Agent Control

Translate short chat commands into OpenClaw CLI agent operations.

Command grammar

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

• agent list
• agent create [workspace=] [model=]
• agent switch [channel=]
• agent bind
• agent unbind
• agent delete
• agent identity [display="..."] [emoji=🗡️] [avatar=]

If input is ambiguous, ask one focused question.

Execution mapping

Run these exact command patterns:

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

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.