[t3dotgg]

The SwiftUI app did a lot of repeated work on hot paths. Each keystroke in a thread re-ran the whole thread view. Each stream event re-parsed tool output, re-folded the full transcript, and decoded JSON twice. Home re-sorted every thread for a pull request cache that nothing read. The app also carried a lot of dead code.

This PR removes that work without changing what users see. Net change is about 380 lines removed (+806 / -1189).

Thread view

• Typing in the composer no longer re-runs the thread view body. A small child view now watches the draft.
• Tool failure and embedded-content flags are computed once per work item. The saved format does not change.
• The turn-folding result is reused until the thread's detail revision, fold state, or feedback changes.
• The transcript skips its workflow scan when nothing it reads has changed.
• Tool icons use a small in-memory cache instead of AsyncImage, so reused cells do not fetch and decode them again.

Home and workspace

• Removed the unused model-level pull request cache and its 30 s poll. It also forced full Home re-sorts.
• Project badge hashing, New Task project grouping, and lastVisitedAt parsing no longer repeat on every render or keystroke.
• The arrange sheet no longer re-diffs on every stream event.

Core and state

• JSONValue decoding no longer runs a re-encode round trip or throws through a chain of DecodingErrors for each node.
• V2 HTTP snapshots are decoded once. V2 control state decodes only the keys it reads.
• Removed a quadratic checkpoint lookup. Lifecycle timestamps are parsed once. Preview text no longer splits and joins the full message.
• Queue commands use the existing tested OrchestrationV2Commands builders.

Other features

• After 512 KB of output, the terminal no longer resets and replays its whole buffer for each chunk.
• Typing a review comment no longer re-diffs the file.

Cleanup: removed unused T3Client RPCs, the dead image preview in Files, production helpers that only tests used, and duplicate provider label and suffix-merge code.

One narrow difference: after a clone reset clears the New Task project and opens the picker, a later snapshot update no longer picks a project behind the picker. This matches the existing rule that an explicit selection never jumps (NewThreadView creationProjectIDs handler). All other New Task selection paths behave the same.

Skipped on purpose: file-backed draft images (the risk is larger than the gain) and lazy diff lines in the PR Files tab (it changes scrolling behavior).

How I checked it

• Full xcodebuild test suite passes after the rebase onto the current SwiftUI branch: 1267 Swift Testing and 510 XCTest. New focused tests cover the work item format after the flag change, JSONValue token decoding, preview text, and terminal buffer trimming. Removed tests that only covered deleted dead code.
• Smoke test on an iPhone simulator, with a signed Debug build paired to an isolated dev server running current main with copied real data. I checked pairing, the Home list and badges, opening and scrolling a long V2 thread with tables, typing, a draft that survives navigation and an app relaunch, and the New Task sheet with its project picker.

Reviewed with sol-loop: 2 rounds with GPT-6.1 Sol on high.

Created with Claude Opus 5.5 in Claude Code. A parallel audit and implementation workflow used Claude Opus 5.5 subagents.

🤖 Generated with Claude Code

[macroscopeapp]

Approvability

Verdict: Not approved

Macroscope's review found this PR not approvable — This is a broad Swift iOS production refactor spanning streaming state, protocol decoding, caches, workspace behavior, thread rendering, and public client APIs. It also changes the New Task project-selection default and pairing-related behavior, so the runtime and user-facing effects require human review.

^{You can add or adjust custom eligibility rules. Learn more.}

[github-actions]

Thread transfer impact

✅ Thread transfer remains within every enforced ceiling.

ℹ️ No successful main baseline artifact is available yet. This run establishes the initial measurement.

Provider	Metric	Main baseline	This PR	Impact	PR ceiling
Codex	Total thread wire	—	4.9 KiB	—	6.8 KiB	✅
Codex	Thread snapshot wire	—	3.8 KiB	—	4.9 KiB	✅
Codex	Live turn WebSocket wire	—	1.2 KiB	—	2.0 KiB	✅
Codex	Live turn WebSocket decoded	—	20.8 KiB	—	29.3 KiB	✅
Codex	Live turn messages	—	1	—	8	✅
Claude	Total thread wire	—	5.0 KiB	—	6.8 KiB	✅
Claude	Thread snapshot wire	—	3.8 KiB	—	4.9 KiB	✅
Claude	Live turn WebSocket wire	—	1.2 KiB	—	2.0 KiB	✅
Claude	Live turn WebSocket decoded	—	21.2 KiB	—	29.3 KiB	✅
Claude	Live turn messages	—	2	—	8	✅

Baseline: unavailable · PR result: 49766f5 · Source CI: success

Scenario and decoded snapshot size

10 historical turns, 5 command tools per turn, 878.9 KiB retained MCP result per historical turn, and a 1.05 MiB retained result in the measured turn.

• Codex decoded thread snapshot: 108.5 KiB
• Claude decoded thread snapshot: 108.8 KiB

Updated in place by a trusted workflow. PR artifacts are strictly validated and never executed.

[t3dotgg]

Note

🤖 Claude Opus 5.5 responding on behalf of Theo

I checked both behavior claims against the diff.

• Pairing: no behavior change. The diff removes PairingURL.pairingURL(fromQRCode:), its two error cases, and PlatformRouteMailbox.peek(). Only tests called them. The app target builds without them. The QR scanner hands its value to ConnectionOnboardingView.applyConnectionString, which never called the removed helper, and PairingService still uses the unchanged PairingURL.resolve.
• New Task: one narrow difference, kept on purpose. requestedInitialProjectID now computes its fallback only while no project is selected and the selection is not explicit. Every path that reads it already requires projectID.isEmpty, so the only case that changes is after a clone reset (explicit selection, empty project, picker open): a later snapshot update no longer picks a project behind the picker. This matches the existing rule in the creationProjectIDs handler (if projectSelectionIsExplicit { return }). The PR description now states this difference.

Agreed that the scope needs human review. Theo will review it before any merge.