CLAUDE.md

Claude Code can read a project-level CLAUDE.md.

The orchestrator can write one so each worker knows its role and boundaries before it starts.

Create one

python "$CC_ORCHESTRATOR_HOME/cc_orchestrator.py" write-claude-md --cwd /path/to/project --role implementation

For review:

python "$CC_ORCHESTRATOR_HOME/cc_orchestrator.py" write-claude-md --cwd /path/to/project --role review

Through MCP, Codex can call:

cc_write_claude_md

What it tells the worker

The generated file tells Claude Code:

• Codex is the controller, planner, reviewer, and final decision maker.
• Claude Code is an external worker process.
• The worker has one assigned role.
• Secrets must not be printed.
• Destructive commands are not allowed unless explicitly requested.
• Unrelated user changes must not be reverted.
• Progress and verification should be reported clearly.

Existing files are protected

If CLAUDE.md already exists, the command is conservative.

Option	Behavior
default	Refuses to overwrite an existing unmanaged file
--append	Adds the managed orchestrator section
--force	Writes a timestamped backup, then replaces the file

Recommended flow

1. Codex plans the work.
2. Codex writes CLAUDE.md for the selected role.
3. Codex launches Claude Code through CLI or MCP.
4. Claude Code follows the role rules.
5. Codex reviews logs, diffs, and verification results.