DeepSeek Harness plugin

semantic-search

A DeepSeek Harness (dsh) plugin for local semantic code search. Builds a fragment-level index of a workspace (symbol-aware chunking for many languages), computes local embeddings (lightweight lexical

Jump to install

Source facts

Repository
JohnXu22786/semantic-search
Latest update
Aug 20, 2026
Category
Docs & Rendering
GitHub stars
0
Format
plugin
Catalog evidence
Upstream dsh.bundle evidence
Evidence path
package.json#dsh.bundle
Checked against
0.1.0-rc.8
Upstream check date
2026-08-20

This evidence comes from the upstream catalog. This site has not installed, run, or security-reviewed the plugin.

Install

Start with a prompt that asks an agent to review the GitHub repository and source. Switch to the command if you want to install it yourself.

Copy this prompt into DSH, Codex, or another agent and ask it to review the GitHub repository and source first.

Do not install or run any commands yet. Read this plugin's GitHub repository, README, and relevant source code. Then answer the questions below clearly and directly so I can decide whether it fits my needs:

1. What is this plugin, and what problem does it solve?
2. Who is it for, and what are its typical use cases?
3. How is it used after installation? Include one minimal example.
4. What known limitations or privacy, security, compatibility, or maintenance risks does it have?
5. Give a clear recommendation: recommend, conditionally recommend, or do not recommend, with reasons.

Distinguish statements documented by the repository, inferences from source code, and unknowns. If evidence is insufficient, say so explicitly. Do not guess or simply repeat the README.

GitHub: https://github.com/JohnXu22786/semantic-search
Plugin: semantic-search
Author: JohnXu22786

Check the source files

Read the README and other files from this plugin directory before installing.

File explorer4 files
README.mdSource · read only
README language

dsh-semantic-search

![CI](https://github.com/JohnXu22786/semantic-search/actions/workflows/ci.yml)

Local semantic code search for DeepSeek Harness (dsh) — and any Node script.

> 中文文档:README.zh.md

sema builds a fragment-level index of a workspace: source is tokenized by a language-aware tokenizer (camel/snake/kebab splitting, CJK n-grams), chunked with symbol-aware boundaries (functions/classes stay intact), and embedded into fixed dimension vectors — locally and dependency-free by default (feature-hashed lexical TF-IDF), or via any OpenAI-compatible embedding endpoint. Queries run a hybrid retrieval (vector cosine + BM25, fused with reciprocal-rank fusion) over the index, so meaning-based search works even when exact terms don't match.

Ships as a dsh plugin bundle — the sema_search, sema_reindex and sema_stats tools on the harness tool registry — plus a standalone sema CLI.

---

Highlights

  • Works offline by default — the built-in lexical provider needs no network,

no model download, no API key. Use it as a fast BM25-plus code search.

  • Symbol-aware chunking — boundaries from a conservative per-language table

(16 languages) keep functions/classes intact; a missed boundary degrades to a plain chunk rather than breaking.

  • CJK-aware tokenizer — n-gram tokenization (default bigrams) aligns Chinese

queries and documents without a segmentation library; full-width punctuation is folded, not treated as a hard break.

  • Hybrid retrieval with RRF — vector cosine + BM25 channels are fused by

reciprocal-rank fusion, so a document found by one channel still ranks.

  • Graceful degradation — if a configured remote provider is unreachable, the

index falls back to the local lexical provider (configurable via allowFallback).

  • Incremental refresh + file watchingsema_reindex diffs by size+mtime,

and an optional watcher keeps the index live.

  • Persistence — the index is saved to <root>/.sema atomically (JSON metadata

+ binary vectors), with staleness detection when the provider/dimension changes.

  • Deterministic — the same workspace and options produce the same index and

the same ranked answers.

Supported languages

TypeScript, JavaScript, Python, Go, Rust, Java, Kotlin, Scala, C, C++, C#, Objective-C, Ruby, PHP, Swift, Bash, plus common data/markup formats (JSON, YAML, TOML, Markdown, HTML, XML...).

---

Installation

As a dsh bundle

The package declares "dsh": { "bundle": { "patch": "./cordis.patch.yml" } }. The patch inserts one plugin row that mounts the bundle and registers sema_search / sema_reindex / sema_stats on ctx.tools.

# from npm (name reserved; publish pending access setup)
npm install -g dsh-semantic-search

# straight from this repository
dsh plugin --profile demo add github:JohnXu22786/semantic-search

# or from a local checkout
dsh plugin --profile demo add /path/to/semantic-search

As a standalone CLI

npm install -g dsh-semantic-search   # or: npm run build && node bin/sema.mjs
sema --help

---

CLI usage

sema index                build the full index from the workspace
sema reindex [--full]     incremental refresh (or full rebuild with --full)
sema search <query...>    hybrid vector + BM25 search, prints top hits
sema stats [--json]       index health, provider, and sizing numbers

Global options:

--root <dir>          workspace root (default: current directory)
--data-dir <dir>      index storage directory (default: <root>/.sema)
--provider <kind>     embedding provider: lexical | openai (default: lexical)
--dim <n>             embedding dimension (lexical default: 4096; openai 0 = auto)
--base-url <url>      OpenAI-compatible embeddings endpoint base URL
--model <name>        embeddings model name (openai only)
--api-key <key>       API key (openai only; env: SEMA_EMBEDDING_API_KEY)
--top <n>             hits to print for search (default: 20)
--json                machine-readable output where supported
--help                show this help

CLI exit codes

  • 0 — success (including a search with zero hits and a --version/--help call).
  • 1 — a runtime failure (config error, build/index/search error).
  • 2 — a usage error: unknown command, unknown flag, or a missing query.

Configuration

The plugin is configurable through the bundle row's config (see cordis.patch.yml for an example), the CLI flags above, or defaults in code:

OptionDefaultMeaning
rootcwdworkspace root to index
dataDir.semaindex storage directory
provider.kindlexicallexical (offline) or openai
provider.dimension0 (lexical: 4096)embedding dimension; 0 = auto-infer from the endpoint
provider.baseUrlhttps://api.openai.com/v1OpenAI-compatible endpoint root
provider.modeltext-embedding-3-smallembedding model name
provider.apiKeyEnvSEMA_EMBEDDING_API_KEYenv var holding the API key
allowFallbacktruefall back to the lexical provider when a remote one fails
include / ignoredefaultsglob sets of files to index / skip
maxLinesPerChunk80hard chunk size upper bound
nGram2CJK n-gram size (1 disables n-gramming)
topK20hits returned by default
rrfK60RRF fusion constant
vectorK300candidates per channel before fusion
autosavetruepersist the index after builds
autoIndextruebuild lazily on first search
watchtruewatch the workspace for changes

Development

npm ci
npm test          # build + run the node:test suite (76 tests)
npm run typecheck
npm run build     # tsc -> lib/

---

License

MIT — see [LICENSE](LICENSE). © 2026 dsh-semantic-search contributors.