cc

Short operator commands for Claude Code relay sessions.

Requires: claude-relay skill must be installed.

Script

All commands execute via:

{baseDir}/scripts/cc.sh <raw-args>

Command routing

Parse user input and route:

• projects / list → list available projects from map + project root
• on <project> / start <project> → start or reuse Claude session (fuzzy match)
• off [project] / stop [project] → stop session (default: last project)
• tail [project] [lines] → show recent output (default: 120 lines)
• status → list active relay sessions
• /cc (no args) → show inline button menu (if platform supports)
• /cc on (no project) → show project picker buttons
• Any other text → relay mode send (see below)

For the "any other text" case: if the first token resolves to an active project session, treat it as explicit project target and use remaining text as the message.

Relay mode

After on <project>, enter relay mode. This is the critical behavior contract:

1. ALL user messages are forwarded to Claude Code — no exceptions. Do NOT interpret, answer, or act on the message yourself. You are a transparent pipe.
2. Forward via scripts/cc.sh send → wait for output → return final result only.
3. The ONLY messages NOT forwarded are cc commands themselves: off, tail, status, projects, /cc.
4. Relay mode ends on off, stop, or /cc menu invocation.

Example:

[relay mode active, project=marvis]
User: "帮我查一下这个 bug 的原因"
→ cc.sh send marvis "帮我查一下这个 bug 的原因"
→ wait for Claude Code output
→ return result to user
WRONG: answering the question yourself

For button specs, output formatting, approval handling, and callback routing, see relay-mode.md.

Key principles

• Never self-answer in relay mode: forward everything, return only Claude Code's output.
• Final result only: one message per interaction, no progress updates.
• Choices → buttons: numbered menus in Claude Code output become inline buttons.
• Tool call discipline: button/menu messages = tool call + NO_REPLY, no surrounding text.

Environment variables