CLI reference

Parsed in packages/cli/src/index.ts and parse.ts. Unknown --flags are ignored (not an error).

Interactive (default)

No subcommand → start or attach, then TUI or REPL.

zox
zox --model anthropic/claude-sonnet-4-20250514 --agent plan
zox --workspace ~/src/app --sandbox host
zox --url http://127.0.0.1:8787 --token "$ZOXX_SERVER_TOKEN"
zox --session sess_...
zox --no-tui --port 9000

Attach requires --token or ZOXX_SERVER_TOKEN.

Command	Purpose
`zox`	Embedded server + TUI/REPL
`zox --url http://127.0.0.1:8787 --token …`	Attach TUI to an existing server
`zox serve`	Headless server (prints token and URL on stderr)
`zox agent run <workspace> <task>`	One-shot task; exits non-zero if the prompt is blocked
`zox export session <id>`	Export a session (`--include-memory` optional)
`zox hooks trust`	Trust this project’s hooks
`zox eval run`	Fixture evals (default: mock tasks)
`zox eval private`	Hidden-grader private suite
`zox eval swe-lite`	SWE-bench Lite subset (Docker)
`zox eval terminal-bench`	Terminal-Bench 2.0 (Docker)
`zox eval archive`	Archive last eval artifacts

`zox serve`

Headless server for CI, remote machines, or a second TUI.

zox serve --port 8787 --sandbox worktree

Stderr prints the token (if generated) and http://127.0.0.1:<port>. Bind is loopback.

`zox agent run`

One-shot: create a session, send <task>, wait, exit.

zox agent run /path/to/repo "Add a --dry-run flag to the CLI"
zox agent run . "Fix the failing test" --model openai/gpt-4.1 --auto-approve

Exit code 1 if the prompt is denied by the guardrail (or equivalent block). --auto-approve approves tools, not Jev prompt ask.

`zox export session`

zox export session <sessionId>
zox export session <sessionId> --include-memory

Exports are redaction-aware; do not assume secrets never appear in raw transcripts you copy by hand.

`zox hooks trust`

Project hooks only run if the project is trusted.

zox hooks trust --workspace .

Global flags

Flag	Type	Notes
`--model`	string	`provider/model`
`--agent`	string	e.g. `build`, `plan`
`--workspace`	path	Session root
`--sandbox`	enum	`host` `worktree` `container` `remote`
`--url`	URL	Skip embed; attach
`--token`	string	Bearer for server
`--port`	number	Default 8787
`--session`	id	Resume
`--no-tui`	bool	REPL
`--include-memory`	bool	Export
`--keep-worktree`	bool	Agent-run lifecycle
`--auto-approve`	bool	Tool permissions only
`--max-turns`	number	Eval / agent-run
`--timeout-ms`	number	Eval

Eval and benchmark flags

These apply to eval subcommands (see Evals):

Flag	Type	Notes
`--limit`	number	Bench subset size
`--k` / `--n-attempts`	number	Trial count
`--skip-eval`	bool	SWE-lite: write predictions only
`--instance-id`	string	Repeatable; SWE-lite
`--task`	string	Repeatable; Terminal-Bench
`--jobs-dir`	path	Terminal-Bench jobs

In-session slash commands

Built-in: /help /model /agent /compact /context /usage /clear /mcp /skill /skills /cancel /sandbox /trace /remember /revert /exit

Details: Slash commands.

Project-defined commands: .zox/commands/<name>.md (see examples/commands/explain.md in the repo).