Limits & Safety

What To Expect Safely

Claw is powerful because it is close to the CLI, but it should be used with clear expectations.

Reasonable expectations:

• Claw runs work inside an isolated cloud runtime
• Claw uses your existing Suzi account context rather than guessing from chat alone
• Claw is optimized for operational workflows, not polished consumer-chat onboarding
• Claw can keep context across a session when the runtime is available

What Not To Assume

• Do not assume every reply is stateless.
• Do not assume /help is a stable command surface.
• Do not assume every notification type is implemented just because Claw can send outbound messages.
• Do not assume the docs can promise a confirmation layer that is not explicitly enforced by the Claw transport itself.

Telegram-Specific Constraints

• Long replies may be split across multiple Telegram messages.
• Rich replies are shaped for Telegram formatting, not raw markdown.
• Files and images are supported, but delivery still depends on Telegram transport constraints.

Operational Caveat

Internally, Claw has more moving parts than the simple user story suggests. Most users do not need the full internal model, but power users should know that some advanced behavior depends on runtime state, account setup, and Telegram connection state.

Best Way To Use Claw Safely

• Ask for concrete outcomes, not vague exploration.
• Keep high-value workflows in a single thread when possible.
• Use the CLI directly when you need exact command control or a fully explicit terminal workflow.

If you want the terminal-first surface, go back to the Suzi CLI docs.