Skip to main content
skill CX Codex

lsp-guide — Claude Code skill: what it does, how to install

lsp-guide gives the agent recipes for the lsp tool: jump to a definition, list usages, read a signature, outline a file, or check what an edit broke.

Apache-2.0 ★ 5,622 updated 2026-10-29

Install into Claude Code · Skills CLI (Vercel) All methods and platforms ↓
npx skills add weave-os/router --skill lsp-guide -a claude-code
Open on GitHub

What lsp-guide does

The skill covers five operations of the lsp tool: definition, references, hover, documentSymbol and diagnostics. Queries go to a real language server for Go, TypeScript/JavaScript, Python and Rust, so imports, aliases and types get resolved rather than matched as text.

Each call needs coordinates — path, line and column, counted from 1, with the column sitting on the identifier's first character. Paths are relative to the working directory. Finding usages takes two moves: locate the declaration first with rg -n --column or documentSymbol, then ask for references.

Expect a slow first query in a new workspace, because the server builds its index; after a timeout the recipe says to retry once. Reference lists are capped, 100 by default. A file type no server handles comes back as advice text rather than an error, and a missing server can be installed through lsp_enable with the user's consent. Lookups also run inside dispatch subagents, which share the parent's warm servers.

What it helps with

  • find every call site of a function instead of text matches
  • read a type signature and doc comment without opening the definition file
  • get a file outline before reading a long module
  • see which errors an edit introduced, straight from the language server
  • tell apart same-named symbols from different packages

Example prompts for the agent

  • Where is <symbol> defined, and where is it used? Go through the language server, not grep.
  • Run diagnostics on <file> after my last edit.
  • Outline <file>, then show the signature and doc comment for <symbol>.

How to install lsp-guide

1Skills CLI (Vercel)

npx skills add weave-os/router --skill lsp-guide -a claude-code

2Manually: put the files in place

~/.claude/skills/lsp-guide/SKILL.md

Copy the whole skill folder; for one project use .claude/skills/lsp-guide/

Step by step, with commands for each platform: How to install skills in Claude Code: five ways

Questions about lsp-guide

How do I install lsp-guide in Claude Code?

Run npx skills add weave-os/router --skill lsp-guide -a claude-code. The skill lands at ~/.claude/skills/lsp-guide/SKILL.md.

What does lsp-guide do?

It hands the agent recipes for the lsp tool: definition lookup, usage search, hover for types and docs, file outlines and compiler diagnostics.

Which languages does it cover?

Go, TypeScript/JavaScript, Python and Rust, each through its own language server. For other file types grep stays the fallback.

Source text

SKILL.md · Apache-2.0
---
name: lsp-guide
description: Recipes for the lsp tool — resolving where a symbol is defined, every place it is used, type signatures and docs, file outlines, and compiler/type errors through a real language server (Go, TypeScript/JavaScript, Python, Rust). Use when navigating or explaining code by symbol, finding callers or usages, checking what broke after an edit, or deciding between lsp and text search.
---

# Using the lsp tool

The `lsp` tool answers *semantic* questions through the language's own server:
it resolves through imports, types, and aliases, where grep only matches text.
Reach for it whenever the question is about a **symbol**; keep grep for plain
strings, comments, and file types no server supports.

## Operations

| Operation | Answers | Needs |
|---|---|---|
| `definition` | where the symbol at a position is defined | `path`, `line`, `column` |
| `references` | every place the symbol at a position is used | `path`, `line`, `column` |
| `hover` | type signature + docs for the symbol at a position | `path`, `line`, `column` |
| `documentSymbol` | the outline (types, functions, methods) of one file | `path` |
| `diagnostics` | compiler/type errors and warnings for one file | `path` |

Positions are 1-based, and `column` must point **at the identifier** (its first
character works). All paths resolve against the working directory.

## Recipes

**"Where is X defined?"** — from any usage site of X:
`{"operation": "definition", "path": "<file>", "line": <L>, "column": <C>}`.
Don't know a usage site? Find one first (see next recipe's step 1).

**"Everywhere X is used"** — two steps, always:

1. Locate the declaration: `rg -n --column '\bX\b'` (the `--column` output is
   exactly the coordinate the next call needs), or `documentSymbol` when you
   already know the file.
2. `{"operation": "references", "path": "<decl file>", "line": <L>, "column": <C>}`.

A grep hit list is **not** a references answer: it includes comments and
strings, misses aliased imports, and can't distinguish same-named symbols.
Step 2 is what makes the answer authoritative.

**"What is this / what's its signature?"** — `hover` at the symbol. Prefer it
over reading the definition file when you only need the type or doc comment.

**"What's in this file?"** — `documentSymbol` before reading a large file; jump
straight to the line the outline gives you.

**"Did my edit break anything here?"** — `diagnostics` on the file after
editing. It reports the server's current analysis of the saved file.

## Behavior worth knowing

- **First query per workspace is slow** (the server indexes — up to a minute
  in a big repo). Later queries on that workspace are fast; don't give up
  after one timeout, retry once.
- **References are capped** (default 100, header says how many exist).
- **Unsupported file type or missing server** returns advice text, not an
  error — fall back to grep for that file. A missing server can be installed
  with the user's consent via `lsp_enable` (see the install-lsps skill for
  toolchain prerequisites).
- Works inside dispatch subagents too (they share the parent's warm servers),
  so symbol lookups are safe to fan out.

© weave-os · Apache-2.0 · weave-os/router

Similar skills

All skills for Agent setup: memory, context, orchestration →

From the same set20

The whole set →
router-session router-session prints the session ID used by Weave Router telemetry and logs. Use it to locate one Codex run in the router dashboard. CX Codex +1 ★ 5,622repo stars force-parent force-parent pins the thread you are in to one model through Weave Router, while subagents keep their own routing. Useful when a single task needs a different model. CX Codex +1 ★ 5,622repo stars models Lists available models in Weave Router and toggles them on or off directly through chat prompts. CX Codex +1 ★ 5,622repo stars router-models router-models is a skill for Claude Code and Codex that lists Weave Router models and lets you enable, disable, or rank them. CX Codex +1 ★ 5,622repo stars debug-claude-session debug-claude-session investigates Claude Code sessions by correlating local transcripts with cloud router logs via timestamps and models. Helps spot wire translation glitches when client rendering breaks. CX Codex +1 ★ 5,622repo stars test-claude-locally The test-claude-locally skill runs the Weave router in local Docker Compose and drives it with claude -p. Use it to reproduce routing bugs and inspect model behavior. CX Codex +1 ★ 5,622repo stars test-codex-locally test-codex-locally stands up the Weave router in docker compose, then drives it with codex exec from a throwaway CODEX_HOME to reproduce routing and marker bugs without touching your global Claude config. CX Codex +1 ★ 5,622repo stars pr-merge-ready pr-merge-ready processes pull request review comments in a single pass, batches edits by file, and avoids burning CI minutes with repeated pushes. CX Codex +1 ★ 5,622repo stars

Project sponsors

Companies that keep this analytics open