improvement(perf): eight verified cuts to workspace cold-load JavaScript

[waleedlatif1]

Summary

Second round of workspace load-time work. Every item below was proposed by one audit pass and then adversarially re-measured by an independent verification pass before implementation; the bar was strict behaviour preservation — nothing changes what renders, when it first renders, or any data path.

• knowledge/[id] drops the tokenizer (~2.5 MB gzip). The document-list route imported one modal through the [documentId]/components barrel, which also exports the chunk editor — and with it js-tiktoken's BPE rank tables. Deep import; the document detail route that genuinely needs exact token counts is untouched.
• prepareBlockState leaves the persistent shell (~476 KB gzip off nearly every route). It was the only function in stores/workflows/utils.ts needing the block registry and the generated tool-outputs artifact, and utils.ts is reached by the workspace shell through the workflow list hooks — so every route paid for a canvas-drop-only helper, including a module-scope JSON.parse of a 5.4 MB string at shell eval. Pure code motion to stores/workflows/prepare-block-state.ts.
• ExecutionSnapshot is lazy (~7.6 MB of source off logs hydration). Done exactly per the "Code-splitting through barrels" procedure in sim-imports.md: deep React.lazy import, the dead barrel re-export deleted (no sideEffects: false, so a leftover re-export silently defeats the split), log-details' sibling imports deepened to break the parent→child barrel cycle, and a local <Suspense fallback={null}> at both render sites — both of which are gated on client-only state, so there is zero SSR/hydration exposure. Exercised end-to-end: the chunk loads when a log detail opens and the modal renders cleanly.
• Contracts barrel stops re-exporting ./tools, ./selectors, ./v1, ./demo-requests (~58 KB gzip on every route). Zod schemas are constructed at module scope, so export * was defeating tree-shaking; per-integration tool contracts and the v1 admin API were shipping to the workspace home page. Zero importers used the barrel path for any of these — verified, and type-check enforces it.
• Streaming csv-parse moves to a server-only module. Its streaming export is a Node Transform, pinning stream polyfills into client bundles through the lib/table barrel. createCsvParser now lives in lib/table/import-stream.ts, deliberately not barrel-re-exported; both consumers are server-side import paths. The client CSV dialog (dynamic csv-parse/sync) is untouched.
• jszip goes dynamic at its two remaining static sites (skill zip extraction, PPTX parsing) — both already-async, user-triggered functions, mirroring the existing dynamic-import pattern in workflow import-export.
• The desktop local-filesystem tool executor goes dynamic in use-chat — with a real failure path: if the chunk fails to load (deploy skew), an error completion is reported so the server-side tool call settles instead of hanging with the dedupe ref already marked handled.

Measurements

Production build, JS downloaded before the load event, vs the previous release:

route	before	after
/home	4.44 MB	3.87 MB
/logs	4.44 MB	3.64 MB
/knowledge	4.22 MB	3.68 MB
/tables	4.17 MB	3.61 MB
/files	4.68 MB	4.10 MB
/w/[id]	4.80 MB	4.67 MB
/home total incl. idle prefetch	8.15 MB	5.52 MB

The knowledge document-list route additionally sheds the 2.5 MB gzip tokenizer chunk, and the largest shared chunk shrank from 2.06 to 1.57 MB.

Rejected on correctness grounds (for the record)

The same audit rejected several tempting wins: swapping <Home key={chatId}> for a narrower key (leaks a frame of the old conversation plus tool-recovery refs), lazy-mounting the editor panel tabs (makes SSR HTML nondeterministic → guaranteed hydration mismatch), per-viewer code-splitting of chat resource tabs (first-paint change on a persisted-tab path), and staleTimes.dynamic (measured no-op; the segment cache already serves these navigations).

Type of Change

• Performance improvement

Testing

Tested against a production build: all workspace routes render with zero page errors, and the lazy snapshot path was driven end-to-end in a real browser (open log detail → View Snapshot → frozen canvas renders). 1,746 tests pass across every touched area; type-check, lint, all 32 audits, and the block-registry audit pass; the tool-registry boundary baseline is retightened to lock the win in.

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		Aug 23, 2026 4:18am

[cursor]

PR Summary

Medium Risk
Mostly import-graph and code-splitting changes with preserved behavior, but lazy snapshot/tool chunks add new failure paths (chunk load, error boundary remount) that can affect logs and client tool completion.

Overview
Cuts workspace cold-load JavaScript without changing what renders or when. Measured drops are on the order of ~0.5–0.8 MB per route (home idle prefetch 8.15 → 5.52 MB).

Biggest splits: ExecutionSnapshot is now React.lazy on logs (preview canvas ~7.6 MB of source off hydration), with a local SnapshotBoundary so a failed chunk toasts/closes instead of taking down the page. prepareBlockState moves out of stores/workflows/utils.ts so the generated tool-outputs artifact (~476 KB gzip) is no longer on the persistent shell. Knowledge document list deep-imports DocumentTagsModal so js-tiktoken stays off that route.

Barrel / polyfill hygiene: contracts barrel no longer re-exports tools/selectors/v1/demo-requests (module-scope Zod). Streaming csv-parse lives in a server-only import-stream module so Node stream polyfills stay out of client table code. jszip and the local-filesystem chat executor are dynamic imports; the latter reports an error completion if the chunk fails to load.

The tool-registry boundary baseline is retightened to lock the smaller graphs.

^{Reviewed by Cursor Bugbot for commit 6331d1e. Configure here.}

[greptile-apps]

Greptile Summary

This PR reduces workspace cold-load JavaScript by moving heavy dependencies and specialized helpers behind direct or dynamic imports while preserving existing rendering and data behavior.

• Lazily loads execution snapshots, desktop filesystem tooling, JSZip consumers, and server-only streaming CSV parsing.
• Removes unnecessary barrel exports and moves block-state preparation off the persistent workspace shell path.
• Adds containment and fallback behavior for snapshot chunk failures and local-filesystem executor load failures.

Confidence Score: 5/5

The PR appears safe to merge.

No blocking failure remains.

Important Files Changed

Filename	Overview
apps/sim/app/workspace/[workspaceId]/home/hooks/use-chat.ts	Dynamically loads the desktop filesystem executor and now contains failures from both recovery imports and completion reporting.
apps/sim/app/workspace/[workspaceId]/logs/components/log-details/components/execution-snapshot/snapshot-boundary.tsx	Adds an error boundary and loading modal that contain rejected snapshot chunks without replacing the logs route.
apps/sim/app/workspace/[workspaceId]/logs/components/log-details/log-details.tsx	Lazily loads snapshots and remounts their boundary across execution and open-state transitions.
apps/sim/app/workspace/[workspaceId]/logs/logs.tsx	Lazily loads the log-preview snapshot with a contained loading and failure path.
apps/sim/lib/table/import-stream.ts	Isolates the Node streaming CSV parser from client-reachable table exports.
apps/sim/stores/workflows/prepare-block-state.ts	Moves registry-dependent block-state preparation out of the broadly imported workflow utilities module.

_{Reviews (4): Last reviewed commit: "fix(logs): preserve snapshot modal while..." | Re-trigger Greptile}

[waleedlatif1]

@greptile

[waleedlatif1]

@cursor review

[waleedlatif1]

@greptile

[waleedlatif1]

@cursor review

[waleedlatif1]

@greptile

[waleedlatif1]

@cursor review

[cursor]

✅ Bugbot reviewed your changes and found no new issues!

Comment @cursor review or bugbot run to trigger another review on this PR

^{Reviewed by Cursor Bugbot for commit 6331d1e. Configure here.}