Works with the AI Agents you already use
gemma code search
Semantic code search for repositories using AST-aware chunking and Google EmbeddingGemma-2.
Free
gemma code search
Example session with this skill installed
Find the implementation of the JWT token refresh logic in the backend.
- Read your context and instructions
- Compiled the gemma code search
Searching for "JWT token refresh logic" in the current repository...
Found 2 matches across 2 files.
Search Results
| Score | File Path | Scope | Snippet |
|---|---|---|---|
| 0.892 | src/auth/service.py | AuthService.refresh_session | def refresh_session(self, token: str):\n claims = self.jwt.decode(token)\n if claims.is_expired():\n raise Unauthorized()\n return self.issue_pair(claims.user_id) |
| 0.845 | src/middleware/auth.ts | verifyRefreshToken | export const verifyRefreshToken = (req: Request) => {\n const token = req.cookies['refresh_token'];\n if (!token) return null;\n return jwt.verify(token, process.env.REFRESH_SECRET);\n}; |
=> [...] continue reading in <src/auth/service.py>: lines 140-210 (query matched 142-148)
Next steps
- Run
code-search index --paths src/auth/service.pyif you just modified the refresh logic. - Inspect the full implementation in
src/auth/service.pyat line 140 to see theissue_pairmethod. - Use
code-search "JWT secret configuration"to find where the environment variables are loaded.
Connects securely to your tools. The creator never sees your data.
About this skill
Finding logic in a massive codebase shouldn't feel like guessing regex patterns. Grep fails when you don't know the exact variable names, and standard LLM context limits break on large repos. This skill provides semantic code search that understands intent, not just strings.
What it does
- Semantic search finding code by meaning or behavior using Google EmbeddingGemma-2.
- AST-aware chunking breaking files into logical functions, classes, and methods for precise results.
- Hybrid ranking fusing exact symbol matches with semantic vectors to ensure high recall.
- Incremental indexing updating only what has changed based on git commits to save time.
- Cross-language support handling Python, JS/TS, Go, Rust, PHP, and Markdown.
How it works
- Index initialization using
code-search index --forceto build the initial vector store of your repo. - Natural language query running
code-search "your query"to find relevant code snippets. - Targeted updates using
--pathsto re-index specific files after you make edits. - Daemon management controlling the shared model process that stays in RAM for fast subsequent searches.
Frameworks & tools
Built with Python and Tree-sitter for syntax-tree parsing. It uses the Google EmbeddingGemma-2 model with either a Torch or ONNX backend, optimized for both GPU and CPU-only environments.
Why this beats prompting it yourself
Raw prompts lack the context of an entire repository and suffer from hallucinated file paths. This skill provides a grounded search index that maps your intent to actual code chunks, ensuring you work with the real source of truth rather than a model's guess.
Use cases
- Locating specific logic like "rate limiting" or "auth middleware" across unfamiliar microservices.
- Understanding how complex tokens or session handlers are managed throughout the stack.
- Discovering existing utility functions to prevent code duplication in large teams.
- Scanning documentation files for architectural decisions using natural language.
Known limitations
Does not auto-reindex dirty or uncommitted files during a standard search. Large text files are split by token limits rather than syntax trees. Requires a manual rebuild if switching between Torch and ONNX backends.
Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for LLM Ops.
Standard search excludes uncommitted changes. Documentation files lack AST chunking. Switching backends requires a full index rebuild. Indexing never auto-resumes after a crash without manual retry.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before buying a skill.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Free to use
No account required to browse
Trust & safety
Security scan
Security scan passed 8 days ago
- Free with an Agensi account
- Accounts
Astral
- Software
- ruff
uv (Python tool runner)
Python
Git
Permissions
Allowed Hosts