DeepSeek Harness Plugin: dsh-tgrep
`dsh-tgrep` is a code search plugin for DeepSeek Harness, designed to address the high query latency of the default `grep` tool in large codebases. It leverages Microsoft's `tgrep` engine, replacing built-in tools via trigram indexing and Agent-plane shadowing to significantly improve search speed without modifying model presets. The plugin features automatic daemon operation, graceful degradation (falling back to `grep` mode when no index is available), and zero runtime dependencies. To use it, ensure the environment includes the `tgrep` binary and observe large-file limits and index-building requirements.
Read Moredsh-code-intel: Symbol-Aware Code Indexing and Hybrid Search for DeepSeek Harness
The DeepSeek Harness plugin dsh-code-intel is designed to address insufficient understanding of local codebases by agents. It supports multilingual code retrieval and outline parsing for TypeScript, JavaScript, Python, and more through symbol-aware indexing, hybrid search, and SQLite persistence. The plugin uses an asynchronous indexing mechanism; on first use, task completion must be awaited via a Job ID. It defaults to offline lexical mode and supports file system watching; to enable hybrid search, configure an OpenAI-compatible embeddings endpoint. Installation requires DSH version >=0.1.0-rc.6 and Node 22+, with secure data privacy.
Read MoreDeepSeek Harness Plugin: semantic-search Local Semantic Code Search
`semantic-search` is a local semantic code search plugin developed for DeepSeek Harness (DSH), designed to overcome the limitation of traditional keyword matching in understanding code semantics through vector indexing. It supports over 16 programming languages and offers core features such as offline embedding, hybrid retrieval (vector + BM25), and incremental updates. The plugin enables efficient code snippet retrieval without requiring an internet connection, significantly improving development and debugging efficiency.
Read MoreIntroducing the DeepSeek Harness Plugin: dsh-workspace-intelligence
`dsh-workspace-intelligence` is a standalone plugin for DeepSeek Harness designed to address the need for agents to look up code structure using only symbolic names — such as class names or function names — without relying on contextual information. It provides deterministic search through the LSP `workspace/symbol` request. The plugin communicates with the LSP via the `ctx` context, does not modify core code, and supports multiple response formats. To use it, Node.js and a language server must be installed, and the server command must be specified in the configuration. It includes concurrent query support, result deduplication, and robust error handling, is licensed under the MIT License, and is well suited for agent-level code structure queries.
Read Moresnapgrep: In-Process Code Search Plugin for DSH
snapgrep is a MIT-licensed tool maintained by Owen718, designed specifically for Pi and DeepSeek Harness (DSH). It aims to accelerate code search through in-process trigram indexing, addressing latency accumulation caused by frequent disk reads. The solution eliminates the need for sidecar processes or daemons, thereby reducing costs associated with startup, synchronization, and debugging. Within DSH, it replaces the built-in `grep` and `glob` tools, ensuring search results are byte-for-byte consistent with ripgrep; for unsupported queries, the system fails closed rather than returning approximate results. For deployment, it does not compile source code but only downloads a runnable addon of approximately 1.2 MB, supporting macOS, Linux (x64/arm64), and
Read More