`static_capture`¶

Loads each workflow in a real browser and screenshots it without executing. Proves the graph renders and no node comes up red.


Needs	`server` (provided by `registration`)
Default	yes
Fails the run	yes -- but skips gracefully if Playwright is unavailable
Source	`orchestration/levels/static_capture.py`, `reporting/screenshot.py`

The cheapest level that exercises the frontend. Everything below it works on Python objects and HTTP; this is the first one that finds out whether a human opening your workflow sees a usable graph.

How it works¶

For each configured workflow: drive a headless browser to the running server, load the workflow JSON into window.app, wait for the graph to render, tidy the view, and screenshot. No prompt is queued, so nothing executes and no model is loaded.

The capture pipeline is shared with execution and does the same framing work before shooting: fit the graph to view, close open panels and alerts, hide the unsaved-changes dot. Screenshots land in screenshots/<workflow>.png.

What it catches¶

A node that renders red -- registered but the frontend cannot build its widget set, usually an INPUT_TYPES the UI cannot express.
A workflow that references a node your pack no longer ships, after a rename or removal.
Frontend JS that breaks the canvas on load.
Widgets that render wrong -- a combo with no options, a missing default.

None of this is visible to validation, which checks the graph as data rather than as a rendered page.

What it does not catch¶

Nothing runs. A workflow that renders perfectly and fails on the first node is a pass here -- that is execution's job.

Graceful skip¶

If Playwright is not installed the level logs and skips rather than failing. A pack with no workflows configured also logs and returns without error.

No workflows means a silent pass

This level reports success when there is nothing to capture. Unlike execution and execution_light, which now fail on a pack with no workflows, static_capture still passes -- it is a screenshot of the graph, so an empty run is not obviously wrong. Check that your workflows are actually discovered -- see what a pack looks like for the folder names ComfyUI and comfy-test recognise.

Config¶

Key	Effect
`[test] res`	capture resolution (viewport height), default 1080
`[test.workflows] cpu` / `cuda`	which workflows are captured on this backend

In the default set. It needs server, so listing it pulls in install and registration.

It is also one of the four terminal levels: a run ends in exactly one of them, chosen by what [test] levels lists (ADR-0012).

static_capture¶