Agent shell
What you'll do
Launch Craik before setup is complete, use the in-runtime command surface, and understand how one-shot prompts behave when the runtime is not yet fully configured.
Launch
Run craik with no subcommand to enter the agent shell entrypoint:
craik
On first launch, Craik does not require operator login, provider credentials, a selected model, or a project. Instead, it prints the current readiness state and the next actions required to make the runtime usable.
Craik Agent Shell
State: unconfigured
Profile: default
Model: not selected
Missing: operator session, provider credentials, active model
Next actions:
- run /auth login or craik auth login
- run /provider login openai, anthropic, gemini, or local
- run /model set <provider/model>
Use craik chat when you want the shell entrypoint to be explicit:
craik chat
One-shot mode
Use -z for quiet one-shot execution:
craik -z "Summarize the current readiness state."
The one-shot path emits only the final response. If Craik is not fully ready, the final response names the blocked state and the next action instead of printing the shell status card.
craik chat -q is the conversational alias:
craik chat -q "What model is active?"
Slash commands
The shell command surface is slash-command first. The same registry is used by the shell and by testable command dispatch.
/help
/status
/setup
/auth
/provider login openai
/model
/sessions
/approvals
/receipts
/skills
/exit
Unknown commands suggest the nearest known command. Readiness-gated commands explain what is missing instead of failing with an opaque stack trace.
Automation
Subsystem CLI commands remain available for scripts:
craik status
craik model status
craik session list
craik profile list
Use the shell for discovery and day-to-day interaction. Use subsystem commands when you need stable JSON output in automation.