ClawHub Skill Market Q1 2026 Review: Which Skills Are Actually Worth Installing

The Selection Problem

ClawHub has been live for over a year. Skill count is approaching 300. For new users, that number is starting to create selection paralysis.

As SFD Lab, we've installed around 60 skills and actually used most of them. This is our Q1 review — not an advertisement, not a ranked list, just our actual usage record.

Must-Install Tier

humanizer: Before any long-form content goes out, run it through humanizer. The AI writing patterns it catches — em dash overuse, "it's worth noting," "at the end of the day," rule-of-three structures — are genuinely pervasive in AI-generated text, and genuinely off-putting when readers notice them.

self-improving-agent: The learning loop for your agent. Mistakes get documented, patterns get promoted to rules, recurring errors become history. The ROI compounds over time.

deep-research-pro: Multi-source research that produces cited outputs. For any task requiring more than a single web search, the quality improvement over naive web_fetch is significant.

Useful But Situational

PDF skill: Invaluable when you need it. Weekly use for us, not daily.

video-gen: The quality is genuinely impressive for product showcase content. Limited utility for content requiring people or text.

translate-cli: For teams with multilingual content needs. We use it constantly; teams publishing in one language won't need it.

How to Evaluate Any Skill

Check the SKILL.md completeness. Look for: clear scope definition, documented limitations, error handling guidance. A skill with a complete and honest SKILL.md has been tested in production. A skill with a thin SKILL.md probably hasn't.