TingdeLiu

Collaborative planning using abstract roles (designer + inspiration + reviewer).

Async via ask, end turn immediately; use when user explicitly delegates to any AI provider (gemini/codex/opencode/droid); NOT for questions about the providers themselves.

AutoFlow entry point. Routes to tp (plan) or tr (run).

Attach the newest context-transfer Markdown from the current project's ./.ccb/history/ into Claude using @file. Use when the user types /continue or asks to resume with the latest transfer file in this project.

Diagnose ROS 2 systems with evidence-first checks across nodes, topics, lifecycle states, QoS, and pipeline data flow. Use when debugging ROS 2 node discovery, topic connectivity, lifecycle, message flow, QoS, or CUDA-related suspicions.

Delegate file I/O to Codex via FileOpsREQ protocol. Use when Claude needs Codex to modify files.

Transform text content into professional Mermaid diagrams for presentations and documentation. Use when users ask to visualize concepts, create flowcharts, or make diagrams from text. Supports process flows, system architectures, comparisons, mindmaps, and more with built-in syntax error prevention.

Report which CCB providers are mounted (session exists AND daemon is online). Outputs JSON.

Set up a Claude Code custom status line (with an optional ASCII buddy/pet easter-egg) AND completion notification sound on Windows or macOS. Use when configuring statusLine in settings.json, when bash/jq/Unicode errors occur, when the user wants a sound alert when Claude finishes a response, or when adding a status-bar pet/companion. Auto-detects OS and Python vs Node.js runtime.

Analyze GitHub projects and generate comprehensive architecture documentation (architecture.md). Use when users request project analysis, architecture summaries, or documentation of code structure. Triggers on queries like "analyze this GitHub project", "generate architecture.md", "summarize project structure", or "document the codebase architecture".

Extract figures and/or tables from PDF files using the TF-ID model (Florence2-based object detection). Outputs cropped images with a Markdown index. Use when user wants to extract figures, tables, or images from a PDF file — e.g., "extract all figures from this PDF", "get the tables from paper.pdf", "extract images from pages 3-7". Requires conda TF-ID environment with pdf2image, transformers, pillow.

Extract and save subtitles from video URLs as plain .txt files. Supports Bilibili (B站) and YouTube. Use when the user provides a Bilibili or YouTube video link and asks to extract, download, or save subtitles/captions. Triggers on phrases like "提取字幕", "下载字幕", "extract subtitles", "get subtitles", or when a bilibili.com / youtube.com / youtu.be URL is provided with a subtitle-related request.

Automatically compress PDF files when they are too large for processing. Use this skill when (1) a PDF file exceeds 10MB and needs compression for efficient processing, (2) user explicitly requests PDF compression, (3) reading a PDF fails with a file size error, or (4) a PDF exceeds the 32MB API limit. Compresses using Ghostscript while maintaining acceptable quality.