CLI & slash commands — zengine docs

Main commands

Command	Purpose
`zagent init`	Create missing home files and folders
`zagent doctor`	Validate settings and test the configured model endpoint
`zagent version`	Print the binary version
`zagent help`	Show command help
`zagent -p "task"`	Run a headless task and print the final answer
`zagent sessions`	List sessions
`zagent show NAME`	Render a saved transcript; avoid using this command against an active engine in this preview
`zagent tools list`	List tools and manifest problems
`zagent tools run NAME --args JSON`	Developer tool test; approval requests are auto-granted in this mode

Run flags

Flag	Meaning
`--home DIR`	Home directory; otherwise `ZAGENT_HOME` or executable directory
`--workspace DIR`, `-w`	Working folder; defaults to current directory
`--session NAME`, `-s`	Create or resume a named session
`--session`	Resume the most recent session for this workspace, or create one
`--new`	Force a new session
`--prompt TEXT`, `-p`	Task text
`--prompt-file PATH`	Read prompt text from a file; `-` reads stdin
`--attach PATH`, `-a`	Add supported media; repeat for several attachments
`--ui headless\|cli\|web`	Choose interface; default is headless
`--output`, `-o`	Print the run transcript
`--output N`	Print the last N transcript lines
`--json`	Print run events as JSON lines
`--verbose`, `-v`	Stream model text and activity
`--quiet`, `-q`	Suppress progress output

Put run flags before a positional prompt. Prefer --prompt when writing scripts. Windows PowerShell examples use .\zagent.exe unless the executable is on PATH.

Slash commands

Command	Effect
`/help`	List built-in and custom commands
`/stop`	Interrupt the ordinary running turn and discard queued prompts
`/note TEXT`	Append a note to the next request
`/compact`	Request a continuation summary
`/status`	Show state, usage, and tasks
`/tasks`	List background tasks
`/tools`	List active tools
`/workspace PATH`	Change workspace when the engine permits it
`/resume`	Retry immediately during a network pause
`/exit`	Detach the interface; agent can keep running
`/quit`	Stop the engine process
`/new`	Ask the client to start a fresh session; the browser provides launch instructions

Output and exit codes

The final answer goes to stdout; progress and diagnostics go to stderr. --json and --output change the stdout format.

Code	Meaning
0	Successful CLI completion
1	Task or execution failure
2	Usage or configuration error
3	Session already owned and cannot be attached
130	Interrupted

For automated business workflows, independently verify the intended outcome. A known remote/background completion limitation in this preview means CLI success is not a sufficient business completion signal.