Shima Workflow Checker

Shima Workflow Checker is a diagnostic node with a personality. Drop it on the canvas and you get a dark little box with a "SCAN WORKFLOW" button. Click it, it scans your whole graph for missing models and broken nodes, then repaints itself - green "LOOKS GOOD!" if the graph passes, or red "ISSUES BELOW:" with a line-by-line report of what's missing. If you've ever loaded someone else's workflow and spent twenty minutes hunting which checkpoint filename is wrong, you already know why this exists.

How it works (and what it doesn't do)

Here's the part people trip on: the Python backend does almost nothing. The node's execute() literally returns the placeholder string "Scan results are displayed in the node's UI widgets." The actual scanning lives in the pack's frontend - js/shima_checker.js registers on the node, draws the chassis and button, and runs a scanGraph() walk of the canvas when you click. So the Workflow Checker is best understood as a frontend tool wearing a node costume, not a backend analysis engine.

Its job: find missing assets (model filenames referenced but not present) and sub-node errors, then render a human-readable report on the node itself. That's the same missing-node pain the whole ecosystem complains about - "feels like I am downloading models and installing missing nodes 90% of the time" - except automated and visible in-graph instead of a wall of console red.

The input that matters

There are none. The checker has zero inputs and one output:

• report (output, STRING) - in practice a stub string; the real report is drawn inside the node's custom UI after a scan

You don't wire it into anything. You just place it, click, and read. If you're the "one checker node per saved workflow" type, drop it near the edge of the canvas so it's visible when you open a downloaded graph.

Installing it

Part of the Shima pack (KDB-USJP/shima_wf). Install via ComfyUI Manager (search "Shima") or:

cd ComfyUI/custom_nodes
git clone https://github.com/KDB-USJP/shima_wf

Then restart. The checker needs the pack's JavaScript (js/shima_checker.js) to load, so if the button never renders, the frontend extension didn't register - a restart usually fixes it, and a hard refresh (Ctrl+Shift+R) fixes the rest.

Gotchas

• The scan is a click-to-run check of your current canvas state, not a live watcher. It won't catch a model you add after the last scan, so re-scan before you trust the green light.
• Because the backend does no real work, the report STRING output won't help you automate anything downstream. If you want machine-readable diagnostics, this isn't the node.
• The visual polish is real, but "LOOKS GOOD!" means "nothing the checker looks for is wrong" - not "this will definitely run." It can't see Python import errors, dependency mismatches, or a VAE that's the wrong one for your model. Treat it as a fast pre-flight, not a guarantee.

It's a nice-to-have rather than essential - but for a pack built around shared, marketplace-sourced islands, a one-click "is anything broken" button is a genuinely thoughtful inclusion. If you build workflows you pass around, keep one in the corner.