What is semantic code search

Semantic code search finds code by meaning and intent rather than exact text, using embeddings to match concepts across the codebase and returning results with file paths, line numbers, relevance scores, and snippets.

How can I improve results

Describe the intent clearly, try broader terms first and then refine with more specific terms. If no good matches exist, fall back to a keyword grep to locate literal strings.

How do I act on found results

Open a result file to read context, then use follow up hints such as callers or symbols to trace usage and relationships, or group results by module to compare implementations.

search

Scanned

npx machina-cli add skill ConaryLabs/Mira/search --openclaw

Files (1)

SKILL.md

2.4 KB

Semantic Code Search

Search the codebase using Mira's semantic search to find code by meaning, not just text.

Query: $ARGUMENTS

Instructions

No arguments -- ask what to search for

If no query is provided ($ARGUMENTS is empty), ask the user what they want to find. Suggest examples:

"authentication middleware"
"error handling patterns"
"database connection logic"

With a query -- semantic search

If a query is provided:

Use the mcp__mira__code tool:

code(action="search", query="<the query>", limit=10)

Present results clearly with:
- File path and line numbers -- linked for navigation
- Relevance score -- how closely the code matches the intent
- Code snippet -- a short preview of the matching code
Group related results if they come from the same module
Omit results with very low relevance scores

When results are empty or low-relevance

If semantic search returns no results or only poor matches:

Suggest refining the query with more specific terms
Try a keyword fallback using Grep for literal string matching
Note that semantic search works best with descriptive intent ("how sessions are created") rather than exact identifiers ("session_start")

Example Output

Found 3 results for "session cleanup":

1. src/session/manager.rs:142-158 (score: 0.84)
   fn cleanup_expired_sessions() -- removes sessions older than TTL

2. src/hooks/stop.rs:45-62 (score: 0.71)
   fn on_stop() -- snapshots and archives the current session

3. src/background/tasks.rs:88-95 (score: 0.65)
   fn schedule_cleanup() -- periodic session maintenance task

Follow-Up Hints

After showing results, suggest next steps:

Read a result file for full context
code(action="callers", function_name="...") -- trace what calls a found function
code(action="symbols", file_path="...") -- explore the structure of a result file
Refine the query if results do not match the intent

Example Usage

/mira:search authentication middleware
/mira:search error handling patterns
/mira:search database connection pooling
/mira:search how are embeddings generated

Source

git clone https://github.com/ConaryLabs/Mira/blob/main/plugin/skills/search/SKILL.mdView on GitHub

Overview

Semantic Code Search lets you find code by concept or intent rather than exact text. It uses Mira's semantic search to match ideas like authentication, error handling, or database logic across the codebase. If no query is given, it prompts you to specify what to search for.

How This Skill Works

With a query, Mira runs a semantic search via the code action search tool with limit=10. Results are shown with file path and line numbers, a relevance score, and a code snippet, and related hits can be grouped by module; low relevance hits are omitted.

When to Use It

When you want to find code that implements a concept such as authentication middleware or error handling patterns
When you need to locate where a feature is implemented or how a function works within the codebase
When searching by meaning to compare different approaches to the same problem across modules
When you want to understand how a feature is constructed by exploring related modules and functions
When results are empty or have low relevance and you need to refine the query or fallback to a keyword search

Quick Start

Step 1: Run a query with the search command, for example /mira:search authentication middleware, or omit the query to be prompted
Step 2: Mira uses semantic search with code action search and returns up to 10 results showing file paths, line ranges, scores, and code snippets
Step 3: Review results, group by module if needed, and use follow up hints like code(action=callers) or code(action=symbols) to explore further

Best Practices

Use descriptive intents rather than single identifiers (example concept like authentication or session management)
Start with high level concepts and then narrow with more specific terms
Limit results to a manageable set (for example 10) and group by module when possible
Review relevance scores first to filter out weak matches
If results are poor, refine the query or try a keyword grep fallback

Example Use Cases

Found 3 results for authentication middleware: src/server/auth/middleware.rs 12-28 (score: 0.88); fn authenticate_user; src/api/auth.rs 45-60 (score: 0.74); fn login; src/hooks/auth.rs 10-22 (score: 0.66); fn require_auth
Found 2 results for error handling patterns: src/errors/patterns.rs 101-118 (score: 0.81); fn map_error; src/utils/error.rs 20-35 (score: 0.72); fn format_error
Found 2 results for database connection logic: src/db/connection.rs 40-58 (score: 0.84); fn connect; src/db/pool.rs 12-28 (score: 0.69); fn create_pool
Found 1 result for how embeddings are generated: src/embeddings/generator.rs 5-28 (score: 0.65); fn generate_embeddings
Found 3 results for how sessions are created: src/sessions/manager.rs 22-38 (score: 0.79); fn create_session; src/sessions/guards.rs 15-28 (score: 0.66); fn start_session; src/logging/session.rs 30-40 (score: 0.61); fn log_session_start

Frequently Asked Questions

Add this skill to your agents