[only21mil]

Problem

In the mobile terminal, tapping Ctrl and then an arrow sends a plain arrow and clears the modifier. Alt, shown as Cmd for macOS hosts, sends an extra escape byte instead of a modified cursor sequence. Word navigation and modified arrows in terminal applications therefore receive the wrong input.

Change

Encode the toolbar's four arrow sequences with cursor modifier 5 for Ctrl or 3 for Meta. Keep the existing paste-chord and character handling. The change is in the input helper shared by iOS and Android.

Scope and approval

This qualifies as a very small, focused fix of an obvious bug without a prior issue. The toolbar already offers these modifiers and arrows; the defect is the bytes emitted when they are combined. The change adds one four-line branch to the input helper and focused cases for all four directions with both modifiers.

Prior PRs: #10699 by sameerr03 included this encoding fix alongside toolbar visibility, repeat, and other input changes. The author closed it after noting that the arrows already existed and withdrew the additional input fixes. This candidate reuses the modified-cursor encoding approach in a focused fix. No maintainer reviewed or rejected that specific fix.

Verification

• vp test run --project unit src/features/terminal/terminalInput.test.ts src/features/terminal/terminalPaste.test.ts, from apps/mobile: the new arrow cases fail against the base helper, with eight failures and 17 passes. With the fix, all 25 tests pass.
• pnpm --dir apps/mobile run typecheck: passed.
• Scoped vp lint --threads=4 --report-unused-disable-directives and vp fmt --threads=4 --check on the two changed files: passed.
• Direct assertions importing the actual helper also fail on all eight modified arrows on the base and pass with this change.

Focused unit tests with the base production helper and new tests: eight failures and 17 passes. The candidate passes all 25 tests.

Native comparison passed on the iOS 26.5 simulator and the Android 16 emulator, base fb990765a4 versus candidate 68923a2e57. In bash, type echo one two three, tap the toolbar modifier and Left, insert X, then press Enter. Meta is labelled Cmd for these macOS test backends.

Platform	Chord	Base output	Candidate output
iOS	Ctrl+Left	one two threXe	one two Xthree
iOS	Cmd+Left	one two three[DX	one two Xthree
Android	Ctrl+Left	one two threXe	one two Xthree
Android	Cmd+Left	one two three[DX	one two Xthree

The Android backend's bash 3.2.57 did not bind the modified cursor sequences by default. Both Android revisions used identical session-only bindings for "\e[1;5D": backward-word and "\e[1;3D": backward-word. Without those bindings, bash 3.2 leaves ;5D or ;3D in the line, as it does for the same keys from any xterm-compatible terminal. Readline 8 binds both sequences to word movement by default. iOS used the backend's default bindings.

iOS base: Ctrl+Left and Meta+Left both fail to move by a word.

iOS candidate: Ctrl+Left and Meta+Left both move by a word.

Android base: both modifiers fail with the same session-only readline bindings.

Android candidate: both modifiers move by a word with the same session-only readline bindings.

The Android byte check confirmed that Ctrl+Left delivered ESC[1;5D and Meta+Left delivered ESC[1;3D.

Device checks cover Left with both toolbar modifiers. Focused unit tests cover all four arrow directions.

Not checked:

• Up, Down and Right on devices (unit tests cover all four directions).
• Shells other than bash, and terminal applications that read modified arrows.

Production fix, tests, PR description and images written with GPT-6.1 Sol in T3 Code Nightly.

🤖 Generated with GPT-6.1 Sol in T3 Code Nightly

[macroscopeapp]

Approvability

Verdict: Needs Human Approval

Macroscope's review found this PR requires human approval — This is a focused mobile terminal bug fix that adds modified escape-sequence handling for all four toolbar arrows, with corresponding tests. It also introduces a line-level ESLint suppression for the control-regex rule, which requires human review under the applicable policy.

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

[coderabbitai]

Warning

Review limit reached

Only developers with an assigned seat can use this organization's usage-based review budget, and seats here are assigned manually. Ask an admin to assign a seat, or change the review continuation mode in Billing.

Next included review available in 11 seconds.

Check out review usage here.

View limit details

Limit details: You’ve used all 10 included reviews currently available.

Learn how review limits work.

Review configuration:

Run configuration

• Source mode: Auto
• Repository: pingdotgg/t3code
• Branch: main
• File: .coderabbit.config.ts
• Commit: f279070
• Layers: Repository TS config
• Review profile: CHILL
• Plan: Advanced
• Run ID: e068832a-159a-43e7-beae-85049675e3b4

Commits

Reviewing files that changed from the base of the PR and between 68923a2 and d2691f0.

Files selected for processing: 2

• apps/mobile/src/features/terminal/terminalInput.test.ts
• apps/mobile/src/features/terminal/terminalInput.ts

No actionable comments were generated in the recent review.

ℹ️ Recent review info

⚙️ Run configuration

• Configuration used: Path: .coderabbit.config.ts
• Review profile: CHILL
• Plan: Advanced
• Run ID: 15ffe568-dd17-46e7-add6-35a3ebac2210

📥 Commits

Reviewing files that changed from the base of the PR and between 8866e8f and 68923a2.

📒 Files selected for processing: 2

• apps/mobile/src/features/terminal/terminalInput.test.ts
• apps/mobile/src/features/terminal/terminalInput.ts

Included review availability: This review used your included allowance. Your plan provides up to 10 included reviews per hour; 5 remain after this review.

Walkthrough

resolveModifiedTerminalInput now converts Linux Ctrl- and Meta-arrow escape sequences into modified terminal sequences. Tests cover all four arrow directions.

Changes: Terminal Arrow Input

Layer / File(s)	Summary
Encode modified arrow-key input apps/mobile/src/features/terminal/terminalInput.ts, apps/mobile/src/features/terminal/terminalInput.test.ts	The resolver converts arrow-key escape sequences to Ctrl- or Meta-modified sequences. Parameterized tests cover directions A through D for both modifiers.

Priority: Low

Estimated code review effort: 2 (Simple) | ~8 minutes

Change: Bug fix

Merge Risk: Minimal · up to 68923

Modified arrow input is encoded correctly for the supported toolbar controls, with no established merge-blocking risk.

Architecture risk: Low · up to 68923

The change affects 1 system.

Changed systems: apps/mobile

Architecture concerns
No architecture-level concerns identified.

Review details

Systems and components

• observed — apps/mobile (service) was modified; 2 changed files map to changed impact.

Before / after behavior

• observed — Modified behavior in apps/mobile/src/features/terminal/terminalInput.test.ts: Added parameterized tests confirming Linux ctrl+arrow sequences are converted to ESC[1;5<arrow> write events for arrows A through D.
• observed — Modified behavior in apps/mobile/src/features/terminal/terminalInput.test.ts: Added parameterized tests confirming Linux meta+arrow sequences are converted to ESC[1;3<arrow> write events for arrows A through D.
• observed — Modified behavior in apps/mobile/src/features/terminal/terminalInput.ts: Added arrow-key handling in resolveModifiedTerminalInput: when input data is exactly an ANSI escape sequence for an arrow key, returns a write containing the corresponding Ctrl- or Meta-modified escape sequence; other input continues through the existing logic.

Pre-merge checks: 4

Check name	Status	Explanation
Title check		The title clearly and concisely describes the main change: applying terminal toolbar modifiers to arrow keys on mobile.
Description check		The description includes all required sections. It explains the problem, implementation, scope rationale, approval context, focused verification, native checks, limitations, and agent disclosure.
Linked Issues check		Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check		Check skipped because no linked issues were found for this pull request.

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create a new PR

• Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

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

[maria-rcks]

Note

Written by gpt-6.1-sol on behalf of Maria

@only21mil, retained your modifier encoding and all eight arrow cases, preserved your commit, and merged current main. Replaced the control-character regex and lint suppression with an exact length/prefix/direction check. Added six fallback cases in the existing test file so other input keeps its prior behavior. The final diff against main is two files, 40 added lines.

Verification:

• Current main (f279070244): all eight modified-arrow cases fail; the remaining input and paste cases pass.
• Blacksmith vp run check:changed: format, lint, both knip scopes, mobile/scripts typechecks, and all 21 input tests pass.
• Actual helper through a bash PTY, with default readline bindings: typing echo one two three, applying Ctrl+Left or Meta+Left, inserting X, and Enter produces one two Xthree. Main produces one two threXe and one two three[DX, respectively.
• Two independent code reviews approved d2691f023c, with no findings.

Unverified this run: native iOS/Android toolbar interaction and captures, other shells, and Up/Down/Right through a native device. Chrome cannot exercise the native mobile terminal. The existing PR's device evidence remains contributor-supplied; it was not reproduced here. No client layout or controls changed.

CI is green on d2691f023ce1cb4b4efada0e7e7c1db875c2b360; GitHub reports CLEAN and MERGEABLE. CodeRabbit marked its check successful but rate-limited the updated-head review. The existing Macroscope human-approval note remains; its lint-suppression concern is removed by this commit. Native mobile verification and human approval remain outstanding.

[maria-rcks]

Note

Written by gpt-6.1-sol on behalf of Maria

Verified in a real dev environment

No changed behavior could be exercised in the running app during this Linux Boat VM verification pass. This is an unverifiable result, not a runtime approval.

Read the current diff, description, discussion, commit authors, and checks. The PR is open at d2691f023ce1cb4b4efada0e7e7c1db875c2b360, with no PR commits authored by juliusmarminge or t3dotgg. GitCafe reports 28 successful checks, zero failing and zero pending; these do not prove native behavior.

Source inspection identifies the user path as native mobile thread terminal → Ctrl or Alt/Cmd toolbar button → arrow button → resolveModifiedTerminalInput → terminal write RPC. The web client does not call this helper, so a Chrome session or a direct helper/PTY exercise would not verify the changed native interaction.

Device prerequisite transcript:

T3 device_list: Agent device access is turned off for this environment.
$ adb devices -l
List of devices attached
[no devices]

Not verified

The native iOS/Android toolbar interaction, modifier reset, plain arrows, character chords and paste behavior remain unverified this pass. No main reproduction, PR runtime comparison, real provider turns, screenshots or recordings were produced. The mobile verification skill explicitly requires stopping when device tools or the selected device are unavailable. No worktree installation, dev server, new checks or code changes were started after this prerequisite failed. Existing contributor captures and earlier helper/PTY results were not reproduced and are not evidence from this pass.

To finish verification, enable native device access and provide an Android device/emulator or iOS host. Reproduce on github/main, then test this exact head using matching app-created data, capture the toolbar interaction and neighboring behaviors, and upload the inspected evidence.

[maria-rcks]

Note

Written by gpt-6.1-sol on behalf of Maria

Verified on Android emulator (API 35) against a real dev server

Verified head d2691f023ce1cb4b4efada0e7e7c1db875c2b360 on the t3 x86_64 Android emulator (ro.build.version.sdk=35). Installed this checkout's matching dev client from the shared cache, ran Metro and an isolated vp run dev backend on port 13773, reversed both ports with adb, and paired the native app. No mocks or demo data. The device panel's boot attempt failed, so this pass used the brief's authorized adb/agent-device fallback.

Created a project through the native app and completed real Codex gpt-6.1-sol turns in two threads through the local proxy. Exercised the terminal in the first thread with the Linux backend's default bash/readline bindings:

• Ctrl+Left moved to the start of the last word. Typing X and Enter printed one two Xthree. The recorded repeat with lowercase x printed one two xthree.
• Alt+Left also moved by a word. Recalling the previous command, tapping Alt+Left, then x and Enter printed one two xxthree.
• Plain Up recalled history. Plain Left after the modified arrow moved one character, producing one two xxthrexe; the modifier was consumed. Ctrl+C still sent byte 03 and exited a foreground raw-input process.
• All eight toolbar combinations were driven on the emulator while a real Node process in that terminal printed received bytes. Ctrl Up/Down/Left/Right delivered 1b5b313b3541, 1b5b313b3542, 1b5b313b3544, 1b5b313b3543. Alt delivered 1b5b313b3341, 1b5b313b3342, 1b5b313b3344, 1b5b313b3343. These are the expected ESC[1;5 and ESC[1;3 cursor sequences.

No code changes or pushes were needed. The working tree remains clean. Existing CI has 28 successful contexts in Cafe (the rendered checks view distinguishes 20 successful and 8 skipped), zero failures and zero pending. No new static checks were run for this verification-only pass.

Before the toolbar interaction on this head:

After Ctrl+Left, x, and Enter:

Nine-second recording, inspected after export and checked for playback and seeking after upload:

All eight native toolbar combinations received by the real terminal process:

App-created project and real Codex response:

Not verified

• iOS cannot run on this Linux host.
• The original problem was not reproduced on github/main in this pass. The before/after captures above show the interaction on the PR head, not a main-versus-PR comparison.
• Paste/clipboard chords, other shells, and interactive terminal applications beyond bash and the Node raw-input process were not exercised.