-V, --version | output the version number |
-p, --print [prompt] | run the prompt without a terminal UI and print the result; text piped on standard input is given to the run after the prompt, and is the prompt when -p has none |
--output-format <format> | output format of --print and attach; values: text, json, stream-json; default: "text" |
--permission-mode <mode> | permission mode of the session; values: manual, acceptEdits, plan, auto, bypassPermissions |
--model <model> | the model of the session, for example devstral-small-2 (momonga models lists them; /model switches it in an interactive session) |
--effort <level> | the reasoning effort the session asks of its model, sent only to a model that declares a reasoning control and moved to the nearest level it accepts (/effort changes it in an interactive session); values: none, minimal, low, medium, high, xhigh |
--dangerously-skip-permissions | run in bypassPermissions mode |
--sandbox-network <setting> | the network of a sandboxed command in this session: deny, ask or allow; wins over the setting sandboxNetwork; values: deny, ask, allow |
--skill <name> | run the prompt with this skill activated |
--image <path> | attach an image to the prompt; repeat for several (needs a model that declares image input) |
--until <command> | a shell command that must exit 0 before the run counts as complete; repeat for several. The runtime runs it after the model stops, never the model, so it must only check and change nothing |
--require <text> | a completion requirement that has no mechanical check; the report lists it as not run; repeat for several |
--mandate <name> | run under the mandate of this name from the user settings (mandates): only what it allows is run, what it asks for is held for your approval while other work goes on, and its budget bounds the run |
--issue <address> | run without a terminal UI to resolve an issue on a declared forge, given by its address as the forge shows it (https://<host>/<owner>/<name>/issues/<number>); needs --mandate; the run ends with a pull request that closes the issue and a comment on it |
--max-cost-usd <usd> | the money limit of the run in US dollars; the run stops before a model request that would pass it |
--verbose | print tool activity to stderr |
--resume [session-id] | reopen a conversation of this folder by its id or a unique prefix of at least 8 characters; without an id, choose from a list |
--continue | reopen the newest conversation of this folder |
--screen-reader | draw the interactive session as plain appended lines for a screen reader |
-h, --help | display help for the command |