stdio

Protocol: stdio://
Default model: None

Key envNone

LicensePro

Hands the navigator’s decisions, or a formatter’s output, to an external agent, a program or a person at a terminal, in place of a model. Each turn ZShot writes a plain-text frame and reads back the agent’s reply.

Endpoint	Channel
`stdio`	Replies on stdin, prompts on stdout
`stdio://3`	One bidirectional file descriptor, such as a `socketpair` the parent passes in
`stdio://?rfd=3&wfd=4`	Separate inherited read and write descriptors
`stdio://?read=/in.fifo&write=/out.fifo`	Named pipes. A plain file gives the reader EOF instead of blocking for the reply

Frames

ZShot writes the rendered prompt, then any screenshot as an <img src="file://…"> line pointing at a per-session temp file, then a <<END>> line. The first frame carries a short protocol preamble. The temp directory is removed when the session ends.

The endpoint’s ?vision= sets when frames carry a screenshot. The agent replies VISION to see the page, and every later frame on that page carries one.

Replies

The agent replies in the navigator action language, ending with a <<END>> line or a blank line:

CLICK @e7, FILL @e7: text, SELECT @e7: Blue, PRESS_KEY Escape, NOTE the total is $42.10, SCROLL DOWN 600 @e12, SCROLL TO @e7, POINTER_DOWN @e5, POINTER_MOVE @e7 TOP, POINTER_MOVE 640:412, POINTER_UP, WAIT 5, READ, VISION, NAVIGATE_TO_TAB 1, ACCEPT, DISMISS, ESCALATE opus, DONE, IMPOSSIBLE, RESTART

An action that names its ref, and its value or options, takes one frame; one that leaves them out gets a follow-up frame asking for them. A fill value stays on one line; write a line break as \n.

Limits

Rejected in server mode. In MCP server mode it must use an inherited descriptor rather than fd 0 or 1.
Bare stdio writes to stdout, so it cannot be combined with -f -.
Exactly one URL.
A navigator and formatter both on stdio share one channel, one atomic exchange at a time.

Options

Option	On stdio
`fd`, `rfd`, `wfd`, `read`, `write`	The channel, as the endpoint table gives it
`vision`	`stall:3` unless set
`timeout`	60 seconds unless set; `0` waits without limit
`frames`	As listed
Navigator options	Every one

Example

CLI Run locally

zshot --navigate "Open the formats page" \
  --navigator-type "stdio" \
  -f shot.png https://zshot-cli.com/