claude-code-ywatanabelisted

# Custom Claude Code Skills for ywatanabe ## Sub-skills * [orchestrator](orchestrator.md) — Master orchestrator: delegate work, stay responsive, never block * [emacs-mcp-server](emacs-mcp-server.md) — Emacs MCP server workarounds, vterm control via eval_elisp * [self-compact](self-compact.md) — Self-compact context via Emacs vterm when auto-compact fails * [master-worker](master-worker.md) — Master-worker architecture: one Telegram bot, screen sessions per project * [environmental-variables](environmental-variables.md) — Env var locations for ywatanabe's secrets * [self-identify-vterm](self-identify-vterm.md) — Pattern for agents to find their own Emacs vterm buffer after compaction ## When to Use - Orchestrating work via Telegram (plotting, speaking, launching agents, monitoring) - Debugging multi-agent Python systems - Setting up Claude Code infrastructure (channels, plugins) - File placement conventions - Controlling Emacs vterm from Claude Code - Launching and monitoring Claude Code instances in Emacs vterm ## Core Guidelines ### Debugging Multi-agent debugging approach: @HOW-TO-DEBUG-with-MULTIPLE_AGENTS.md ### Claude Code Channels (Telegram) One bot serves all projects. Master agent at `~/proj` holds the Telegram channel. Workers run per-project in screen sessions. See [master-worker.md](master-worker.md) for the full setup. Key files: - Screen functions: `~/.bash.d/all/010_claude/020_claude_agents_in_screen.src` - Channel functions: `~/.bash.d/all/010_claude/010_