Semantic Skill Search
Semantic Skill Search
Sidekick's skill search understands natural language. Instead of requiring exact keyword matches, it uses AI embeddings to match what you're trying to do against what each skill is capable of — so you can find the right skill by describing your intent in plain English.
Overview
When you type a query into the skill search bar, Sidekick:
- Converts your query into a vector embedding using a language model
- Compares that embedding against pre-indexed embeddings for every skill's name, description, and capability metadata
- Returns results ranked by semantic similarity
This means you don't need to remember exact skill names or use the same words a skill author used. You just describe what you want to accomplish.
Where It Works
Semantic search is available in two places:
- Installed Skills — search across skills you've already connected to your Sidekick account
- Skill Marketplace — discover new skills from the marketplace by describing your use case
Example Queries
| What you type | What gets found |
|---|---|
Help me schedule meetings | Calendar & Scheduling skills |
Keep an eye on my pull requests | GitHub / GitLab monitoring skills |
Summarise my emails in the morning | Email triage and digest skills |
Turn off the lights when I leave | Smart home automation skills |
ClawHub Compatibility
Semantic search works across all skills, including the 13,000+ community skills available via ClawHub. Skills imported from ClawHub are indexed the same way as native Sidekick skills — their SKILL.md capability descriptions are embedded and made searchable immediately on import.
Tips for Better Results
- Describe the outcome, not the tool.
Remind me about important emailsworks better thanemail plugin. - Use natural phrases. Full sentences like
Help me manage my calendarare understood as well as short keywords. - Combine concepts. Queries like
automatically reply to Slack messages when I'm busycan surface multi-capability skills.
Fallback Behaviour
If semantic matching returns low-confidence results, Sidekick falls back to keyword-based filtering as a secondary signal to ensure precision. Both signals are combined in the final ranking.