Skip to main content

​
Overview

The Coding capability lets CORE spawn coding agents like Claude Code on your machine. Start a coding session from WhatsApp, check its progress from the web dashboard, and let it work autonomously while you’re away from your computer. Sessions run in the background and persist their output, so you can resume or read results anytime.

​
Supported Agents

Currently supported:
  • Claude Code - Anthropic’s CLI coding agent
More agents coming soon. The system is extensible to support any CLI-based coding agent.

​
Commands

​
Start a Session

corebrain coding start
This prompts you for:
  • Agent to use (e.g., claude-code)
  • Working directory
  • Initial prompt
Or provide everything directly: 
corebrain coding start --agent claude-code --dir /path/to/project --prompt "Fix the auth bug"

​
List Sessions

corebrain coding list
Shows all sessions with their status (running, completed, closed).

​
Read Session Output

corebrain coding read <sessionId>

# Read last 50 lines
corebrain coding read <sessionId> --lines 50 --tail

​
Resume a Session

corebrain coding resume <sessionId>
Prompts for a new message to continue the session.

​
Close a Session

corebrain coding close <sessionId>
Stops a running session.

​
Gateway Tools

When the gateway is running, CORE can use these coding tools:

​
coding_start_session

Start a new coding session: 
{
  "agent": "claude-code",
  "prompt": "Fix the authentication timeout issue",
  "dir": "/Users/me/projects/myapp"
}
Optional parameters:
  • model - Specific model to use
  • systemPrompt - Custom system prompt

​
coding_resume_session

Resume an existing session: 
{
  "sessionId": "abc-123",
  "prompt": "Now add tests for the fix"
}

​
coding_read_session

Read session output: 
{
  "sessionId": "abc-123",
  "lines": 100,
  "tail": true
}

​
coding_close_session

Close a session: 
{
  "sessionId": "abc-123"
}

​
coding_list_sessions

List all sessions (no parameters).

​
Session Lifecycle

1

Start

CORE starts a session with an agent, directory, and prompt. The session gets a unique ID.
2

Running

The agent runs in the background. Output is streamed to a session file.
3

Monitor

CORE (or you) can read the session output at any time while it’s running.
4

Complete

The agent finishes its task. Session status changes to “completed”.
5

Resume or Close

Resume with a new prompt to continue, or close to clean up.

​
Configuration

Configure agents before first use: 
corebrain coding config --agent claude-code
This sets up the command path and any agent-specific options.

​
Use Cases

Remote bug fixing: You’re at dinner, a P0 bug comes in. Message CORE: “Fix the payment processing error in checkout.ts”. CORE spawns Claude Code, provides context from memory, and works on the fix. Async code reviews: “Review the changes in the auth PR and suggest improvements.” CORE starts a session, analyzes the code, and sends you a summary when done. Automated refactoring: “Migrate all API routes from Express to Hono.” CORE works through the codebase file by file, and you check in periodically to see progress. Night shifts: Set a reminder: “At 2am, run the database migration script and report any errors.” CORE handles it while you sleep.

​
Notes

  • Sessions are stored in ~/.corebrain/sessions.json
  • Session output is read from the agent’s native session files
  • Multiple sessions can run concurrently
  • Sessions persist across gateway restarts