/search-code
Hybrid code search — semantic + keyword + BM25
> /plugin marketplace add sagarmk/beacon-plugin> /plugin install beacon@claude-code-beacon-plugin
Command definition
search-code.mddescription: "Hybrid code search — semantic + keyword + BM25"
argument-hint: <query> [query2] [--top-k N] [--threshold F]
allowed-tools: [Bash, Read, Glob]
/search-code
Search the codebase using Beacon hybrid search (semantic embeddings + BM25 keyword matching + identifier boosting).
Single query
1. Run `node ${CLAUDE_PLUGIN_ROOT}/scripts/search.js "$ARGUMENTS"` 2. Parse the JSON results — each match has `file`, `lines`, `similarity`, `score`, and `preview` 3. Review the `preview` field first — it often contains enough context to answer the query 4. Only read the actual source files if the preview is insufficient to answer the question 5. Summarize findings and cite file:line references
Multi-query batch
Pass multiple quoted queries for a single HTTP round-trip:
node ${CLAUDE_PLUGIN_ROOT}/scripts/search.js "auth flow" "session handling"Returns `[{query, matches}, ...]` grouped by query.
Options
- `--top-k N` — number of results (default: 10)
- `--threshold F` — minimum score cutoff (default: 0.35)
- `--path <prefix>` — scope results to files under a directory (e.g. `--path src/middleware/`)
- `--no-hybrid` — disable hybrid scoring, use pure vector search
Result fields
- `file` — file path
- `lines` — matched line range (e.g. "45-78")
- `similarity` — vector cosine similarity
- `score` — final hybrid score (semantic + BM25 + identifier boost)
- `preview` — first 300 chars of matched code chunk
Read more
description: "Hybrid code search — semantic + keyword + BM25" argument-hint: <query> [query2] [--top-k N] [--threshold F] allowed-tools: [Bash, Read, Glob]
/search-code
Search the codebase using Beacon hybrid search (semantic embeddings + BM25 keyword matching + identifier boosting).
Single query
1. Run `node ${CLAUDE_PLUGIN_ROOT}/scripts/search.js "$ARGUMENTS"` 2. Parse the JSON results — each match has `file`, `lines`, `similarity`, `score`, and `preview` 3. Review the `preview` field first — it often contains enough context to answer the query 4. Only read the actual source files if the preview is insufficient to answer the question 5. Summarize findings and cite file:line references
Multi-query batch
Pass multiple quoted queries for a single HTTP round-trip:
node ${CLAUDE_PLUGIN_ROOT}/scripts/search.js "auth flow" "session handling"Returns `[{query, matches}, ...]` grouped by query.
Options
- `--top-k N` — number of results (default: 10)
- `--threshold F` — minimum score cutoff (default: 0.35)
- `--path <prefix>` — scope results to files under a directory (e.g. `--path src/middleware/`)
- `--no-hybrid` — disable hybrid scoring, use pure vector search
Result fields
- `file` — file path
- `lines` — matched line range (e.g. "45-78")
- `similarity` — vector cosine similarity
- `score` — final hybrid score (semantic + BM25 + identifier boost)
- `preview` — first 300 chars of matched code chunk
Semantic code search plugin for Claude Code using hybrid vector search + BM25. 98% accuracy, 5x faster than grep.
Other commands on beacon.
More commands from other plugins.
build
CommandImplement tasks incrementally — build, test, verify, commit. Add "auto" to run the whole plan in one approved pass.

