npx claudepluginhub jrc2139/claude-plugin-sema --plugin semaSemantic code search and code intelligence for Claude Code.
claude plugins marketplace add jrc2139/sema-plugins
claude plugins install sema
git clone https://github.com/jrc2139/sema-plugins.git
claude plugins link ./sema-plugins/packages/claude
The sema binary (v0.2.0+) must be installed:
curl -fsSL https://sema.sh/install.sh | sh
Use the /sema skill or run sema directly:
# Hybrid search (default) - best for natural language questions
sema "where is authentication handled?"
sema "error handling patterns" -l python
# Keyword search (-k) - fast, no model loading, good for identifiers
sema -k "parseArgs"
sema -k "ConfigLoader"
# Search within a directory
sema "API endpoints" src/
sema "database queries" ./backend/
# With filters
sema "error handling" -l zig -n 10
sema -g "src/**/*.ts" "authentication"
sema --exclude "**/tests/*" "main function"
Direct database queries for symbol lookup and structural filtering (requires a previously built index):
# Find symbol definitions
sema find parseConfig # Find by exact name
sema find handle --prefix # Prefix match
sema find init --kind function # Filter by kind
sema find MyClass --exported # Only exported symbols
# Find symbol references
sema refs AuthHandler # Find chunks that reference a symbol
# Query by structural metadata
sema query --kind function --exported --no-docstring # Undocumented public functions
sema query --kind method --min-complexity 10 # Complex methods
sema query --role test --language zig # All test code in Zig
sema query --parent MyClass --kind method # All methods in MyClass
Auto-Indexing: First search automatically indexes your codebase (no manual sema index required)
-k): Builds FTS-only index instantly, no model loadingServer Lifecycle: Hybrid search spawns sema serve in the background for fast searches (~50ms)
Search: Queries are sent to the server for instant results (subsequent runs)
<path>: Search within directory (positional, e.g., sema "query" src/)-k, --keyword: BM25 text search (no model loading, instant)-n <num>: Max results (default: 5)-l <lang>: Filter by language (python, javascript, zig, etc.)-g <pattern>: Filter by file path glob (e.g., "src/**/*.ts")--exclude <pattern>: Exclude files matching glob pattern (e.g., "**/tests/", ".md")-c, --compact: File paths only| Mode | Flag | Best For | Speed |
|---|---|---|---|
| Hybrid | (default) | Natural language questions | ~50ms |
| Keyword | -k | Exact identifiers, function names | ~5ms |
By default, sema respects .gitignore files when indexing. You can also use .semaignore files for sema-specific exclusions (same syntax as .gitignore):
# .semaignore - exclude from sema but keep in git
generated/
*.min.js
vendor/
.semaignore files work at any directory level, just like .gitignore.
To index everything (ignore no patterns):
sema index --no-gitignore .
Or in config:
{
"respect_ignore": false
}
Ultra-compressed communication mode. Cuts ~75% of tokens while keeping full technical accuracy by speaking like a caveman.
Intelligent prompt optimization using skill-based architecture. Enriches vague prompts with research-based clarifying questions before Claude Code executes them
Persistent memory system for Claude Code - seamlessly preserve context across sessions