panel-node-pack-sync

Keep the panel node-pack in sync with the orchestrator

The orchestrator (comfyui-mcp, from npm) and the sidebar panel (comfyui-agent-panel on the Comfy Registry, repo comfyui-mcp-panel) ship separately. Updating one does not update the other. A new orchestrator driving an old panel fails in confusing ways — a bridge command the panel simply doesn't implement, a feature that exists in the docs but not in the sidebar — and those failures are hard for a user to diagnose. This skill closes that gap.

The two rules that outrank everything else here

1. Never report a sync that did not happen. This whole feature sits on top of the fabricate-success fixes (#639/#641): ComfyUI-Manager reports its queue "drained" even when it never enqueued anything, and a .bak-style copy in custom_nodes can shadow the real panel in the browser. So the only version you may ever tell the user is the one read back from disk after the fact. If the tool throws, the sync FAILED — say so plainly. Do not soften it, do not retry it into a success, do not report the version you intended to install.

2. Never move a pinned user. A pin is a promise. If the user pinned the panel, you warn and stop. You do not unpin for them, you do not "just this once", you do not sync anyway because the new version is obviously better. Offer to clear the pin, and act only if they say yes.

Step 1 — Look before you touch anything

install_panel(action='status')

This never errors. Read these fields:

Do not compute the comparison yourself. sync.decision already accounts for the pin, shadow copies, dev symlinks, remote/cloud mode, and unreadable versions.

Step 2 — Act on sync.decision

Step 3 — Sync (decision: "sync")

install_panel(action='sync')

That single call re-checks the decision at execution time (the pin may have been set a second ago), runs the update through the hardened, verified path, and re-reads the pack from disk afterwards. Read the result:

• synced: true → it really moved. Report verifiedVersion — that is the version observed on disk after the op, not the one we asked for. Then tell the user ComfyUI must be RESTARTED to load it (restartRequired: true); this never auto-restarts. Now read stillBehind, which is tri-state:
  • false → the panel provably meets what the orchestrator needs. Done.
  • true → the update applied but did not close the gap. Say so; do not round it up to "you're current now".
  • null → it landed, but the resulting version (e.g. nightly) can't be compared, so whether the mismatch is fixed is unknown. Say exactly that. null is not false — never report it as "you're fine".
• synced: false → nothing was changed. decision says why (pinned-warn, up-to-date, blocked, …). This is a normal outcome, not a failure.
• The tool errored → the sync FAILED. The error text names the cause (ComfyUI-Manager's stale-3.x silent no-op, a shadow copy, an unverifiable post-state) and the fix. Relay it. Never describe a failed sync as "completed with warnings" or "probably fine after a restart".

For a user who is on Comfy Desktop, restart via the Manager reboot endpoint rather than killing the process.

Step 4 — Pinned (decision: "pinned-warn") — WARN, DO NOT SYNC

The user deliberately held the panel where it is. Tell them three things and then stop:

1. A newer panel that matches their orchestrator exists (requiredPanelVersion).
2. They are pinned (say to what, and where the pin lives — pin.source).
3. How to get off it, if they want to.

Your orchestrator (comfyui-mcp 0.48.32) expects panel 0.11.28+, and you're on 0.11.3 — but you've pinned the panel to 0.11.3, so I haven't changed anything. Want me to clear the pin and update? I'd unpin and then sync; ComfyUI needs a restart afterwards.

Only if they say yes:

install_panel(action='unpin')      # clears the persisted pin
install_panel(action='sync')       # then Step 3

If pin.source is "env", unpin cannot clear it — the pin comes from the COMFYUI_MCP_PANEL_PIN environment variable. Tell the user to unset it (or set it to off) in their environment or ~/.comfyui-mcp/.env and restart the orchestrator. Do not edit their environment for them, and do not report them as unpinned — install_panel(action='unpin') returns the still-active pin in that case, and its note says exactly this.

Step 5 — Blocked

• Shadow copy (shadows non-empty): a dir like .comfyui-agent-panel.bak-0.11.28 in custom_nodes is also served as a web extension, and a dot-prefixed name wins by sort order — so the browser may be loading the old panel no matter what the disk says. Nothing can be verified until it's gone. Tell the user to move it out of custom_nodes (not just rename it) and hard-refresh the ComfyUI tab, then re-run Step 1.
• Unreadable pin: ~/.comfyui-mcp/panel-settings.json exists but couldn't be parsed, so we cannot prove the user isn't pinned — and we refuse to move them on a guess. Ask them to fix or delete that file (or set COMFYUI_MCP_PANEL_PIN=off), then re-run Step 1.

Pinning on request

If the user wants to stay on their current panel (they're mid-project, a newer panel regressed something, they're testing):

install_panel(action='pin', version='<installedVersion from status>', reason='<why>')

version is required — never invent one. Pass the installedVersion from Step 1 to pin them where they already are. A pin records intent only: it does not change what is installed.

While it's set, everything that could move the panel refuses — not just install_panel. The panel is an ordinary custom node pack, so the generic node tools are a second door into the same operation, and they are guarded too:

• install_custom_node / update_custom_node / reinstall_custom_node targeting the panel by any spelling — the registry id, the repo name, or a git URL including ref-carrying forms like …/[email protected] and …/comfyui-mcp-panel/tree/main. These also route through the verified path automatically, so the version they report is re-read from disk like sync's.
• id="all", and update_all — a bulk update moves the panel along with everything else. ComfyUI-Manager can't update everything-except-one-pack, so while pinned these refuse outright. If the user wants the rest updated, either unpin first or update the other packs individually by id. Say that plainly rather than quietly unpinning to make all work.
• fix_custom_node, panel_install_node and panel_update_node refuse a panel target outright — pinned or not. They report success as soon as the ComfyUI-Manager queue drains, which proves nothing, and there's no verified equivalent to route them into. Use install_panel instead; don't work around the refusal.

Prefer install_panel throughout — it's the one with status, sync and the pin.

When to run this at all

• Right after the orchestrator updates (self_update, a fresh npm i -g, or a version in the ENVIRONMENT line that's newer than last you saw).
• When a panel/bridge command fails in a way that smells like version drift — "panel is too old", a graph_*/ui_* command the panel doesn't implement, a documented sidebar feature that isn't there.
• Whenever the user asks to update, pin, or unpin the panel.

Be proportionate: this is a one-line check. If decision is up-to-date and the user didn't ask, don't narrate it — just carry on with what they actually wanted.

Absolute rules

• A sync that didn't move bytes is a FAILURE. Report the thrown error, never a success.
• Report verifiedVersion (re-read from disk), never the target version, never nightly.
• A pin is never overridden, not even "temporarily". Unpin requires the user's explicit yes.
• Never touch a dev symlink — it's someone's working repo.
• Always say a restart is required after a sync lands; nothing here auto-restarts ComfyUI.