agent [OPTIONS]
| Flag | Default | Description |
|---|
-p, --prompt <TEXT> | — | Execute a single prompt and exit (non-interactive) |
-m, --model <MODEL> | claude-sonnet-4-20250514 | Model to use |
--api-base-url <URL> | auto-detected | API endpoint URL |
--api-key <KEY> | from env | API key (prefer env var) |
--provider <NAME> | auto | LLM provider: anthropic, openai, or auto |
--permission-mode <MODE> | ask | Permission mode: ask, allow, deny, plan, accept_edits |
--dangerously-skip-permissions | false | Skip all permission checks |
-C, --cwd <DIR> | current dir | Working directory |
--max-turns <N> | 50 | Maximum agent turns per request |
-v, --verbose | false | Enable verbose output |
--dump-system-prompt | false | Print the system prompt and exit |
-h, --help | — | Show help |
--version | — | Show version |
| Variable | Equivalent flag | Description |
|---|
AGENT_CODE_API_KEY | --api-key | API key (highest priority) |
ANTHROPIC_API_KEY | --api-key | Anthropic API key |
OPENAI_API_KEY | --api-key | OpenAI API key |
AGENT_CODE_API_BASE_URL | --api-base-url | API endpoint URL |
AGENT_CODE_MODEL | --model | Model name |
# Interactive mode with Anthropic
ANTHROPIC_API_KEY=sk-ant-... agent
# One-shot with OpenAI
OPENAI_API_KEY=sk-... agent --model gpt-4o --prompt "explain main.rs"
# Local Ollama
agent --api-base-url http://localhost:11434/v1 --model llama3 --api-key x
# CI: fix tests without asking
agent --dangerously-skip-permissions --prompt "fix the failing tests"
# Read-only exploration
agent --permission-mode plan
# Debug: see what the LLM receives
agent --dump-system-prompt