Agent Control

Manage OpenClaw isolated agents from chat with short commands. Use when the user asks to create/list/switch/bind/delete agents, route channels to a specific agent, or set an agent identity without manually typing full CLI syntax.

Translate short chat commands into OpenClaw CLI agent operations.

Command grammar

Accept these commands (case-insensitive, trim extra spaces):

• agent list
• agent create <name> [workspace=<path>] [model=<id>]
• agent switch <name> [channel=<channel[:accountId]>]
• agent bind <name> <channel[:accountId]>
• agent unbind <name> <channel[:accountId]>
• agent delete <name>
• agent identity <name> [display="..."] [emoji=🗡️] [avatar=<path>]

If input is ambiguous, ask one focused question.

Execution mapping

Run these exact command patterns:

• List:
  • openclaw agents list
• Create:
  • openclaw agents add <name> --workspace <path> --model <id>
  • Omit optional flags when missing.
  • Default workspace when omitted: ~/clawd/agents/<name>.
• Switch (route this channel/account to an agent):
  • openclaw agents bind --agent <name> --bind <channel[:accountId]>
  • If channel is omitted, infer from current surface when possible (e.g. webchat).
• Bind / unbind:
  • openclaw agents bind --agent <name> --bind <binding>
  • openclaw agents unbind --agent <name> --bind <binding>
• Delete:
  • Require explicit confirmation in the same turn for destructive action.
  • Then run: openclaw agents delete <name>
• Identity:
  • openclaw agents set-identity --agent <name> [--name <display>] [--emoji <emoji>] [--avatar <path>]

Response style

After each operation, return:

1. A one-line result (success/failure)
2. The next useful command for the user

Keep it brief and practical.

Safety rules

• Treat agent delete as destructive: confirm before running.
• Never run unrelated shell commands.
• If a command fails, show the error and one concrete fix.

scripts/

Use scripts/example.py as a deterministic helper wrapper for command parsing/execution when needed.