AnyAgent

Cline

Use Cline through AnyAgent: what you can ask for, how it behaves, and the flags underneath.

The cline adapter drives cline --json --auto-approve true, detected as the cline binary on PATH; import cline from anyagent/cline. The Cline CLI is Apache-2.0 open source and bring-your-own-key (BYOK): configure a provider once with cline auth -p openai -k your_api_key_here (openrouter, anthropic, openai, and gemini all work).

What you can ask for

model and effort (closed vocabulary: none, low, medium, high, xhigh) work on the CLI’s own flags; systemPrompt and schema are provided by AnyAgent. readOnly, resume, mcp, and model listing are undeclared and throw: no read-only run exists, since headless Cline auto-approves every tool and plan mode still runs shell commands. authStatus() reads the provider settings file and never runs the CLI, whose config subcommand needs a TTY. The full grid is on the support matrix.

Good to know

  • No session id is revealed headless, so result.sessionId stays absent; carry context forward in the next prompt instead.
  • The prompt must be positional, so a prompt larger than the OS argv limit needs agent.raw.
  • systemPrompt is provided by AnyAgent because Cline’s -s replaces the system prompt entirely; reach -s via extraArgs if replacement is what you want.
  • Usage arrives in the final run_result, with cache line items and cost; a finishReason other than "completed" throws Invocation with Cline’s own message.

Under the hood

Every run executes cline --json --auto-approve true with the prompt as a positional argument. Run options append flags:

OptionFlag
model-m <model>
effort--thinking <level>

On this page