cyberstrike run sends a prompt to an agent and streams the response to your terminal. It is the main entry point for scripting, CI pipelines, and headless automation.
message positional argument accepts one or more words that are joined into a single prompt. You can also pipe text via stdin.
Flags
Run a named command instead of a free-form prompt. The
message positional arguments are passed as the command’s arguments.Continue the most recent session instead of starting a new one.
Session ID to continue. Takes precedence over
--continue when both are provided.Fork the session before continuing. Creates a new branch from the existing session history. Requires
--continue or --session.Share the session and print the share URL after the run completes.
Model to use, specified as
provider/model (e.g. anthropic/claude-sonnet-4-5). Overrides the default model for this run only.Name of the agent to use (e.g.
web-application). If the agent is not found or is a subagent, the default agent is used.Output format. Choose
default for human-readable formatted output or json for newline-delimited JSON events suitable for piping to other tools.One or more file paths to attach to the message. May be specified multiple times. Directories are attached as
application/x-directory.Title for the session. If the flag is present but no value is given, the first 50 characters of the message are used.
Base URL of a running CyberStrike server to attach to (e.g.
http://localhost:4096). When set, no local server is started.Directory to run in. When using
--attach, this is the directory on the remote server.Port for the local server that
run starts internally. Defaults to a random port when omitted. Use this to expose the local server for external inspection or --attach from another process.Model variant controlling provider-specific reasoning effort (e.g.
high, max, minimal).Show thinking blocks in the output when the model supports extended reasoning.
Examples
JSON events
When--format json is set, each event is a newline-delimited JSON object containing a type field and a timestamp (Unix milliseconds). Common event types:
| Type | When |
|---|---|
text | The model produced a text block |
reasoning | The model produced a thinking block (requires --thinking) |
tool_use | A tool was invoked and completed |
step_start | A reasoning step began |
step_finish | A reasoning step finished |
error | The session encountered an error |
Permission requests are automatically rejected in
run mode. If an agent needs interactive permission, switch to the TUI or configure permission rules in your cyberstrike.json.