osborn 0.5.3 → 0.5.5

package/.claude/settings.local.json

@@ -0,0 +1,9 @@
+{
+  "permissions": {
+    "allow": [
+      "Bash(ps:*)",
+      "Bash(osascript:*)",
+      "Bash(curl -s http://localhost:3000)"
+    ]
+  }
+}

package/.claude/skills/markdown-to-pdf/SKILL.md

@@ -0,0 +1,29 @@
+# Skill: Markdown to PDF
+
+Export Markdown documents as formatted PDF files.
+
+## When to use
+When the user wants to create a PDF from a Markdown file, spec, or research findings.
+
+## How to execute
+
+Option 1 — Using md-to-pdf (best quality):
+```bash
+npx --yes md-to-pdf "<MARKDOWN_PATH>"
+```
+This creates a PDF alongside the source file with the same name.
+
+Option 2 — Using pandoc (if available):
+```bash
+pandoc "<MARKDOWN_PATH>" -o "<OUTPUT_PATH>.pdf" --pdf-engine=wkhtmltopdf
+```
+
+Option 3 — Using markdown-pdf:
+```bash
+npx --yes markdown-pdf "<MARKDOWN_PATH>" -o "<OUTPUT_PATH>.pdf"
+```
+
+## Output
+- Save the PDF to the session workspace (e.g., `library/{name}.pdf`)
+- Confirm the output path and file size to the user
+- If the source is spec.md, name the output `spec-export.pdf`

package/.claude/skills/pdf-to-markdown/SKILL.md

@@ -0,0 +1,28 @@
+# Skill: PDF to Markdown
+
+Convert PDF documents to readable Markdown text.
+
+## When to use
+When the user provides a PDF file path and wants to read, search, or work with its contents.
+
+## How to execute
+
+Option 1 — Using the built-in Read tool:
+The Read tool can directly read PDF files. Use `pages` parameter for large PDFs (max 20 pages per request).
+
+Option 2 — Full extraction via CLI (for better formatting or batch processing):
+```bash
+npx --yes pdf-parse-cli "<PDF_PATH>"
+```
+
+Option 3 — Using pdftotext (if available):
+```bash
+pdftotext -layout "<PDF_PATH>" -
+```
+
+## Output
+Save the converted content to the session workspace as `library/{filename}.md` with:
+- Document title and source path at the top
+- Preserved heading structure where detectable
+- Tables converted to Markdown tables where possible
+- Page numbers as section markers

package/.claude/skills/playwright-browser/SKILL.md

@@ -0,0 +1,75 @@
+# Skill: Playwright Browser Automation
+
+Automate web browser interactions — navigate pages, click buttons, fill forms, take screenshots, and extract content.
+
+## When to use
+- Navigate to a URL and interact with it
+- Click buttons or links by their text or role
+- Fill form fields and submit data
+- Take screenshots of web pages
+- Extract text or structured data from pages
+- Automate multi-step web workflows (e.g. join a room, test a UI flow)
+
+## How to execute
+
+Uses `@playwright/cli` via npx — no global install needed. Token-efficient: uses element references (e.g. `e15`) instead of pixel coordinates.
+
+### First time only — install browser binaries
+```bash
+npx playwright install chromium
+```
+
+### Step 1 — Open a URL
+```bash
+npx @playwright/cli open https://localhost:3000
+```
+
+### Step 2 — Get page structure and element references
+```bash
+npx @playwright/cli snapshot
+```
+Returns an accessibility tree with element IDs like e1, e2, e15. Use these in subsequent commands.
+
+### Step 3 — Interact with elements
+```bash
+npx @playwright/cli click e15
+npx @playwright/cli fill e3 "some text"
+npx @playwright/cli press e3 Enter
+npx @playwright/cli select e7 "optionValue"
+npx @playwright/cli check e9
+npx @playwright/cli hover e12
+```
+
+### Take a screenshot
+```bash
+npx @playwright/cli screenshot --path=/tmp/page.png
+```
+
+### Close the browser
+```bash
+npx @playwright/cli close
+```
+
+### Named sessions (persistent state across commands)
+```bash
+npx @playwright/cli -s=myflow open https://localhost:3000
+npx @playwright/cli -s=myflow snapshot
+npx @playwright/cli -s=myflow fill e3 "abc123"
+npx @playwright/cli -s=myflow click e5
+npx @playwright/cli -s=myflow close
+```
+
+## Complete example — join Osborn voice room
+```bash
+npx @playwright/cli open http://localhost:3000
+npx @playwright/cli snapshot
+npx @playwright/cli fill e3 "abc123"
+npx @playwright/cli click e4
+npx @playwright/cli screenshot --path=/tmp/osborn-joined.png
+npx @playwright/cli close
+```
+
+## Notes
+- Runs headless by default. Add --headed to see the browser window.
+- Install browsers first if needed: npx playwright install chromium
+- Element IDs are session-scoped — run snapshot again after page changes

package/.claude/skills/youtube-transcript/SKILL.md

@@ -0,0 +1,24 @@
+# Skill: YouTube Transcript
+
+Fetch and save transcripts from YouTube videos.
+
+## When to use
+When the user asks to get a transcript, subtitles, captions, or summary from a YouTube video URL.
+
+## How to execute
+
+yt-dlp is installed on this system. Use this exact command:
+
+```bash
+yt-dlp --skip-download --write-auto-sub --sub-lang en --convert-subs srt -o "/tmp/yt-%(id)s" "<VIDEO_URL>"
+```
+
+This downloads auto-generated English subtitles as an SRT file to /tmp/yt-{video-id}.en.srt
+
+Then read the SRT file and strip the timing markers to get clean transcript text.
+
+## Output
+Save the cleaned transcript to the session workspace as `library/youtube-{video-id}-transcript.md` with:
+- Video title and URL at the top
+- Cleaned transcript text (strip SRT timing markers and duplicate lines)
+- Key timestamps preserved as section markers if meaningful breaks exist

package/dist/claude-llm.d.ts

@@ -11,6 +11,7 @@ import { type McpServerConfig } from '@anthropic-ai/claude-agent-sdk';
 import { EventEmitter } from 'events';
 export interface ClaudeLLMOptions {
     workingDirectory?: string;
+    sessionBaseDir?: string;
     permissionMode?: 'default' | 'acceptEdits' | 'bypassPermissions';
     allowedTools?: string[];
     eventEmitter?: EventEmitter;
@@ -18,6 +19,8 @@ export interface ClaudeLLMOptions {
     continueSession?: boolean;
     mcpServers?: Record<string, McpServerConfig>;
     model?: string;
+    voiceMode?: 'direct' | 'realtime';
+    skipTTSQueue?: boolean;
 }
 /**
  * Claude LLM - Wraps Claude Agent SDK for LiveKit
@@ -113,13 +116,38 @@ export declare class ClaudeLLM extends llm.LLM {
      * Check if checkpoints are available
      */
     hasCheckpoints(): boolean;
-    chat({ chatCtx, toolCtx, connOptions, }: {
+    /**
+     * Interrupt the current Claude query gracefully (like pressing Esc).
+     * Stops current tool execution but keeps the process alive.
+     * Returns true if interrupted, false if no active query.
+     */
+    interruptQuery(): Promise<boolean>;
+    /**
+     * Hard abort all active queries (like Ctrl+C).
+     * Kills subprocesses. Next message will spawn new processes.
+     */
+    abortQuery(): void;
+    /**
+     * Rewind file changes to a specific checkpoint.
+     * Uses the most recently added query (most likely to have the rewind capability).
+     */
+    rewindToCheckpoint(checkpointId?: string): Promise<boolean>;
+    /**
+     * Check if there are active queries that can be interrupted
+     */
+    hasActiveQuery(): boolean;
+    /** Add an active query (called from ClaudeLLMStream when query starts) */
+    setActiveQuery(q: any): void;
+    /** Remove an active query (called from ClaudeLLMStream when query completes) */
+    removeActiveQuery(q: any): void;
+    chat({ chatCtx, toolCtx, connOptions, abortController, }: {
         chatCtx: llm.ChatContext;
         toolCtx?: llm.ToolContext;
         connOptions?: APIConnectOptions;
         parallelToolCalls?: boolean;
         toolChoice?: llm.ToolChoice;
         extraKwargs?: Record<string, unknown>;
+        abortController?: AbortController;
     }): llm.LLMStream;
 }
 /**