AI Agent Hub
Back to plugins
🖥️

dsh-typelens

Client Updated 2026.08.28

Run the following command in DeepSeek Harness:

dsh plugin install fantasyce/dsh-typelens

Paste the following prompt into your AI chat to install this plugin:

Run dsh plugin install fantasyce/dsh-typelens in an environment where pnpm is available, then restart the affected DSH profile; the plugin source is at https://github.com/fantasyce/dsh-typelens

About this plugin

When driving agents through DeepSeek Harness, a recurring friction is that an agent reading a file has no inherent understanding of the surrounding type structure, and an edit produces no immediate diagnostic feedback unless the developer manually invokes an LSP tool. TypeLens removes that extra step: after every successful file read it appends only the complete type declarations needed to understand the visible code, and after every edit it follows with bounded TypeScript diagnostics, new errors first.

All analysis happens locally. TypeLens makes no outbound requests and never persists source-bearing state. Output is ranked, deduplicated, cycle-safe, and never cut mid-signature. Multiple bounds keep results predictable: workspace scope, file size, timeout, recursion depth, result count, cache size, and token ceiling. A built-in circuit breaker pauses automatic analysis after repeated operational failures while still leaving explicit tools and the health endpoint available, so an analysis hiccup can never turn a successful file operation into a failure.

If you write TypeScript, JavaScript, TSX, or JSX through DSH and want agents to automatically gain type context and immediate post-edit diagnostics without remembering to call an LSP tool, TypeLens is a lightweight plugin built for that. It exposes four explicit tools (lookup a type, list bounded types, run diagnostics, inspect configuration and health state), lifecycle hooks, a health API, and a Web settings panel so you can dial in analysis behavior and limits to match your workflow.

Use Cases

  • Agents automatically receive full type declarations after every successful file read without invoking an LSP tool
  • Bounded TypeScript diagnostics appear immediately after edits, with new errors listed first
  • Type semantic context is attached to agent steps entirely locally with no outbound requests

Best For

  • Developers who drive agents through DSH to write TypeScript, JavaScript, TSX, or JSX
  • Teams that want agents to stop re-reading files or manually requesting type context
  • Privacy-sensitive users who prefer fully local analysis with zero outbound requests and fail-open safety