-paste-ready code across BLIP-2's main tasks, weakened by redundant inlined workflow classes, missing validation on the batch operation, and advanced content that should be offloaded to the existing reference files.

Suggestions

Move the three full workflow classes (ImageCaptioner, VisualQA, ImageSearchEngine) into references/advanced-usage.md and keep only concise quick-start snippets in SKILL.md to reduce redundancy.

Add an explicit verification step to the batch-processing example (e.g., check that output length matches input count, or log per-image failures) so the batch workflow has a validation checkpoint.

Relocate the 'Common issues' table to references/troubleshooting.md and the 'Performance optimization' section to references/advanced-usage.md, leaving SKILL.md as a lean overview with clearly signaled links.

Dimension	Reasoning	Score
Conciseness	The body is mostly efficient, BLIP-2-specific code, but the three full workflow classes (ImageCaptioner, VisualQA, ImageSearchEngine) largely restate the quick-start snippets, adding redundant padding that could be trimmed.	3 / 5
Actionability	Abundant copy-paste-ready, executable code covering the common cases (captioning, VQA, retrieval, quantization), with minor gaps where ITM and feature-extraction snippets reference undefined 'raw_image'/'device' variables.	4 / 5
Workflow Clarity	The workflows are runnable inference patterns rather than sequenced destructive operations, and the batch-processing example performs a batch operation without any validation or verification checkpoint, capping this dimension at 3.	3 / 5
Progressive Disclosure	Two real reference files exist and are clearly signaled one level deep, but substantial advanced content (three full workflow classes, performance optimization, common issues) is inlined in SKILL.md when it could live in the existing reference files.	3 / 5
	Total	13 / 20 Passed

Description

83%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, well-structured description that clearly states both what the skill does and when to use it with concrete, natural trigger terms. The main weaknesses are second-person voice (penalizing specificity) and minor overlap risk with sibling multimodal skills.

Dimension	Reasoning	Score
Specificity	Names four concrete actions (image captioning, visual question answering, image-text retrieval, multimodal chat) for comprehensive coverage, but the second-person phrasing 'Use when you need...' triggers the voice penalty reducing specificity by one.	4 / 5
Completeness	Explicitly answers both 'what' (framework bridging frozen image encoders and LLMs) and 'when' with concrete trigger phrases, matching the top anchor.	5 / 5
Trigger Term Quality	Strong natural keywords (image captioning, visual question answering, image-text retrieval, multimodal chat, zero-shot) with some synonyms, but a few common natural phrasings are missing.	4 / 5
Distinctiveness Conflict Risk	Clear BLIP-2 niche with distinct task triggers, but the named tasks overlap with closely related multimodal skills such as CLIP, LLaVA, and InstructBLIP, creating minor conflict risk.	4 / 5
	Total	17 / 20 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: 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.