feat: missing tools + scoped approval system

.gitignore

@@ -6,3 +6,6 @@ webview/node_modules/
 webview/dist/
 .lucent/
 package-lock.json
+.playwright-mcp/
+docs/regression-screenshots/
+docs/regression-report-*.md

.playwright-mcp/console-2026-03-20T17-22-00-332Z.log

@@ -0,0 +1 @@
+[     548ms] [ERROR] Failed to load resource: the server responded with a status of 404 (Not Found) @ http://localhost:5173/favicon.ico:0

.playwright-mcp/console-2026-03-20T18-43-51-205Z.log

@@ -0,0 +1 @@
+[     332ms] [ERROR] Failed to load resource: the server responded with a status of 404 (Not Found) @ http://localhost:5173/favicon.ico:0

.playwright-mcp/console-2026-03-21T07-20-13-106Z.log

@@ -0,0 +1 @@
+[     814ms] [ERROR] Failed to load resource: the server responded with a status of 404 (Not Found) @ http://localhost:5173/favicon.ico:0

.playwright-mcp/console-2026-03-21T08-11-48-416Z.log

@@ -0,0 +1,12 @@
+[   13898ms] TypeError: Cannot read properties of undefined (reading 'trim')
+    at Object.sendMessage (http://localhost:5177/src/stores/chat.ts:33:18)
+    at http://localhost:5177/src/App.tsx:62:21
+    at dispatch (eval at evaluate (:301:30), <anonymous>:2:37)
+    at eval (eval at evaluate (:301:30), <anonymous>:6:3)
+    at UtilityScript.evaluate (<anonymous>:303:16)
+    at UtilityScript.<anonymous> (<anonymous>:1:44)
+[   14604ms] TypeError: Cannot read properties of undefined (reading 'trim')
+    at Object.sendMessage (http://localhost:5177/src/stores/chat.ts:33:18)
+    at http://localhost:5177/src/App.tsx:62:21
+    at dispatch (eval at evaluate (:301:30), <anonymous>:2:37)
+    at eval (eval at evaluate (:301:30), <anonymous>:13:7)

.vscodeignore

@@ -4,6 +4,7 @@ webview/src/**
 webview/node_modules/**
 node_modules/**
 *.map
+**/*.map
 .gitignore
 tsconfig.json
 esbuild.config.mjs
@@ -12,3 +13,6 @@ scripts/**
 marketing/**
 docs/**
 .github/**
+.playwright-mcp/**
+.lucent/**
+lucent-code-*.vsix

docs-site/.gitignore

@@ -0,0 +1,3 @@
+node_modules/
+build/
+.docusaurus/

docs-site/README.md

@@ -0,0 +1,22 @@
+# Lucent Code Docs
+
+Docusaurus 3.x documentation site for [Lucent Code](https://lucentcode.dev).
+
+## Local development
+
+```bash
+npm install
+npm start        # dev server at http://localhost:3000
+npm run build    # production build → build/
+```
+
+## Cloudflare Pages deployment
+
+| Setting | Value |
+|---|---|
+| Root directory | `docs-site` |
+| Build command | `npm run build` |
+| Build output directory | `build` |
+| Node.js version | `20` |
+
+Set the `NODE_VERSION` environment variable to `20` in the Cloudflare Pages project settings.

docs-site/docs/developer/adding-skills.md

@@ -0,0 +1,98 @@
+---
+sidebar_position: 2
+title: Adding Skills
+description: How to write a skill file, the frontmatter spec, and how skills are loaded from different sources.
+---
+
+# Adding Skills
+
+A **skill** is a Markdown file with a YAML frontmatter header. The body is the prompt that gets sent to the AI when the skill is invoked.
+
+## Skill File Format
+
+```markdown
+---
+name: my-skill
+description: One sentence describing what this skill does
+---
+
+Your prompt here. Write clear, structured instructions for the AI.
+
+Include examples, output formats, and any constraints you want enforced.
+```
+
+**Frontmatter fields:**
+
+| Field | Required | Description |
+|---|---|---|
+| `name` | Yes | Slug used in the `/` picker and for `use_skill` tool calls |
+| `description` | Yes | Shown in the skill picker and system prompt — keep it to one sentence |
+
+The `name` and `description` values may optionally be quoted (single or double quotes — both are stripped).
+
+## Built-in Skills
+
+Built-in skills live in `src/skills/builtin/` as individual `.md` files. They're bundled at build time by esbuild's text loader.
+
+To add a new built-in skill:
+
+1. Create `src/skills/builtin/your-skill.md` with frontmatter
+2. Import it in `src/skills/builtin/index.ts`:
+   ```typescript
+   import yourSkill from './your-skill.md';
+   export const BUILTIN_SKILLS: readonly string[] = [
+     // ... existing skills ...
+     yourSkill,
+   ];
+   ```
+3. Build: `npm run build`
+
+## User Skills (`~/.claude/skills/`)
+
+Create a directory per skill:
+
+```
+~/.claude/skills/
+  my-skill/
+    SKILL.md       ← skill content (frontmatter + body)
+  another-skill/
+    SKILL.md
+```
+
+Lucent Code reads each subdirectory's `SKILL.md` on startup.
+
+## Skill Sources
+
+Skills are loaded from multiple sources, appearing in the picker grouped by source label:
+
+| Label | Source |
+|---|---|
+| `builtin` | Bundled with the extension |
+| `claude` | `~/.claude/skills/` and `~/.claude/plugins/cache/` |
+| `github` | Fetched from a GitHub repository |
+| `npm` | Fetched from an npm package via unpkg.com |
+| `marketplace` | Fetched from the Superpowers registry |
+| `local` | Loaded from a local directory path |
+
+## SkillRegistry API
+
+`src/skills/skill-registry.ts` — central registry:
+
+```typescript
+class SkillRegistry {
+  load(content: string, source?: string): void    // parse and index one skill
+  clear(): void                                    // reset all loaded skills
+  get(name: string): Skill | undefined
+  getSummaries(): SkillSummary[]                  // name + description + source
+  getContent(name: string): string | undefined    // full body
+}
+```
+
+`getSummaries()` is called when building the system prompt (lightweight — no body content). `getContent()` is called when the AI invokes `use_skill`.
+
+## Writing Good Skills
+
+- **Be specific about output format.** The AI will follow explicit format instructions. Vague instructions produce inconsistent output.
+- **Show don't tell.** Include a concrete example of the expected output in the skill body.
+- **Keep descriptions scannable.** The description appears in the AI's system prompt — one sentence is enough.
+- **YAGNI.** Don't add instructions for edge cases that won't come up. Longer prompts don't mean better results.

docs-site/docs/developer/architecture.md

@@ -0,0 +1,103 @@
+---
+sidebar_position: 1
+title: Architecture
+description: How the extension is structured — the VS Code host process, the SolidJS webview, and the postMessage protocol between them.
+---
+
+# Architecture
+
+Lucent Code is a VS Code extension split into two processes that communicate via VS Code's webview postMessage API.
+
+## High-Level Structure
+
+```
+┌─────────────────────────────────────────────────────────────┐
+│ VS Code Extension Host (Node.js)                            │
+│                                                             │
+│  src/extension.ts          ← activation entry point        │
+│  src/chat/chat-handler.ts  ← OpenRouter API calls          │
+│  src/skills/               ← skill registry & sources      │
+│  src/tools/                ← editor tool implementations   │
+│  src/mcp/                  ← MCP client manager            │
+│  src/context/              ← LSP context gatherer          │
+│  src/search/               ← codebase indexer              │
+└──────────────────┬──────────────────────────────────────────┘
+                   │  postMessage (WebviewMessage / ExtensionMessage)
+┌──────────────────▼──────────────────────────────────────────┐
+│ Webview (SolidJS, bundled by esbuild)                       │
+│                                                             │
+│  webview/src/App.tsx           ← root component            │
+│  webview/src/components/       ← UI components             │
+│  webview/src/services/         ← vscode API bridge         │
+└─────────────────────────────────────────────────────────────┘
+```
+
+## Message Protocol
+
+All communication uses typed messages defined in `src/shared/types.ts`.
+
+**Webview → Extension** (`WebviewMessage`):
+```typescript
+type WebviewMessage =
+  | { type: 'sendMessage'; content: string; images?: string[]; model: string }
+  | { type: 'getModels' }
+  | { type: 'setModel'; modelId: string }
+  | { type: 'toolApprovalResponse'; requestId: string; approved: boolean; scope?: ApprovalScope }
+  | { type: 'listFiles'; query: string }
+  // ... more
+```
+
+**Extension → Webview** (`ExtensionMessage`):
+```typescript
+type ExtensionMessage =
+  | { type: 'streamChunk'; content: string }
+  | { type: 'streamEnd'; usage?: Usage }
+  | { type: 'modelsLoaded'; models: OpenRouterModel[] }
+  | { type: 'toolApprovalRequest'; requestId: string; toolName: string; args: Record<string, unknown>; diff?: DiffLine[] }
+  | { type: 'usageUpdate'; lastMessageCost: number; sessionCost: number; creditsUsed: number }
+  // ... more
+```
+
+The webview sends messages via `vscode.postMessage()`. The extension sends messages via `panel.webview.postMessage()`.
+
+## Key Source Files
+
+| File | Purpose |
+|---|---|
+| `src/extension.ts` | Activation, panel creation, message routing |
+| `src/chat/chat-handler.ts` | Builds prompts, calls OpenRouter API, streams responses, routes tool calls |
+| `src/skills/skill-registry.ts` | Loads and indexes skills from all sources |
+| `src/skills/sources/` | Skill source implementations (builtin, claude-code, github, npm, marketplace, local) |
+| `src/tools/editor-tools.ts` | Implements editor tool calls (write_file, rename_symbol, etc.) |
+| `src/mcp/mcp-client-manager.ts` | Spawns and manages MCP server subprocesses |
+| `src/context/context-gatherer.ts` | Collects LSP context (file, selection, diagnostics) |
+| `src/search/indexer.ts` | Vector indexing for @codebase search |
+| `src/shared/types.ts` | **All shared types** — read this first |
+| `webview/src/App.tsx` | Root SolidJS component, message handler |
+| `webview/src/components/ChatMessage.tsx` | Renders a single message with markdown, code blocks, tool cards |
+| `webview/src/components/ChatInput.tsx` | Input with @mentions, skill chips, file attachment |
+| `webview/src/components/ModelSelector.tsx` | Model picker dropdown |
+
+## Build System
+
+esbuild bundles both the extension and the webview:
+
+```bash
+npm run build         # production build
+npm run watch         # incremental rebuild on file change
+```
+
+Config: `esbuild.config.mjs`. Notable: `.md` files are loaded as text strings (used for built-in skills).
+
+## Data Flow: Sending a Message
+
+1. User types in `ChatInput.tsx` and presses Enter
+2. Webview posts `{ type: 'sendMessage', content, model }` to extension
+3. `extension.ts` routes to `ChatHandler`
+4. `ChatHandler` builds the system prompt (active file, skills list, capabilities, LUCENT.md)
+5. OpenRouter API called with streaming enabled
+6. Each token streamed back as `{ type: 'streamChunk', content }`
+7. If AI calls a tool: `{ type: 'toolApprovalRequest' }` sent to webview
+8. User approves → webview posts `{ type: 'toolApprovalResponse', approved: true }`
+9. Tool executed, result appended to messages, API called again
+10. `{ type: 'streamEnd', usage }` sent when done; `{ type: 'usageUpdate' }` updates status bar

docs-site/docs/developer/building-locally.md

@@ -0,0 +1,111 @@
+---
+sidebar_position: 5
+title: Building Locally
+description: How to set up your development environment, run a debug build, and work with the esbuild bundler.
+---
+
+# Building Locally
+
+## Prerequisites
+
+- Node.js 20+
+- VS Code 1.85+
+- Git
+
+## Setup
+
+```bash
+git clone https://github.com/lucent-org/lucent-code
+cd lucent-code
+npm install
+```
+
+This installs dependencies for both the extension host (`src/`) and the webview (`webview/`). They share a single root `package.json`.
+
+## Development Build
+
+Press **F5** in VS Code to launch the Extension Development Host — a separate VS Code window with the extension loaded from source.
+
+Or manually:
+
+```bash
+npm run watch        # rebuild on every file change (esbuild incremental)
+```
+
+Then open the Command Palette (`Ctrl+Shift+P`) → **Developer: Reload Window** to pick up changes.
+
+## Production Build
+
+```bash
+npm run build        # one-off production build
+npm run package      # build + vsce package → .vsix file
+```
+
+Output:
+- `dist/extension.js` — bundled extension host
+- `dist/webview.js` — bundled webview
+
+## esbuild Config
+
+`esbuild.config.mjs` bundles both targets. Key configuration:
+
+```javascript
+// Extension host — CommonJS for VS Code
+{
+  entryPoints: ['src/extension.ts'],
+  platform: 'node',
+  format: 'cjs',
+  external: ['vscode'],
+  loader: { '.md': 'text' },   // built-in skills imported as strings
+}
+
+// Webview — ESM/browser
+{
+  entryPoints: ['webview/src/index.tsx'],
+  platform: 'browser',
+  format: 'iife',
+}
+```
+
+The `.md` text loader allows skill files to be imported directly:
+```typescript
+import codeReview from './code-review.md';  // returns file content as string
+```
+
+## Shared Types
+
+`src/shared/types.ts` is imported by both the extension host and the webview. It's the single source of truth for the message protocol, model types, and conversation types.
+
+## TypeScript Declaration for `.md` Imports
+
+`src/markdown.d.ts` declares the module type so TypeScript is happy:
+
+```typescript
+declare module '*.md' {
+  const content: string;
+  export default content;
+}
+```
+
+## Testing
+
+Run regression test scripts against a live OpenRouter API:
+
+```bash
+node scripts/test-skills-full.mjs        # test all built-in skills
+node scripts/test-code-review.mjs        # test code-review skill specifically
+```
+
+These scripts call the OpenRouter API directly — you need `OPENROUTER_API_KEY` set in your environment (or a `.env` file).
+
+## Packaging for Marketplace
+
+```bash
+npm run package
+```
+
+Generates `lucent-code-X.Y.Z.vsix`. Install locally with:
+
+```bash
+code --install-extension lucent-code-X.Y.Z.vsix
+```