feat(workflow-editor): open block palette on edge drag-release with auto-connect

[TheodoreSpeaks]

Summary

• Releasing a connection drag on empty canvas now opens the command palette so you can pick a block to connect to
• The chosen block lands at the drop position and is auto-wired from the source handle
• The palette is restricted to connectable sections only (Blocks, Tools, Tool operations) via a new typed sections allowlist on the search-modal store — triggers/pages/workflows/etc. are hidden since you can't connect into them
• The restriction self-resets, so normal palette opens (toolbar, keyboard, context menu) are unaffected

Type of Change

• New feature

Testing

Tested manually. Lint and check:api-validation:strict pass.

Checklist

• Code follows project style guidelines
• Self-reviewed my changes
• Tests added/updated and passing
• No new warnings introduced
• I confirm that I have read and agree to the terms outlined in the Contributor License Agreement (CLA)

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

1 Skipped Deployment

Project	Deployment	Actions	Updated (UTC)
docs	Skipped		Jul 11, 2026 3:06am

[cursor]

PR Summary

Medium Risk
Changes core canvas connection and block-add flows (edge creation, container boundaries, collaborative adds) without automated test updates noted in the PR.

Overview
Releasing a connection drag on empty canvas (when onConnect did not already complete) now opens the universal search palette instead of doing nothing. The open call passes pendingConnect (source node/handle + screen drop coordinates) and limits sections to blocks, tools, and tool operations.

The search modal store gains typed SearchSection / PendingConnect state; open(options) can set section allowlists and pending connect, and setOpen / close clear those so normal palette opens stay unchanged. The modal only renders groups whose section key is allowed.

Palette picks dispatch add-block-from-toolbar with pendingConnect copied from the store. The workflow canvas treats that as completing the drag: it routes through handleToolbarDrop at the release point with a forcedSource, using resolveEdge so wiring follows the same container-boundary rules as onConnect (including loop/parallel start handles), with presetOperation still applied on drop paths.

^{Reviewed by Cursor Bugbot for commit f5443b7. Bugbot is set up for automated code reviews on this repo. Configure here.}

[greptile-apps]

Greptile Summary

This PR adds edge drag-release block creation in the workflow editor. The main changes are:

• Opens the search palette when a connection is released on empty canvas.
• Limits that palette to connectable block and tool sections.
• Places the selected block at the release point and wires it from the dragged handle.
• Adds one-shot search modal state for section filtering and pending connection context.

Confidence Score: 5/5

This looks safe to merge.

• No blocking issues found in the changed code.

Important Files Changed

Filename	Overview
apps/sim/app/workspace/[workspaceId]/w/[workflowId]/workflow.tsx	Adds pending connection placement and container-aware forced-source wiring for palette selections.
apps/sim/app/workspace/[workspaceId]/w/components/sidebar/components/search-modal/search-modal.tsx	Filters rendered search sections and includes pending connection context on connectable selection events.
apps/sim/stores/modals/search/store.ts	Extends the search modal store with resettable section filtering and pending connection state.
apps/sim/stores/modals/search/types.ts	Adds typed search sections and the pending connection payload shape.

_{Reviews (7): Last reviewed commit: "refactor(workflow-editor): collapse drag..." | Re-trigger Greptile}

[TheodoreSpeaks]

@greptile review

[TheodoreSpeaks]

Thanks — all four findings were legit and are fixed in 8da6b27:

1. Drop coordinate offset (Greptile P1) — confirmed against reactflow v11 source: screenToFlowPosition subtracts the pane rect internally (the deprecated project warning literally says "no need to subtract the react flow bounds anymore"). Removed the .workflow-container bounds subtraction; now passes raw client coords like handleCanvasPointerMove does for the live cursor.
2. Unscoped pending consume (Greptile P1) — the pending ref is now consumed only while the restricted connect-palette it opened is actually up (isOpen && sections), so a toolbar/sidebar/command-list add-block-from-toolbar can't inherit its position/source.
3. Forced edge skips boundary checks (Bugbot) — the source→new-block edge is now forced only when it stays root-to-root (source exists, not inside a container, not a container-start handle), matching onConnect's cross-boundary rejection.
4. Palette drop ignores container placement (Bugbot) — releases that land inside a loop/parallel container now delegate to the container-aware handleToolbarDrop (parenting, clamping, in-container auto-connect). The cross-boundary forced edge is intentionally dropped there.

Minor known limitation from (4): a tool operation released directly inside a container loses its preset operation (the container path doesn't thread presetOperation) — niche, and the block still adds correctly.

@greptile review

[TheodoreSpeaks]

Round 2 addressed in 7a5f66b. Both remaining findings shared one root cause — I was correlating via ambient store state (isOpen && sections) instead of the selection itself. Replaced that with an explicit correlation token:

• onConnectEnd generates a token, stores it on pendingConnectRef, and passes it to open({ sections, connectToken }).
• The palette stamps the store's connectToken onto its add-block-from-toolbar event (both block and tool-operation selections).
• handleAddBlockFromToolbar consumes the pending connect only when the event's token matches the pending token.

This resolves:

• "Pending add still unscoped" / "Unrelated add-block steals pending" (Greptile + Bugbot) — a toolbar/sidebar/command-list event carries no matching token, so it can never inherit the pending position/source, even while the modal is open.
• "Sections reset drops connect placement" (Bugbot, store.ts) — a plain open() now also clears connectToken, so a later pick can't force-wire a stale source; worst case the connect intent is safely dropped (no wrong edge), never misapplied.

The two Cursor comments still anchored to 31ae9f8 (forced-edge boundary + container placement) were fixed in the prior commit 8da6b27 and are outdated.

@greptile review

[TheodoreSpeaks]

Fixed in 2d5ce1e — the last real item ("Tool preset dropped in container"). Threaded presetOperation through handleToolbarDrop (optional param, additive for existing callers) and pass it in the in-container delegation, so a tool-operation pick released inside a loop/parallel keeps its chosen operation. No more known limitations on this flow.

The other Cursor comments still anchored to 31ae9f8 (forced-edge boundary + container placement) are outdated — fixed in 8da6b27.

@greptile review

[cursor]

Cursor Bugbot has reviewed your changes using default effort and found 2 potential issues.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit 2d5ce1e. Configure here.}

[TheodoreSpeaks]

Round 4 in 40347c4.

"In-container connect wrong edge" (Medium, real) — fixed by unifying the edge logic. handleToolbarDrop now takes an optional forcedSource; when present it wires from that exact handle iff it stays within the resolved container context (same-parent, or a container-start handle into its own child — matching onConnect's boundary rules), otherwise no edge. The drag-release path now delegates entirely to handleToolbarDrop({ ..., forcedSource }) for both root and in-container drops, so:

• edges always come from the handle you actually dragged (no more nearest-child guessing),
• container placement/clamp/parenting is correct,
• cross-boundary drags produce no edge,
• presetOperation is preserved.

This also let me delete the bespoke root-vs-container branching in handleAddBlockFromToolbar (net simpler).

"Pending connect cleared too early" (Low) — acknowledged, leaving as-is: the palette closes itself on any pick and the close subscription clears the pending ref regardless of success, so deferring the clear wouldn't preserve intent across a rejection. The only rejection reachable from this restricted palette is a single-instance-duplicate block (triggers are excluded from it), which is rare and degrades gracefully to re-dragging.

The three Cursor comments anchored to 31ae9f8/7a5f66b (forced-edge boundary, container placement, tool preset) are outdated — fixed in 8da6b27/2d5ce1e.

@greptile review

[TheodoreSpeaks]

Simplification pass (f5443b7): collapsed the drag-release correlation machinery — the module ref, the generated connectToken, and the subscribe cleanup effect — into a single pendingConnect field on the search-modal store. The palette stamps it onto its selection event, so its presence is the correlation (no token to match); the store's own open/close lifecycle clears it (no cleanup effect). Net −25 lines, behavior unchanged. @greptile review