Use machine-readable output and capability discovery instead of parsing human-readable help in scripts or agent workflows.
Discover the current contract
The response describes supported commands, authentication, browser modes, output schemas, lifecycle events, and recommended agent recipes.
Agents should inspect the contract at the start of a workflow rather than hard-code assumptions about flags or JSON fields.
Authenticate in CI
Credential priority is access token, API key, then stored credentials from octoparse auth login.
Do not print, commit, or include credentials in shared logs, screenshots, or generated documentation.
Choose an output mode
--json returns one result envelope.--jsonl streams foreground run events as JSON Lines.- Human output is intended for interactive terminal use.
See Output and exit codes for schemas, lifecycle events, stdout/stderr behavior, and failure codes.
Create tasks with an agent
Agents should use detect --agent or the prepare / preview / apply workflow, not --auto.
Follow machineContract.recipes.createTaskFromUrlWithAgent from the capabilities response. Only pass a trusted executable to --agent-command.
Confirm state-changing commands
Commands that rename, move, delete, update, or change schedules may require --yes. Use capabilities --json to discover confirmation requirements before constructing automation.
Internal commands
env switches internal API environments, and runs inspects internal local artifacts. They are not part of the primary user workflow. Prefer public task-ID and lot-ID commands unless Octoparse support instructs you otherwise. 