feat(swift-ios): choose the reasoning level from the composer by saphid · Pull Request #7344 · pingdotgg/t3code

saphid · 2026-08-17T22:41:04Z

Important

Current delivery status: blocked. Theo's SwiftUI target advanced to c49bcc5d8a56324eb0ad9cf1b19d5dcc6ca0caa0 with overlapping model-option work after this PR's proof and checks were captured. The head remains textually mergeable, but the existing CI, tests, and media are bound to the earlier target and must not be treated as exact-live-base validation.

Upstream delivery

Delivery: blocked
Validated against Theo commit: c49bcc5d8a56324eb0ad9cf1b19d5dcc6ca0caa0
Depends on: none
Merge order: this PR only, after refreshed live-base validation
Validation status: current head 2524de5c55914a9364c56f1dcff7c5953b74ea98 has green required CI and successful CodeRabbit, Cursor Bugbot, and Macroscope correctness reviews with no unresolved threads. A merge-tree simulation against the live target exited 0, but Theo's new picker-level model options overlap this composer-level trait UI semantically. Focused native tests, Simulator proof, and AI review must be repeated on the integrated result before human merge review.
Dark mode evidence: yes
Clean screenshot: https://github.com/user-attachments/assets/eee3fbae-456d-4253-a8af-0ff45b7ee377
Annotated screenshot: https://raw.githubusercontent.com/saphid/t3code/6dcfcf115e9cea4ac2aee0464bc89c23db520f6f/proof/pr-7344/2524de5c/traits-interaction.gif
Clean video: https://github.com/user-attachments/assets/b17c2f48-294f-46da-ad87-9da57ba01e12
Annotated video: https://github.com/user-attachments/assets/22c313e1-94a5-400c-b124-4b6dedb07d49

Problem

The native SwiftUI composer exposed only Codex reasoning, while the Electron composer renders the selected model's full option-descriptor catalog. That left Service Tier unavailable and made the native behavior diverge as providers and models changed.

Change

Render generic descriptor-backed trait sections in descriptor order from the selected model; absent and unsupported sections stay hidden.
Match the Electron treatment for current and default choices, descriptions, and the compact trigger.
Route every choice through the existing provider/model selection and turn-dispatch path.
Preserve provider/model changes, inherited selections, default materialization, and prompt-injected hidden effective values when a neighboring trait changes.
Add focused coverage for reasoning, service tier, descriptor order, trigger text, default/current states, unsupported descriptors, hidden prompt-injected values, and persisted effective dispatch.

For the live Codex descriptor, the menu shows Reasoning followed by Service Tier, with Standard marked Default. The backend currently supplies Fast's description as 1.5x speed, increased usage without terminal punctuation; SwiftUI renders descriptor text verbatim. The focused fixture also covers the supplied punctuated form.

Verification

Exact base: b67837984e5650888b1eed0e1cd7fff83625f696
Exact head: 2524de5c55914a9364c56f1dcff7c5953b74ea98
Base is an ancestor; GitHub reports MERGEABLE / CLEAN with no conflicts.
Focused native tests: 32 passed, 0 failed, 0 skipped.
swiftc -parse passed.
git diff --check passed.
CI at exact head: all required checks passed; intentional preview/static-analysis jobs are skipped and Macroscope Approvability is neutral.
All four published review threads are resolved; the three current findings were addressed in 46b2288 and 2524de5.
Direct Claude Opus high review was attempted read-only on the frozen change. The process exited 1 with HTTP 429, zero tokens, and no model usage, so no Claude model participated and cross-provider review was skipped because capacity was exhausted.

Exact-head proof

Built and installed only on the leased canonical proof simulator from exact head. Installed executable SHA-256: c68c7d9d1ba53b96fa54347f29d84db26765b95e00e9bfb5dcaca3c3a1e5b2d9. Exact-base executable SHA-256: 0ea482ec8e37519c86fa5529ac541b605e9e01f286e336039ec0ebf3ef971cf2.

Dark mode — exact base → exact head

Interaction

Light mode — exact base → exact head

Raw interaction capture: 76.758 seconds, SHA-256 1f14953d7c351a31498a776a9daef347a058258af7b3eed27dfb234a4a1fdbf4. Complete timeline validation passed all 5 declared actions. Every published attachment is anonymously readable; both MP4s return video/mp4, nonzero full bodies, and 206 byte ranges. The clean video rendered successfully in the shared browser.

Owning lane issue: saphid/t3code-personal#110.

Built with GPT-5.6 Sol via Codex in T3 Code.

coderabbitai · 2026-08-17T22:41:14Z

Important

Review skipped

Auto reviews are disabled on this repository. Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: Repository UI

Review profile: CHILL

Plan: Pro Plus

Run ID: 0fda1ea5-6762-467b-a742-8559ba6ed64a

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

🔍 Trigger review

Warning

Your free Security trial is over. An organization admin can activate Security or dismiss this notice.

_{Comment @coderabbitai help to get the list of available commands.}

macroscopeapp · 2026-08-17T22:48:25Z

Approvability

Verdict: Not approved

Macroscope's review found this PR not approvable — The PR introduces a new descriptor-backed traits picker in the iOS composer and changes which model-option values can be selected and sent on existing message flows. Its scope is contained, but the new production workflow and cross-layer selection behavior warrant human review.

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

t3dotgg · 2026-08-19T08:42:04Z

Need this updated and also screenshots/vids

saphid · 2026-08-20T06:47:00Z

@t3dotgg Updated: the branch is now on the current t3code/rebuild-mobile-app-swift base (961a3ae), with the final selector changes at 6cf8887; GitHub reports it 0 commits behind and mergeable. The requested media is still pending: the required canonical proof simulator lock (~/.t3/locks/swiftui-proof-sim.lock) is currently held, so I did not disturb its owner. We still need both light- and dark-mode captures plus a short selector video, and I’ll attach them in a follow-up when that simulator is available.

The composer's reasoning level was read-only, so changing it meant opening the model picker and reconfiguring the model. Make it an inline selector instead. The levels are read from the selected model's own option descriptor, so the client never assumes which levels a provider exposes, and choosing one writes the same selection shape the model picker's configuration screen writes through the composer's existing selection binding. Descriptors that are not a choice list, and values the model does not offer, keep the previous read-only and hidden behavior.

On-device review of the selector asked for two changes. The two prompt-tier "ultra" levels are not wanted in the composer, so the control excludes exactly the `ultracode` and `ultrathink` choice ids and documents that as the single product rule in an otherwise descriptor-driven control. A level excluded here is still displayed while it is the effective one, so the composer keeps telling the truth about a setting made elsewhere, but it is never offered or checked. The composer sits at the bottom of the screen, where an adaptive menu flips its contents and puts the highest level nearest the finger. Pin the menu to the descriptor's declared order so the lowest level reads first.

saphid · 2026-08-21T02:22:45Z

@t3dotgg Updated again: I rebased the branch onto the current t3code/rebuild-mobile-app-swift head (c2276320) and pushed exact head b849a8800. The PR body now contains exact-head light and dark screenshots plus a short interaction recording (Low → High, then reopen with High checked). Focused FeatureComposerPowerTests are 25/25 passing, and the exact-head SwiftUI simulator build succeeded. I also fixed the selector label to meet the 44-point minimum tap target in b849a8800. GitHub reports the PR mergeable; no inline human review threads currently exist.

…bile-app-swift' into feat/issue110-reasoning-selector

saphid · 2026-08-30T04:04:08Z

@t3dotgg Updated against the live SwiftUI target and refreshed the requested proof.

PR head: 696f9d428e5069da90e5d758a45017d1c9b0d409
live target: b67837984e5650888b1eed0e1cd7fff83625f696 (ancestor of the PR head)
exact-head dark-first and light before/after screenshots plus comparison GIFs are now in the PR body
exact-head interaction GIF plus clean and annotated 23-second MP4s are linked in the PR body
the installed simulator app reported the exact PR SHA and its executable matched the preserved build byte-for-byte
the three-action proof packet passed validation; current CI native tests and the main checks are green, with Cursor's head rerun still pending

The prior b849a880 media has been replaced rather than presented as current proof.

cursor

Cursor Bugbot has reviewed your changes using high effort and found 1 potential issue.

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

^{Reviewed by Cursor Bugbot for commit 46b2288. Configure here.}

saphid · 2026-08-30T05:54:44Z

@t3dotgg Updated to the superseding acceptance target at exact head 2524de5c55914a9364c56f1dcff7c5953b74ea98.

The SwiftUI composer now renders every supported model-option descriptor section in descriptor order. The Codex proof shows Reasoning followed by Service Tier, Standard as Default, Fast with the live descriptor text, and retained/effective selections through the existing turn-dispatch path. Unsupported sections stay hidden; provider/model changes, inherited values, prompt-injected hidden values, and default materialization are covered by focused tests.

Focused verification: 32 tests passed, 0 failed; swiftc -parse and git diff --check passed. Exact base b67837984e5650888b1eed0e1cd7fff83625f696 remains an ancestor. GitHub reports no conflicts.