pingdotgg/t3code. All authors. Drafts included. Default branch main. 570 issue assessments and 925 PR assessments. Initial inventory: 570 open issues and 924 open PRs. Current assessed open inventory: 570 issues and 924 PRs. Final reconciliation: 2026-09-01T11:57:41.491617+00:00.
Request. Add slash commands, skills, and file mentions to a new mobile task.
Audit finding. Merged PR #8587 already adds the shared command menu to new-task and existing-thread composers, with project or worktree-scoped search and plan-mode gating. Main also resets the caret when the draft owner changes, but does it after render rather than before render as this PR proposes. The primary feature is handled on React Native Android and iOS, leaving only the draft-switch timing and built-in-command differences to decide.
Recommendation. Keep open: partial fix. Reduce this PR to any reproducible draft-switch or unsupported-command gap left after PR #8587.
Confidence high. Release: In stable source. PR readiness: Needs work or a decision.
Check: PR #7523. Current head ce5d27c2: required checks are absent from canonical latest-commit metadata: Test, Check, Mobile Native Static Analysis, Release Smoke. GitHub reports merge conflicts.
Merged pr: PR #8587. Addresses the part identified in this finding; merge commit e89800895924 was checked against repository release ancestry.
Limits. Required CI checks are not present on the current head in canonical check metadata. Release status refers to repository tag ancestry; mobile app-store and OTA rollout was not verified.
Request. Optionally display used context tokens beside the meter with absolute-token warning colors.
Audit finding. Main still has an icon-only meter and no showContextWindowTokenCount setting. This patch adds fixed warning thresholds at 160,000 and 250,000 tokens across models, which is a product choice separate from simply exposing the count. The percentage proposal is also open, and this branch commits PR-only screenshot files that should not ship.
Recommendation. Keep open: decision needed. Choose the meter presentation and warning policy, then remove PR-only screenshots from the branch.
Confidence high. Release: Not applicable. PR readiness: Needs work or a decision.
Pr: PR #7681. The other open proposal displays percentage instead of optional absolute-token warnings.
Limits. The committed screenshot binaries were not visually inspected. No current-head result for required checks: Test, Check, Mobile Native Static Analysis, Release Smoke.
Request. Switch among recently visited threads with a held-modifier preview and commit on release.
Audit finding. Main offers sidebar-order thread traversal but no recentThreads commands, visit-order store, or preview switcher. This branch contains the still-open modified-Tab recorder prerequisite, and its default chord is reserved by browsers, so the default interaction is desktop-specific. Alternative user bindings are required for web.
Recommendation. Keep open: work remains. Merge the recorder prerequisite, rebase this branch, and verify modifier release and cancellation in desktop.
Confidence high. Release: Not applicable. PR readiness: Needs work or a decision.
Pr: PR #7548. The switcher branch includes this unmerged recorder dependency.
Limits. No current-head result for required checks: Test, Check, Mobile Native Static Analysis, Release Smoke. An older related reference could not be retrieved from GitHub. It is excluded from verified dependencies and fix evidence.
Request. Render streamed model reasoning as bounded Thinking and Thought activities on all clients.
Audit finding. Main now drops non-assistant content deltas before loading a thread, so reasoning remains absent. This patch reuses tool activities across clients, while the other open proposal adds a reasoning message channel and intentionally omits mobile rendering. Neither approach is landed, and the mobile scope prevents treating them as exact duplicates.
Recommendation. Keep open: decision needed. Choose one reasoning representation and rebase it around the current ingestion fast path and OpenCode steering behavior.
Confidence high. Release: Not applicable. PR readiness: Needs work or a decision.
Request. Fix forward lifetime and stale-navigation races in the SSH preview feature branch.
Audit finding. The base SSH preview API is absent from main, so this is a dependent feature fix rather than an obsolete main-branch patch. The reviewed route manager adds generations and defers retired-route cleanup until use ends. Its SSH readiness and shutdown work must remain paired with the forwarding implementation.
Recommendation. Keep open: work remains. Merge the lifecycle corrections into the forwarding stack for one integration review.
Confidence medium. Release: Not applicable. PR readiness: Needs work or a decision.
Observed GitHub metadata. GitHub state OPEN. Author ralphsmith80. Updated 2026-08-20T21:52:36Z. Draft no. Target t3code/add-ssh-preview-port-forwarding. Head c20acc8796c33b1a14052323dbf940fb5faebac8. Branch fix/pr-4039-review-feedback. Size +844 / -171, 12 files, 7 commits. Mergeability MERGEABLE. Merge state UNSTABLE. Target is not main. Metadata snapshot 2026-09-01T11:57:41.491617+00:00.
Checks. Required: Test: not captured, Check: not captured, Mobile Native Static Analysis: not captured, Release Smoke: not captured. Observed: 10 passed, 0 failed, 0 pending, 3 skipped. Checks captured 2026-09-01T11:50:38.676311+00:00.
Reviews. GitHub review decision UNKNOWN. Current-head approvals 0. Current-head change requests 0. Unresolved review threads 0. Counts do not replace review of the findings.
Evidence read. body yes, discussion yes, full diff no, current source yes, history yes.
Pr: PR #4039. This is the active unmerged forwarding prerequisite.
Limits. The 61,997-byte diff was not fully read. Review covered the named feature paths and current main, not every file or test. The collected latest head has no results for these required checks: Test, Check, Mobile Native Static Analysis, Release Smoke.
Request. Add an opt-in desktop sleep assertion that keeps an idle remote host reachable while the app runs.
Audit finding. Main has no idle-host sleep assertion, and the closed active-turn proposal does not provide one. The diff restores and applies the setting in the main process, but its persistence-and-apply path is not serialized and global Restore defaults omits the new setting. Those code gaps remain alongside the product decision about idle power use and the persistent warning.
Recommendation. Keep open: decision needed. Fix persistence/apply ordering and Restore defaults before the sleep-setting product review.
Confidence high. Release: Not applicable. PR readiness: Needs work or a decision.
Pr: PR #5535. Verified closed without merge. It protected local active turns rather than idle remote hosting.
Comment: Discussion comment. Confirmed: the diff does not add the new preference to global reset tracking or its patch.
Limits. Current-head required checks were not reported: Check, Mobile Native Static Analysis, Release Smoke, Test. GitHub reports merge conflicts with main on the collected head.
Request. Exclude disabled provider instances from transcript scanning and usage display.
Audit finding. The merged usage-view fix already filters empty provider rows and charts using actual activity. The server still scans legacy Claude and Codex homes without checking instance enablement, so disabled providers with history can still contribute. The current service also supports Grok, which the older two-provider patch must preserve when rebased.
Recommendation. Keep open: partial fix. Add enabled-instance source resolution for all currently supported usage providers without undoing activity-based display filtering.
Confidence high. Release: In stable source. PR readiness: Needs work or a decision.
Request. Make service install restart and repair even an already-current background service.
Audit finding. Main still returns early when the installed service is current. The proposed change removes that guard for install, update, and shared onboarding reconciliation, so every caller can restart a live service. This is a command-behavior decision, not a fix already supplied by launchd support.
Recommendation. Keep open: decision needed. Decide whether restart-on-every-reconcile should apply to onboarding as well as explicit service install.
Confidence high. Release: Not applicable. PR readiness: Needs work or a decision.
Request. Use recent commit bodies as examples for generated commit-message style.
Audit finding. Main still reads twenty non-merge subjects and has no commit-body examples. Merged PR 8804 now adds local repository instruction files to the same policy, which this older diff does not contain. The feature remains useful but must retain that newer instruction behavior when its consumer-specific examples are integrated.
Recommendation. Keep open: work remains. Rebase the commit-body examples onto the current policy while retaining repository instruction files.
Confidence high. Release: Not applicable. PR readiness: Needs work or a decision.
Request. Update branch status when an external Git process changes HEAD.
Audit finding. Main has no HEAD watcher for external branch changes. The proposal adds one, but its post-retain load can still overwrite a fresher watcher result in the shared cache. The current diff also never resets watcher backoff after a healthy period, matching the two remaining review findings.
Recommendation. Keep open: work remains. Serialize or version the handoff reload so it cannot overwrite a newer watcher result.
Confidence high. Release: Not applicable. PR readiness: Needs work or a decision.
Request. Continue worktree bootstrap from the selected local branch when optional origin preparation fails.
Audit finding. Main falls back when origin is absent, but a failed check, fetch, or remote-ref resolution still aborts bootstrap. Merged draft recovery work in PR 7664 does not change that selection rule. This proposal intentionally starts from potentially stale local history and reports the fallback only in server logs.
Recommendation. Keep open: decision needed. Decide whether Start from origin may silently fall back before accepting the server change.
Confidence high. Release: Not applicable. PR readiness: Needs work or a decision.
Request. Show context usage percentage beside the context ring without opening the popover.
Audit finding. The current meter exposes percentage in its accessible label and popover, not as visible trigger text. The manual-compaction changes kept the same icon-only trigger. The separate token-count proposal is still open and offers a different presentation, so this needs a UI choice.
Recommendation. Keep open: decision needed. Choose the visible percentage or optional token-count presentation for the meter.
Confidence high. Release: Not applicable. PR readiness: Needs work or a decision.
Request. Replace endless mobile file-preview loading with failure, disconnected, and retry states.
Audit finding. Main now has a tagged asset URL state for chat images, but workspace file previews still use the nullable wrapper and treat null as loading. The complete PR adds disconnected and retry handling to those file screens. Its hook return shape differs from the current chat-image hook, so it needs integration with the landed API rather than a blind apply.
Recommendation. Keep open: partial fix. Extend the current asset-state hook without breaking chat callers, then update both file preview components.
Confidence high. Release: In stable source. PR readiness: Needs work or a decision.
Request. Keep finding a merged PR after a branch pushed without upstream configuration is deleted and pruned.
Audit finding. Merged PR 6216 preserves lookup when branch remote and merge configuration survives deletion. Main still treats a branch with neither that configuration nor a remote-tracking ref as unpublished. This PR covers the remaining push-without-upstream case by removing that optimization, at the cost of more empty provider lookups.
Recommendation. Keep open: partial fix. Retain the no-upstream deletion test and decide how to avoid excessive lookup traffic.
Confidence high. Release: In stable source. PR readiness: Needs work or a decision.
Limits. Latest-commit metadata has no required runs for Test, Check, Mobile Native Static Analysis, Release Smoke. GitHub reports conflicts with the target branch.
Request. Refresh an open Git diff when a sibling thread completes work in the same checkout.
Audit finding.PR 8803 now refreshes panels after active-thread mutations, but the mutation ID still comes from the active thread. Main does not observe sibling completions in DiffPanel. This proposal adds a same-checkout tracker, and must be adapted to the newer mutation refresh path rather than restoring the removed active-turn effect.
Recommendation. Keep open: work remains. Integrate sibling checkout events with the current workspace-mutation refresh hook.
Confidence high. Release: Not applicable. PR readiness: Needs work or a decision.
Request. Windows terminal shortcuts and links should reach their intended clipboard, shell, and browser targets.
Audit finding. Merged PR #7678 already supplies textarea priming and correct native copy-event ownership, including IME protection. Main still uses Electron's standard Edit menu and the asynchronous preview-or-browser link choice, so the other parts of this PR remain outstanding. Reapplying its old clipboard code would replace newer safeguards, and its global non-HTTP rejection goes beyond terminal links.
Recommendation. Keep open: partial fix. Split out and review the remaining native accelerator and link-opening changes against the current clipboard code.
Confidence high. Release: In stable source. PR readiness: Needs work or a decision.
Request. Explain rejected Git hooks and retain the failed commit message and selected files.
Audit finding. Main still filters trace2 child-exit records by child_class and clears commit inputs before success. The diff fixes hook exit-code reporting and preserves drafts in web and mobile. Its web completion callbacks can still run after a repository switch, so the new scope-reset effect alone does not prove old requests cannot alter a newer draft.
Recommendation. Keep open: work remains. Add a scope-aware completion guard before reviewing the commit draft changes.
Confidence high. Release: Not applicable. PR readiness: Needs work or a decision.
Request. Show provider compaction progress and results in web and mobile thread views.
Audit finding. Main still maps Claude compaction to waiting and has no persisted compaction detail or start clock. The proposed provider mappings and optional session fields address that gap, but the patch includes the still-open heartbeat dependency and claims migration IDs already used on main. The separate Codex-only and cross-provider proposals are related, not landed replacements.
Recommendation. Keep open: work remains. Resolve the heartbeat dependency and migration IDs before reviewing the complete compaction change.
Confidence medium. Release: Not applicable. PR readiness: Needs work or a decision.
Limits. Provider mappings, lifecycle changes, contracts, and sidebar code were reviewed, but not the complete 38-file diff. Required CI checks are absent at the reviewed head: Test, Check, Mobile Native Static Analysis, Release Smoke. GitHub reports merge conflicts at the reviewed head.
Request. Track viewed PR files across sessions and devices, with provider-specific storage.
Audit finding. Main has no viewed-file capability and still disables Azure diffs. The complete 52-file patch adds viewed tracking, persistence, and a new Azure Code tab. It deliberately leaves Azure context expansion reading the latest iteration even when displayed hunks belong to an older one, an acknowledged correctness gap.
Recommendation. Keep open: decision needed. Split Azure diff and identity changes from viewed-file tracking before maintainer review.
Confidence high. Release: Not applicable. PR readiness: Needs work or a decision.
Comment: Discussion comment. Author leaves the reported diff-expansion version mismatch unfixed. The proposed Azure provider still selects the latest iteration.
Check: GitHub check. Latest-commit checks: Test success, Check success, Mobile Native Static Analysis skipped, Release Smoke success.
Limits. GitHub reports conflicts with the target branch.
Request. Configure desktop composer spellcheck with supported dictionaries or automatic OS and keyboard-language selection.
Audit finding. Main has native spelling suggestions but no spellcheck preference fields, dictionary discovery, or session language configuration. The patch adds those controls and fixes the reported Windows startup delay. Its latest keyboard map still maps the default Canadian ca layout to English, and checkbox labels omit the visible unavailable-dictionary state from their accessible names. These two review findings remain in the diff.
Recommendation. Keep open: work remains. Correct the Canadian layout mapping and accessible unavailable-state labels, then rebase the spellcheck controls.
Confidence high. Release: Not applicable. PR readiness: Needs work or a decision.
Comment: Discussion comment. The Canadian layout finding is still present in the latest mapping.
Comment: Discussion comment. The latest checkbox aria-label still omits unavailable dictionary state.
Limits. The documentation screenshot binary was not visually inspected. No current-head result for required checks: Test, Check, Mobile Native Static Analysis, Release Smoke.
Request. Remove terminal escape sequences from OpenCode inventory and previously stored model options.
Audit finding. Main now loads inventory through HTTP, so fresh CLI inventory corruption no longer reaches the production picker. Stored agent and variant options are still read and sent without terminal cleanup, so existing polluted threads are not repaired. The PR addresses that remaining case but also changes shared string-option and trim helpers for every provider, which needs narrower scope or focused cross-provider tests.
Recommendation. Keep open: partial fix. Retain stored OpenCode option repair and limit or verify the shared string normalization changes.
Confidence high. Release: In stable source. PR readiness: Needs work or a decision.
Merged pr: PR #8480. Removed production CLI inventory calls.
Limits. Canonical latest-commit checks do not include Check, Test, Mobile Native Static Analysis, Release Smoke. GitHub reports a merge conflict with current main.
Request. Show running-terminal status on mobile thread rows and headers without constant animation work.
Audit finding. Main has no running-terminal row indicator. The collected head cdf0285 adds row dots with an endless interpolated 800 ms pulse, not the stepped pulse, terminal glyph, and header coverage promised by the body. Earlier header review threads are resolved because that code is no longer in this diff, so they do not prove the current patch meets the stated scope.
Recommendation. Keep open: work remains. Replace the continuous pulse with an approved low-work indicator and align the body with the actual row and header coverage.
Confidence high. Release: Not applicable. PR readiness: Needs work or a decision.
Request. Restore genuine child liveness when a turn starts before child registration.
Audit finding. Merged PR 7937 correctly stops trailing interacted events from reviving completed children. Main still does not replay an already recorded live child turn when late registration arrives, and does not retain a terminal preregistration error. This residual ordering fix remains needed and overlaps PR 5723's settlement state.
Recommendation. Keep open: partial fix. Combine the early-start and early-settlement requirements into one reviewed child lifecycle change.
Confidence high. Release: In stable source. PR readiness: Needs work or a decision.
Merged pr: PR #7937. fix(server): stop completed Codex threads from staying stuck on working
Limits. Production changes were inspected, but the full integration-test diff was not reviewed. Latest-head required check results are absent for Test, Check, Mobile Native Static Analysis, Release Smoke. GitHub reports a conflict with main. Any passing checks apply to the existing head, not a rebased result.
Request. Collapse the resting desktop composer and move its controls into the thread toolbar.
Audit finding. Main still has only the existing mobile collapse path, not the proposed 50-pixel desktop resting state. The patch changes focus and wheel handling, control portals, image layout, and timeline footer anchoring across 28 files. Recent banner and spacing fixes do not implement this interaction, and the full interaction diff needs review before a merge recommendation.
Recommendation. Keep open: decision needed. Decide on the resting interaction and complete a full review of its focus, scroll, and anchoring changes.
Confidence high. Release: Not applicable. PR readiness: Needs work or a decision.
Commit: Commit 660cddd3bc98. The merged composer spacing fixes are not the resting-state feature.
Check: GitHub check. Current head 7c6a28c passes Test, Check, and Release Smoke. Full interaction review remains outstanding.
Limits. The 28-file patch was inspected for state, scroll, and overlay scope but not reviewed line by line in full. The full historical review discussion was not read; the body, current thread status, and available maintainer direction were checked. Head advanced to 7c6a28c during the audit. The range-diff preserves all 34 feature commits; the merge resolution only removes obsolete client auto-settle reads, but the full new comparison remains unreviewed.
Request. Handle Codex permission-profile approvals in both clients and settle them before Stop.
Audit finding. Merged PR 8058 handles MCP app elicitation, not item/permissions/requestApproval, which remains absent from main. The proposed handler grants a permission profile but forwards only its optional reason, so an approval can omit the capabilities being granted. Its Stop logic also needs review against the later concurrent request-handler and stale-approval recovery changes.
Recommendation. Keep open: partial fix. Expose the actual permission profile in canonical approval details and rebase the Stop path.
Confidence high. Release: In stable source. PR readiness: Needs work or a decision.
Merged pr: PR #8058. fix(codex): show app access approval prompts
Merged pr: PR #5195. fix(server): recover stale Codex approval callbacks
Comment: Discussion comment. Discussion evidence checked against the submitted source change.
Limits. Latest-head required check results are absent for Test, Check, Mobile Native Static Analysis, Release Smoke. GitHub reports a conflict with main. Any passing checks apply to the existing head, not a rebased result.
Request. Add fx as an opt-in ACP provider for Gateway, Codex, and Grok accounts.
Audit finding. Main has no fx driver, and supporting Codex or Grok directly does not supply fx-managed account routing or its ACP model catalog. The inspected change adds a dedicated driver and makes ACP authentication optional for providers that already manage credentials. The author verified only unauthenticated real startup, while generic ACP remains a separate open architecture proposal.
Recommendation. Keep open: decision needed. Decide whether fx should be a built-in driver or use generic ACP.
Confidence medium. Release: Not applicable. PR readiness: Needs work or a decision.
Check: GitHub check. Head CI: Test success, Check success, Mobile Native Static Analysis skipped, Release Smoke success.
Limits. The 29-file provider diff was not fully reviewed. ACP startup, model-selection support, settings, and install guidance were inspected. The author did not run authenticated fx Codex or Grok turns.
Request. Disable composer spellcheck to stop incorrect underlines in other languages.
Audit finding. The main composer ContentEditable still inherits browser spellcheck. This patch disables it for both web and desktop, whereas the related language-settings proposal preserves configurable spellcheck and another issue asks to preserve native corrections. These are different product choices, so neither proposal is a landed replacement.
Recommendation. Keep open: decision needed. Choose between disabling composer spellcheck and providing spelling settings before merging either proposal.
Confidence high. Release: Not applicable. PR readiness: Needs work or a decision.
Request. Accept newer ChatGPT account plans without failing the entire Codex account response.
Audit finding. Merged PR 8447 adds all five named plan values, including Edu Plus and Edu Pro, to the generated schema. Main still uses a closed literal union, while this PR adds a schema-local fallback for future unknown strings. The known account failure is fixed, but the forward-compatibility behavior still needs a decision.
Recommendation. Keep open: partial fix. Reframe this PR around unknown future plan values and review that fallback separately.
Confidence high. Release: In stable source. PR readiness: Needs work or a decision.
Limits. Latest-head required check results are absent for Test, Check, Mobile Native Static Analysis, Release Smoke. GitHub reports a conflict with main. Any passing checks apply to the existing head, not a rebased result.
Request. Update the checked-in dev container to install the repository toolchain with vp.
Audit finding. Main still uses Debian, Bun install, no Rust feature, and no declared non-root user. The diff moves setup to Ubuntu with Node 24, Rust, GitHub CLI, and isolated worktree state. The author completed container builds, but never completed its first dependency install, and Codespaces pairing remains untested.
Recommendation. Keep open: work remains. Complete one clean dependency install and paired dev start in the declared container configuration.
Confidence high. Release: Not applicable. PR readiness: Needs work or a decision.
Request. Discover Claude skills and native slash commands for the active workspace.
Audit finding. Main still probes Claude capabilities against ServerConfig.cwd rather than the composer workspace. The PR adds a workspace RPC, but the composer falls back to snapshot skills for a valid empty result and still reads snapshot skills for slash-menu skill entries. The newer workspace-skills PR covers more clients and providers but explicitly excludes native slash commands, so it is not a full replacement.
Recommendation. Keep open: work remains. Keep native slash-command discovery separate and use the newer workspace-skills path for skill menus.
Confidence high. Release: Not applicable. PR readiness: Needs work or a decision.
Request. Attach dropped files to a sidebar thread after opening that thread.
Audit finding. Main accepts files in the chat workspace but Sidebar has no file-drop handlers or cross-thread pending-drop store. The diff adds row targets and an environment-scoped handoff without automatically sending. The handoff has only one animation-frame retry if the composer is absent, and its tests cover the store rather than delayed mounting and navigation.
Recommendation. Keep open: decision needed. Add a delayed-composer handoff test and review file drops across thread navigation.
Confidence high. Release: Not applicable. PR readiness: Needs work or a decision.
Source: apps/web/src/components/Sidebar.tsx:1040. Main accepts files in the chat workspace but Sidebar has no file-drop handlers or cross-thread pending-drop store.
Check: GitHub check. Latest-commit checks: Test success, Check success, Mobile Native Static Analysis skipped, Release Smoke success.
Limits. An older related reference could not be retrieved from GitHub. It is excluded from verified dependencies and fix evidence.
Request. Keep desktop Connect auth polling until a legacy credential receives its account ID.
Audit finding. This patch targets the browser-login architecture in the still-open parent PR, not the current Clerk-based desktop path. Main has no DesktopConnectAuthProvider or connectAuthState module. The diff adds pending-identity polling and avoids a false onboarding prompt, but cannot stand alone on main.
Recommendation. Keep open: work remains. Fold this retry change into the browser-login parent PR before its integration review.
Confidence high. Release: Not applicable. PR readiness: Needs work or a decision.
Observed GitHub metadata. GitHub state OPEN. Author gnustella-lab. Updated 2026-08-22T14:21:24Z. Draft no. Target t3code/move-electron-auth-to-browser. Head 6998c9e9d108318e19a3ffc69e26a686ba08a460. Branch fix/pr7483-linux-auth-recovery. Size +134 / -9, 4 files, 3 commits. Mergeability MERGEABLE. Merge state UNSTABLE. Target is not main. Metadata snapshot 2026-09-01T11:57:41.491617+00:00.
Checks. Required: Test: not captured, Check: not captured, Mobile Native Static Analysis: not captured, Release Smoke: not captured. Observed: 10 passed, 0 failed, 0 pending, 3 skipped. Checks captured 2026-09-01T11:50:57.635739+00:00.
Reviews. GitHub review decision UNKNOWN. Current-head approvals 0. Current-head change requests 0. Unresolved review threads 0. Counts do not replace review of the findings.
Evidence read. body yes, discussion yes, full diff yes, current source yes, history yes.
Request. Allow mobile tasks in non-Git projects when New worktree is the configured default.
Audit finding. Main mobile mode selection does not check isRepo, while web already forces the current checkout for non-repositories. The diff adds a shared rule, waits for connected status, and rechecks queued tasks with a bounded query. It also revalidates queued payloads after the await, so the offline replay path is explicitly covered rather than only the composer.
Recommendation. Keep open: work remains. Review the mobile online and queued non-repository cases on current main.
Confidence high. Release: Not applicable. PR readiness: Needs work or a decision.
Limits. Latest-commit metadata has no required runs for Test, Check, Mobile Native Static Analysis, Release Smoke. GitHub reports conflicts with the target branch.
Request. Expose the native Codex Goal lifecycle through thread commands and live client state.
Audit finding. The current runtime exposes turns, rollback and feedback but no Goal operations or Goal RPCs. This PR adds commands and subscriptions for both clients while leaving native Codex authoritative. Stop would still interrupt only the turn, not pause the Goal, and the shared session-recovery locking change needs review with other handoff work.
Recommendation. Keep open: decision needed. Approve the Goal command and Stop behavior before reviewing the native API and session-lock changes.
Confidence high. Release: Not applicable. PR readiness: Needs work or a decision.
Limits. Goal contracts and runtime wiring were inspected, but the full 34-file lifecycle and client diff was not reviewed. Only selected current review findings were read. The complete discussion and resolved review history were not reviewed. Latest-head required check results are absent for Test, Check, Mobile Native Static Analysis, Release Smoke. GitHub reports a conflict with main. Any passing checks apply to the existing head, not a rebased result. An older related reference could not be retrieved from GitHub. It is excluded from verified dependencies and fix evidence.
Request. Append sanitized terminal output with byte-bounded compaction instead of repeatedly rewriting all history.
Audit finding. Main still rewrites the full history and caps by newline count, so carriage-return-only output can keep growing and increase disk writes. The patch adds ordered appends, 8/12 MiB compaction, and attach and shutdown tests, but the latest unresolved truncate-failure finding is valid: a failed clear, restart, or compaction is only logged, then later appends use the stale file. Its append-failure fallback and separate snapshot history fix earlier review issues, while queue fairness remains a separate change in PR #8966.
Recommendation. Keep open: work remains. Recover a failed truncation before accepting later appends, with a fault-injection test for clear and compaction.
Confidence high. Release: Not applicable. PR readiness: Needs work or a decision.
Request. Fix several persistence, protocol, cleanup, and scheduling failures in the Effect runtime.
Audit finding. Main now bounds raw ACP and Codex queues, and the projection bootstrap merge replays all unapplied events. It still has a global provider command worker and a synchronous five-second SQLite busy timeout, which the inspected PR changes. Those remaining mechanisms and the broader attachment, retry, and teardown scope are not replaced by the recent performance merges.
Recommendation. Keep open: partial fix. Split the remaining SQLite, command scheduling, and persistence fixes after removing landed queue and replay work.
Confidence medium. Release: In nightly source. PR readiness: Needs work or a decision.
Merged pr: PR #8187. perf(server): cut idle CPU use and stop provider event leaks
Merged pr: PR #7538. fix(server): replay all un-applied events during projection bootstrap
Limits. The 51-file runtime diff was not fully reviewed. SQLite retry, keyed worker, ACP termination, startup subscription, and recent merged overlaps were inspected.
Request. Recall earlier prompts and image attachments with unmodified arrow keys.
Audit finding. Main has no prompt-history navigation. The proposed history is built from timelineMessages, whose local feedback rows are appended after normal messages, so the reviewed /feedback ordering defect remains in the submitted code. The patch also assumes image-only attachments and needs adaptation to the current file and video types.
Recommendation. Keep open: work remains. Fix feedback ordering in prompt history and adapt recall to the current attachment union.
Confidence high. Release: Not applicable. PR readiness: Needs work or a decision.
Request. Show failed, input-needed, or unread-completion dots on projects in the switcher.
Audit finding. The current searchable project combobox still renders only each project icon, name, and settings gear. The patch adds an attention rollup but targets the replaced MenuRadioItem path and the removed client-side auto-settling calculation. Porting those parts is necessary, while the attention feature itself is still absent.
Recommendation. Keep open: work remains. Port the attention rollup to the current combobox and server-owned settled state.
Confidence high. Release: Not applicable. PR readiness: Needs work or a decision.
Request. Copy selected untracked local files into each new worktree using .worktreeinclude.
Audit finding. Main initializes submodules but has no .worktreeinclude copy step. The proposal uses Git pattern matching at the common worktree-creation boundary and avoids replacing tracked files or writing through destination symlinks. Copies are best-effort and can add a sixty-second listing plus unbounded file-copy time before setup.
Recommendation. Keep open: decision needed. Decide whether to adopt .worktreeinclude and set an acceptable copy limit before merge.
Confidence high. Release: Not applicable. PR readiness: Needs work or a decision.
Limits. Latest-commit metadata has no required runs for Test, Check, Mobile Native Static Analysis, Release Smoke. GitHub reports conflicts with the target branch.
Request. Route web Lucide imports through a generated icon module and enforce that boundary.
Audit finding. Main still permits direct Lucide imports and has no generated icon module. The import-only changes preserve the icon names, but the new custom TSX scanner misses a later icon import after a self-closing JSX element whose attribute contains a greater-than comparison. I reproduced that failure with the PR parser, so this is unfinished tooling rather than a safe mechanical refactor.
Recommendation. Keep open: decision needed. Decide whether to add this import boundary, then replace or repair the TSX scanner before review.
Confidence high. Release: Not applicable. PR readiness: Needs work or a decision.
Request. Record durable cancellation of pending user questions after a successful turn interrupt.
Audit finding. Main still has no provider-independent cancellation after interrupt, while merged PR 5127 only covers Claude session shutdown. The patch rereads activities after interrupt, but the fresh read does not wait for provider ingestion and its timestamp-plus-ID sort can reopen a request resolved at the same timestamp. Those two review findings remain valid in the diff, and current interrupt-failure recovery must also be preserved.
Recommendation. Keep open: partial fix. Base cancellation on durable request ordering and completed provider resolution before integrating current interrupt recovery.
Confidence high. Release: In stable source. PR readiness: Needs work or a decision.
Merged pr: PR #5127. Claude session-stop cleanup is on main and is an ancestor of v0.0.37.
Comment: Discussion comment. Equal-timestamp ordering finding remains in the reviewed diff.
Limits. Required CI checks are absent at the reviewed head: Test, Check, Mobile Native Static Analysis, Release Smoke. GitHub reports merge conflicts at the reviewed head.
Request. Add the Factory Droid native protocol provider with approvals, resume, and rewind.
Audit finding. Main has no Droid driver, and generic ACP does not supply the native spec handoff and rewind behavior this PR targets. The inspected shared changes replace count-only rollback and move provider rewind before filesystem restore for all providers. The last maintainer comment objects to the permanent support cost of this scope, so product approval needs confirmation before more merge work. No recent merge supplies the Droid integration.
Recommendation. Keep open: decision needed. Confirm whether maintainers want the Droid integration after the support-cost objection.
Confidence medium. Release: Not applicable. PR readiness: Needs work or a decision.
Comment: Discussion comment. The maintainer comment rejects this scope on support-cost grounds, not as already fixed.
Limits. The 65-file, 11,424-addition diff was not fully reviewed. Driver setup, shared rollback preparation, cursor persistence, and checkpoint ordering were inspected. The live account recording predates the current head. Top-level comments and current unresolved findings were reviewed, but the full history of 146 review threads was not read. Required checks have no results on this head: Test, Check, Mobile Native Static Analysis, Release Smoke.
Request. Run thread bootstrap for HTTP dispatch as well as WebSocket dispatch.
Audit finding. Main still sends HTTP bootstrap commands directly to the engine while WebSocket runs the setup flow. The extraction adds HTTP success and rollback coverage, but it copies the older bootstrap without the deletion-reactor drain added by the merged first-send retry fix. That drain must move with the shared implementation or retrying a thread ID can race cleanup.
Recommendation. Keep open: work remains. Rebase the shared bootstrap on main and preserve the deletion-reactor drain in both transports.
Confidence high. Release: Not applicable. PR readiness: Needs work or a decision.
Merged pr: PR #8226. Merged retry fix added the drain that this extraction must preserve.
Limits. Required current-head checks are absent: Test, Check, Mobile Native Static Analysis, Release Smoke. GitHub reports merge conflicts with the current base.
Request. Use themed app menus on desktop and add styled browser text-editing menus.
Audit finding. Main still routes desktop app menus through the native bridge. The patch changes that routing and adds browser-wide text commands, but its Cut handler deletes selected input text even if both clipboard attempts fail. That unresolved finding matches the diff, so this cannot be treated as a cosmetic icon change or a safe merge.
Recommendation. Keep open: decision needed. Fix the Cut data-loss path and obtain a decision on replacing native desktop app menus.
Confidence high. Release: Not applicable. PR readiness: Needs work or a decision.
Request. Keep a user active override through later turn, session, and request activity.
Audit finding. Main still clears both active and settled overrides on activity, including after the server-side settlement change. Merged PR 7454 limits repeat settlement from a closed PR, but does not implement durable active pins. This patch changes an explicit lifecycle policy rather than closing the whole repeated-settlement report.
Recommendation. Keep open: decision needed. Decide whether active pins survive new work, then adapt the tests to current server settlement.
Confidence high. Release: Not applicable. PR readiness: Needs work or a decision.
Limits. Required CI checks are absent at the reviewed head: Test, Check, Mobile Native Static Analysis, Release Smoke. GitHub reports merge conflicts at the reviewed head.
Request. Let mobile browse environment-local dev servers and attach annotated screenshots to a draft.
Audit finding. Main has no mobile dev-server proxy capability or route. The reviewed proxy persists its signing key but keeps redeemed ticket IDs only in memory, so restart restores the validity of a consumed ticket. Its redirect filter also accepts a backslash authority form, and the review identifies mobile capture and session-close races alongside the missing iOS entry point.
Recommendation. Keep open: decision needed. Fix the proxy ticket, redirect, and mobile cancellation defects before the shared-gateway design review.
Confidence medium. Release: Not applicable. PR readiness: Needs work or a decision.
Comment: Discussion comment. The reviewed ProxyAccess code uses a persistent key and process-local consumed ticket map.
Comment: Discussion comment. The redirect filter only checks a leading slash and rejects a double slash, not a backslash authority.
Check: GitHub check. Required check is FAILURE on the collected latest head.
Limits. No rebuilt Android or iOS client was tested. The 108,027-byte diff was not fully read. Review covered the named feature paths and current main, not every file or test. Check is not passing on the latest head. Its failure log was not inspected. GitHub reports merge conflicts with the current base.
Request. Keep timeline text out of translucent composer banners without changing the glass opacity setting.
Audit finding. Merged work restored the banner attachment mask, and the new ComposerBanner still uses that mask. This PR now adds a separate bottom mask to MessagesTimeline, which main still renders with only the top fade. The seam repair is covered, but the proposed timeline occlusion remains absent and needs a fresh check with the rewritten banners.
Recommendation. Keep open: partial fix. Rebase the remaining timeline-mask change and verify text occlusion with the current banner components.
Confidence high. Release: In nightly source. PR readiness: Needs work or a decision.
Request. Scroll the sidebar to threads selected outside it while leaving direct row navigation still.
Audit finding. Main scrolls highlighted search results but has no route-aware active-thread scroll hook. The patch covers both sidebar implementations, rendered off-preview active rows, and reduced motion, but also changes project-filter visibility and introduces navigation-source state. The new hook has no focused tests or current interaction recording, so its no-scroll cases still need verification.
Recommendation. Keep open: work remains. Add focused navigation-source coverage and record external navigation, sidebar clicks, and settings round trips.
Confidence high. Release: Not applicable. PR readiness: Needs work or a decision.
Request. Make the generated worktree branch prefix configurable.
Audit finding. Main still hard-codes t3code/ when renaming the generated worktree branch. The diff adds a server setting and web control, but leaves temporary branch names unchanged and adds no mobile settings control. The user-facing scope is therefore generated final branch names, not every worktree branch.
Recommendation. Keep open: decision needed. Decide the prefix-setting scope and document temporary-name and mobile-control limits.
Confidence high. Release: Not applicable. PR readiness: Needs work or a decision.
Request. Remember merge methods and allow a project-specific default on web and desktop.
Audit finding. Main has no persistent merge-method defaults. Theo closed this proposal with a stated product decision against the added settings, then Julius reopened it on August 28 without an explanatory comment. The feature is still absent, but that conflicting maintainer history requires a decision rather than automatic archival.
Recommendation. Keep open: decision needed. Ask the maintainers to resolve the merge-defaults decision recorded in the close and reopen history.
Confidence high. Release: Not applicable. PR readiness: Needs work or a decision.
Request. Restore active composer tasks after relaunch even when the saved plan is outside the activity window.
Audit finding. Merged banner work already derives composer task progress from persisted active-plan activities instead of process-local shell progress. Main still does not pin an older active plan outside the 500-activity window or finalize matching turn.aborted events as this PR proposes. The remaining server changes need to target the new snapshot ID selectors and an unused migration number, since migration 42 now stores linked pull requests.
Recommendation. Keep open: partial fix. Keep the uncovered plan-pinning and abort-finalization changes, update the snapshot paths, and renumber the migration.
Confidence high. Release: In nightly source. PR readiness: Needs work or a decision.
Request. Open a sent skill mention in the file panel, including skills outside the project.
Audit finding. Main still renders skill mentions as static spans. The diff introduces rooted file tabs and a v12 persistence migration, but a link leaving the skill directory retains that root and can be rejected by the workspace file API. That current-head review finding remains unresolved, and mobile has no equivalent action.
Recommendation. Keep open: decision needed. Resolve parent-relative links to their actual file root before reviewing rooted skill tabs.
Confidence high. Release: Not applicable. PR readiness: Needs work or a decision.
Request. Capture another desktop application window and attach its image and available accessibility data to a thread.
Audit finding. Main has preview-browser capture but no native window-capture path. The proposed onboarding still offers capture on unsupported X11 and shows a Shift-pair shortcut that Wayland cannot register. A maintainer reopened the PR after an earlier closure over permission, privacy, and native support policy. The remaining platform decision and incomplete native review rule out a merge or obsolete-PR recommendation.
Recommendation. Keep open: decision needed. Confirm the supported platforms and permission policy, then fix capability-aware onboarding before the native review.
Confidence medium. Release: Not applicable. PR readiness: Needs work or a decision.
Comment: Discussion comment. Earlier maintainer closure names permission, privacy, and native support policy.
Comment: Discussion comment. Latest onboarding finding remains in the reviewed head: no capture-capability check and an unusable Wayland default.
Limits. The 124-file diff and native dependency changes were not fully reviewed. No native permission, global shortcut, or window capture was exercised. The latest Macroscope correctness review was skipped because its cost limit was reached.
Request. Keep a CopilotKit and OpenRouter review-agent demo separate from the product merge queue.
Audit finding. The current body explicitly says demo-only and do not merge. An earlier maintainer closure note says this product and infrastructure bundle needs a smaller owned plan, while the PR is open again. The new OpenRouter runtime and CopilotKit panel are absent from main, so the owner must confirm whether to retain the demo or archive it.
Recommendation. Keep open: decision needed. Confirm whether the prior maintainer decision to remove this demo bundle from review still stands.
Confidence high. Release: Not applicable. PR readiness: Needs work or a decision.
Comment: Discussion comment. Prior maintainer closure note requires a smaller owned plan before considering the feature.
Limits. Runtime, settings, and review-helper changes were inspected, but the full 38-file demo and generated lockfile diff was not reviewed line by line. The full historical review discussion was not read; the body, current thread status, and available maintainer direction were checked. Canonical latest-commit metadata has no results for these required checks: Test, Check, Mobile Native Static Analysis, Release Smoke.
Request. Delete inactive managed tunnels and let hosts recover with the same public hostname.
Audit finding. Main supports explicit tunnel release on orderly shutdown, but has no inactive-tunnel sweep or registered host recovery. The reviewed reaper and provider changes gate deletion on recovery support, recheck inactive state, preserve hostnames, and default cleanup to off. The author has not run the Cloudflare sleep/wake canary, and this broad lifecycle diff still needs a complete integration review before any deletion rollout.
Recommendation. Keep open: work remains. Complete the isolated Cloudflare recovery canary and keep cleanup off until it passes.
Confidence medium. Release: Not applicable. PR readiness: Needs work or a decision.
Pr: PR #8788. Separate restart-backoff work must be reconciled with this supervisor change.
Limits. The 36-file diff is about 354 KB. Reaper, migration, provider recovery, runtime, and rollout sections were inspected, but the full registration and test diff was not reviewed. The author explicitly reports that the live Cloudflare sleep/wake canary has not run. GitHub reports merge conflicts with the current base.
Request. Open linked pull requests from both mobile thread menus.
Audit finding. Main can present a linked PR badge but the v1 and v2 thread menus have no open-linked-PR action. The diff adds provider-aware browser actions for open, closed, and merged requests and reports browser failures. The existing linked-thread feature is therefore not a replacement for this mobile entry point.
Recommendation. Keep open: work remains. Review the browser action in both mobile thread menu variants.
Confidence high. Release: Not applicable. PR readiness: Needs work or a decision.
Request. Publish mobile updates skipped while their compatible native build was still running.
Audit finding. Main only publishes fingerprint-gated updates on push and has no completion retry. The patch adds scheduled retries and fixes the reviewed rollback-preservation cases, but separates scheduled and release concurrency groups. Those groups can publish different commits at the same time, and the diff has no newer-published-commit guard, so an older run can overwrite a newer OTA.
Recommendation. Keep open: work remains. Serialize the OTA publication step across both trigger types and add a test that rejects stale publication.
Confidence high. Release: Not applicable. PR readiness: Needs work or a decision.
Request. Keep mobile image bytes out of WebSocket turn messages, including on older servers.
Audit finding. Merged mobile file support and background uploads now send image IDs to upload-capable environments and retain uploads for queued retries. Main still deliberately falls back to inline images when upload support is absent or unknown, which this PR would reject with an upgrade error. The large-message failure is therefore covered for current servers but not the full compatibility scope.
Recommendation. Keep open: partial fix. Reduce this PR to the legacy-server behavior decision and any missing send-boundary guard.
Confidence high. Release: In nightly source. PR readiness: Needs work or a decision.
Request. Keep an opt-in SSH agent channel alive for desktop-managed remote servers.
Audit finding. Main has no retained SSH agent channel or saved forwarding option. The diff adds them, but verified review findings show that password-auth failures lose their diagnostic text and an existing managed server is stopped before the new socket is validated. The forwarding choice also survives dialog cancellation, so reopening the form for a different host can retain an earlier opt-in.
Recommendation. Keep open: work remains. Fix forwarding authentication, pre-stop socket validation, and per-host opt-in reset before integration.
Confidence high. Release: Not applicable. PR readiness: Needs work or a decision.
Request. Show a reconnecting status when an active thread loses its event stream.
Audit finding. Main does not use event-stream health in the Working label or timer. The diff adds that presentation for web and desktop, but subscribeDynamic still drains a transport error without marking shell or detail state detached, so that dead-stream path continues to report Working. The author confirms this remaining recovery dependency, and mobile does not yet consume the new helpers. Non-config transport failures still drain without onExpectedFailure. That path can leave shell/detail presentation live and still show Working.
Recommendation. Keep open: work remains. Connect transport-failure notification to the new stream-health state before claiming complete drop visibility.
Confidence high. Release: Not applicable. PR readiness: Needs work or a decision.
Latest main change. Non-config transport failures still drain without onExpectedFailure. That path can leave shell/detail presentation live and still show Working. Keep the known display/recovery gap and coordinate it with general subscription recovery.
Limits. Required current-head checks are absent: Test, Check, Mobile Native Static Analysis, Release Smoke. GitHub reports merge conflicts with the current base.
Request. Show Codex context compaction and elapsed time while it runs.
Audit finding. Main does not show Codex item-compaction while it runs. The PR adds that flow but retains the separate legacy thread-compacted activity, so one compaction can produce two completed rows. Its live region also includes a timer that changes every second, creating repeated announcements.
Recommendation. Keep open: work remains. Unify legacy and item compaction into one activity and keep the ticking timer outside the live region.
Confidence high. Release: Not applicable. PR readiness: Needs work or a decision.
Check: GitHub check. Latest head required checks: Test success, Check success, Mobile Native Static Analysis skipped, Release Smoke success.
Comment: Discussion comment. Discussion evidence checked against the submitted source change.
Limits. All production changes were inspected, but the full UI and ingestion test diff was not reviewed. GitHub reports a conflict with main. Any passing checks apply to the existing head, not a rebased result.
Request. Fix clipped Android text when the app runs in a desktop window with different display density.
Audit finding. Main still unlocks orientation only at the tablet-width threshold and does not rebind React Native display metrics from the Activity. The patch adds desktop mode and density handling plus a new Android heading, while the later merged header-overflow change is not a DeX density fix. The author confirms that no after result exists on DeX, and this native change cannot be delivered or proved by an OTA alone.
Recommendation. Keep open: work remains. Build the Android native candidate and verify the same DeX or Pixel Desktop window before approving it.
Confidence medium. Release: Not applicable. PR readiness: Needs work or a decision.
Comment: Discussion comment. Review discussion checked against the source.
Check: GitHub check. Head 5ad82f3: Check, Test, and Release Smoke pass. Mobile Native Static Analysis is skipped.
Limits. No desktop-window Android runtime result or after screenshot is available. The current head conflicts with main, and post-rebase checks are not available.
Request. Sign Windows native addons so Smart App Control does not block the packaged desktop backend.
Audit finding. Main still omits native Windows signing. The proposed extension-only match also selects Linux .node files retained in server.asar.unpacked, as the current build script and electron-builder signing filter confirm. The unresolved review identifies a real cross-format signing gap, separate from the older missing-addon packaging defects.
Recommendation. Keep open: work remains. Select Windows PE binaries for signing and verify that WSL ELF natives are excluded from the signing inputs.
Confidence high. Release: Not applicable. PR readiness: Needs work or a decision.
Limits. Azure signing and Smart App Control startup were not executed. Required Check, Test, Release Smoke, Mobile Native Static Analysis checks are absent on the collected head.
Request. Support OpenCode 2 while retaining OpenCode 1 compatibility.
Audit finding. Main still uses the existing OpenCode client and has no OpenCode 2 compatibility bridge. The PR adds one, but its current commit has a review that reproduced twelve failures, including agent rules bypassing supervised mode and allowing tools during background generation. The source still caches session permissions without server enforcement and sends order together with cursor when paging messages. It also lacks permission.list, question.list, and session status, message, and children methods now required by main.
Recommendation. Keep open: work remains. Rework the OpenCode 2 bridge against current main and rerun the documented permission, reconnect, pagination, and form cases.
Confidence high. Release: Not applicable. PR readiness: Needs work or a decision.
Comment: Discussion comment. The same commit was tested on two OpenCode betas. The review reports twelve failures and says not to merge.
Comment: Discussion comment. Verified source mechanism: configured agent rules can override global mode rules, while session permissions are only cached.
Check: GitHub check. Latest commit: Check success, Test success, Mobile Native Static Analysis skipped, Release Smoke success.
Limits. The two-beta runtime matrix in the existing review was not rerun during this read-only audit. GitHub reports a merge conflict with current main.
Request. Avoid inheriting a base branch PR when a feature branch tracks a non-default integration branch.
Audit finding. Merged PR 7317 prevents inheritance from the default branch, but main still allows a non-default base such as origin/dev to lend its PR to a new branch. This proposal extends the guard and fixes suffix cases. It intentionally suppresses legitimate differently named upstreams after local branch renames, so that tradeoff still needs acceptance.
Recommendation. Keep open: partial fix. Decide how renamed local branches should retain a legitimate PR before accepting the new guard.
Confidence high. Release: In stable source. PR readiness: Needs work or a decision.
Source: apps/server/src/git/GitManager.ts:1004. Merged PR 7317 prevents inheritance from the default branch, but main still allows a non-default base such as origin/dev to lend its PR to a new branch.
Merged pr: PR #7317. Handles default-branch upstream inheritance only. Its merge commit is on the pinned main history.
Comment: Discussion comment. Author explicitly accepts the remaining renamed-branch ambiguity. The current guard still suppresses the lookup.
Limits. Latest-commit metadata has no required runs for Test, Check, Mobile Native Static Analysis, Release Smoke. GitHub reports conflicts with the target branch.
Request. Recover Android composer position after an interrupted keyboard dismissal and resume.
Audit finding. Main already has a thread-only keyboardStateSuspect guard, but new-task, terminal, and review sticky views lack the shared recovery proposed here. The patch replaces the thread guard with focus or keyboard-show recovery and adds terminal focus events. It must preserve the newer iOS always-enabled sticky-view behavior, and the body has no Android runtime proof.
Recommendation. Keep open: partial fix. Verify recovery with both dismissed and still-visible Android keyboards across all four input screens.
Confidence medium. Release: Not applicable. PR readiness: Needs work or a decision.
Request. Increase iOS Markdown row height when drawn text exceeds its first measured size.
Audit finding. Main still enumerates native text lines without first forcing layout at the current text-container width. The React wrapper has no measured-line minHeight correction, so the narrow two-file fix is not on main and does not cover Android. The PR targets an orchestration branch and its collected 888-file diff includes that parent work. The Check failure is a server Layer type error in inherited orchestration tests, not a native layout error.
Recommendation. Keep open: work remains. Restack or extract the two-file iOS layout fix and verify font scaling and width changes.
Confidence high. Release: Not applicable. PR readiness: Needs work or a decision.
Observed GitHub metadata. GitHub state OPEN. Author mwolson. Updated 2026-09-01T05:36:36Z. Draft no. Target t3code/codex-turn-mapping. Head 138cb2a7f38c0a9d4b484827218c3e23018ccaac. Branch fix/mobile-ios-markdown-stale-content. Size +181675 / -85512, 888 files, 240 commits. Mergeability CONFLICTING. Merge state DIRTY. Target is not main. Metadata snapshot 2026-09-01T11:57:41.491617+00:00.
Check: GitHub check. Current head 138cb2a7: Check=FAILURE. The log reports TS2375 in orchestration-v2/ProviderSessionManager.test.ts:836, an inherited server Layer error. GitHub reports merge conflicts.
Limits. The complete 12.84 MB, 888-file parent-branch diff was not reviewed. The task-specific two-file head commit was read in full.
Request. Open desktop thread routes from operating-system deep links, including cold starts.
Audit finding. Main has no thread-link listener or renderer bridge. This patch adds buffering, subscription teardown, and argv delivery, but its UUID-only parser excludes IDs accepted by the entity contracts. Later commits resolve the service-injection findings, while the author explicitly leaves the hard-renderer-reload delivery gap open.
Recommendation. Keep open: work remains. Align ID parsing with the contracts and handle renderer reload before the final desktop interaction review.
Confidence high. Release: Not applicable. PR readiness: Needs work or a decision.
Commit: Commit 41adccc83e81. Recent HTTP route work explicitly preserved long valid thread IDs.
Comment: Discussion comment. The author confirms that hard reload does not run the unsubscribe cleanup and leaves that delivery gap open.
Limits. Current-head screenshots and interaction video are explicitly missing. Windows and Linux argv behavior has no reported live-platform verification.
Request. Prevent Homebrew Codex upgrades from leaving a quarantined binary that cannot pass background probes.
Audit finding. Main still uses ordinary brew upgrade, so the reported quarantine case is not handled here. The PR adds --no-quarantine for every Homebrew-managed provider, but the current upstream upgrade command no longer declares that option. This is a verified compatibility concern, not a safe command-only fix.
Recommendation. Keep open: work remains. Replace the removed Homebrew flag with a currently supported and tested update approach.
Confidence high. Release: Not applicable. PR readiness: Needs work or a decision.
Request. Let a specific overlay timeout reach the agent before the broker gives up.
Audit finding. Main still sends the full broker timeout to the host and drops host timeout-stage detail. The complete patch reserves response time and carries a closed stage value for overlay, navigation, and viewport failures. It overlaps the earlier host-budget patch and the newer navigation-start protocol, while deliberately leaving desktop waitFor and snapshot text unchanged.
Recommendation. Keep open: work remains. Combine the timeout-budget choice with navigation-start handling and keep one tested broker contract.
Confidence high. Release: Not applicable. PR readiness: Needs work or a decision.
Request. Post deterministic advisory comments about PR contribution requirements.
Audit finding. Main has vouch and size automation but no deterministic contribution-description evaluator. The diff adds trusted-base execution and advisory comments, plus new template requirements. Its related vouched-guide PR 8274 was closed unmerged, so the richer checks stay disabled unless maintainers adopt that guide or an equivalent.
Recommendation. Keep open: decision needed. Decide whether to adopt the advisory workflow and its contribution-template requirements.
Confidence high. Release: Not applicable. PR readiness: Needs work or a decision.
Request. Add isolated side conversations that answer questions without interrupting the active coding turn.
Audit finding. Main has no side-question RPC or matching text-generation operation. The patch adds a separate ephemeral conversation and cancellation path across five providers and both clients. A prior maintainer closure note explicitly rejects adding this second protocol, but the PR is open again without a recorded replacement product decision.
Recommendation. Keep open: decision needed. Confirm the prior maintainer rejection before continuing review of the reopened side-conversation proposal.
Confidence high. Release: Not applicable. PR readiness: Needs work or a decision.
Comment: Discussion comment. Prior maintainer closure note rejects a second ephemeral side-conversation protocol.
Check: GitHub check. Current-head Test, Check, and Release Smoke succeed. Mobile Native Static Analysis is skipped.
Limits. Provider operations, coordinator, prompts, and contracts were inspected, but the full 47-file UI and test diff was not reviewed line by line. Codex tool isolation relies on read-only execution and a no-tool instruction rather than a complete tool-disable switch. The full historical review discussion was not read; the body, current thread status, and available maintainer direction were checked.
Request. Capture and show turn diffs only after the provider completes the turn.
Audit finding. Main still dispatches and captures a placeholder on turn.diff.updated, and attaches changed files to an assistant message. The proposal moves capture to completion and renders a separate settled-turn row, including replacement of older mid-turn checkpoints. It adds new turn-ownership state and projection reads, so rapid follow-ups and multi-provider completion ordering need focused review.
Recommendation. Keep open: work remains. Review completion ownership against the current provider event subscription changes and verify final file content.
Confidence high. Release: Not applicable. PR readiness: Needs work or a decision.
Request. Provide the Windows native helper pipe needed by Codex Computer Use.
Audit finding. Current Codex session startup does not provision a Computer Use helper or replace Desktop-owned pipe configuration. MCP app approval support does not create that native transport. The PR adds a Windows helper bridge and trusted shim, which needs a complete Windows process-lifecycle and permission review.
Recommendation. Keep open: work remains. Complete the Windows bridge review and verify a real helper-backed Computer Use session.
Confidence high. Release: Not applicable. PR readiness: Needs work or a decision.
Limits. Runtime bridge configuration was inspected, but the full embedded shim and helper lifecycle diff was not reviewed. No Windows runtime verification was performed. Only selected current review findings were read. The complete discussion and resolved review history were not reviewed. Latest-head required check results are absent for Test, Check, Mobile Native Static Analysis, Release Smoke.
Request. Track account-wide referral points and award a referrer once after a referred account links an environment.
Audit finding. Main has no referral ledger or referral endpoints. The reviewed service uses transactions and unique award keys, while pending-award recovery still selects all matching pending accounts before limiting concurrency. This feature needs an agreed award policy and a relay migration rollout, not closure as an existing fix.
Recommendation. Keep open: decision needed. Review the award policy and bound pending-award recovery before planning the relay rollout.
Confidence medium. Release: Not applicable. PR readiness: Needs work or a decision.
Limits. The 50-file client, relay, and migration diff was not fully reviewed. No hosted relay state or live account was used. GitHub reports merge conflicts at the reviewed head.
Request. Carry explicit skill invocation metadata through clients and provider adapters.
Audit finding. Main still sends prompt text without explicit skill ranges, so skill deduplication does not solve this request. The patch adds native Codex inputs and document expansion for Claude, Cursor, Grok, and OpenCode, plus mobile draft and outbox metadata. Its ordinary-send wiring does not cover the early plan-follow-up path, which can still send a selected skill as plain text.
Recommendation. Keep open: work remains. Carry skill metadata through plan follow-up before completing the cross-provider review.
Confidence high. Release: Not applicable. PR readiness: Needs work or a decision.
Comment: Discussion comment. The downstream stack review identifies the shared plan-follow-up gap, which is still absent from this base patch.
Check: GitHub check. Current-head Test, Check, and Release Smoke succeed. Mobile Native Static Analysis is skipped.
Limits. The contract, provider boundaries, and key client persistence paths were reviewed, but the full 50-file diff and all added tests were not reviewed line by line.
Request. Keep Codex thread resume usable when generated protocol schemas lag the CLI.
Audit finding. Merged PR 8346 adds the completed subagent activity kind that caused the reported failure. This PR goes further by decoding only consumed open-response fields and warning about dropped notifications, neither of which is in main. Its fallback from any decode failure to a new native thread can lose conversation continuity and conflicts with strict-resume handoff requirements.
Recommendation. Keep open: partial fix. Separate narrow response decoding from the fresh-thread fallback and review it with the resume-payload fix.
Confidence high. Release: In stable source. PR readiness: Needs work or a decision.
Request. Use the selected local worktree base when origin has no matching branch.
Audit finding. Main still aborts bootstrap when resolveRemoteTrackingCommit fails after fetching origin. This diff catches every resolution error and keeps the local base, while the related PR #8751 uses an explicit missing-ref result. These are competing fixes for the same local-only branch failure.
Recommendation. Keep open: work remains. Consolidate this change with PR #8751 after comparing missing-ref and real Git error handling.
Confidence high. Release: Not applicable. PR readiness: Needs work or a decision.
Request. Keep skill selection usable while answering a pending question.
Audit finding. Main still shares the composer between draft and pending-answer text, so the picker problem remains relevant. The proposed fix still uses an append-only isPlainInsert check, which misses skill-query replacement and mid-answer insertion, and it hardcodes the new plainText mode to false. The unresolved review findings match the submitted code, including an unscoped capture-phase Escape handler.
Recommendation. Keep open: work remains. Repair programmatic answer replacement and scoped dismissal, then remove the unreachable plain-text mode.
Confidence high. Release: Not applicable. PR readiness: Needs work or a decision.
Request. Allow the selected provider package to run install scripts during pnpm global updates.
Audit finding. Main still builds pnpm add -g without allow-build. The merged npm fix does not change pnpm or Bun update commands. This PR closes the pnpm postinstall gap, but its unconditional flag rejects pnpm versions older than 10.4.
Recommendation. Keep open: work remains. Decide whether to require pnpm 10.4 or version-gate allow-build.
Confidence high. Release: Not applicable. PR readiness: Needs work or a decision.
Request. Load Azure DevOps PRs with nullable fields and authenticated comments, and check out same-repository source branches.
Audit finding. Main still rejects nullable Azure fields and reads comments through az rest, so this adapter fix remains needed. The latest diff fixes several earlier fork findings, but an empty cross-repository identity still falls through to getRepositoryCloneUrls with an empty repository. The latest unresolved review identifies that exact path, so checkout must reject incomplete fork identity before merge.
Recommendation. Keep open: work remains. Reject incomplete fork repository identity before clone lookup and add a materialization test.
Confidence high. Release: Not applicable. PR readiness: Needs work or a decision.
Request. Limit decoded activity payload bytes for full and windowed thread detail reads.
Audit finding. Merged snapshot work now loads client activity payloads in sequential batches of 25 and projects each batch before retaining it. It does not enforce a byte budget, and raw detail reads still fetch up to 500 full payloads, so this PR is not fully covered. Its proposed 16 MiB selection also drops older activities, unlike the current client batching, and must be reconciled with both paths.
Recommendation. Keep open: partial fix. Rework the byte-limit proposal against the separate client and raw detail-read paths and measure the remaining peak memory.
Confidence high. Release: In nightly source. PR readiness: Needs work or a decision.
Request. Avoid a full origin fetch before creating a start-from-origin worktree.
Audit finding. Main still fetches all origin branches before resolving the base commit. This diff uses an existing local origin ref without refreshing it, and fetches one branch only when that ref is absent. The speed improvement changes the meaning of Start from origin because stale local refs can select an old commit.
Recommendation. Keep open: decision needed. Decide whether Start from origin promises a fresh remote commit before choosing this fast path.
Confidence high. Release: Not applicable. PR readiness: Needs work or a decision.
Request. Keep server startup and file-tree listing usable when the native workspace index cannot load.
Audit finding. Main still imports the native FileFinder eagerly and has no creation-failure listing fallback. The diff adds degraded listing, but a failed or timed-out git ls-files falls into a raw directory walk that does not honor Git ignore rules. The unresolved empty-directory and fsmonitor/ignore findings remain visible in the latest implementation.
Recommendation. Keep open: work remains. Preserve ignore behavior when the Git fallback fails and decide how to include empty directories.
Confidence high. Release: Not applicable. PR readiness: Needs work or a decision.
Request. Show Grok child agents and workflows through the existing task event stream.
Audit finding. Main Grok handling does not map subagent_spawned, subagent_progress, or workflow_updated into task events. The Grok reliability merge and the small ACP Task classification PR do not provide a live child roster. This diff adds the missing mapping plus startup buffering and watchdog updates, which require review for stale-session events and bounded retained task state.
Recommendation. Keep open: work remains. Review task identity and startup buffering against the current Grok adapter.
Confidence high. Release: Not applicable. PR readiness: Needs work or a decision.
Request. Close an already-open WebSocket when its authenticated session is revoked.
Audit finding. Main only marks the session revoked and leaves an established RPC socket running. The diff adds process-local revocation waiters and a socket race, which covers revocation through the running server. However, t3 auth creates its own auth runtime and writes the database directly, so it cannot wake those waiters despite the PR documentation promising that behavior.
Recommendation. Keep open: work remains. Make CLI revocation notify the running server and test closure after a confirmed WebSocket subscription.
Confidence high. Release: Not applicable. PR readiness: Needs work or a decision.
Request. Keep the Windows desktop backend running in a system tray and show the local active-agent count.
Audit finding. Main quits when the last Windows window closes and has no tray service. The patch adds close and minimize settings and counts primary plus WSL sessions, but shouldHideForTray checks only settings and quitting, not whether tray registration succeeded. Its new window-all-closed registration guard cannot help after the close handler prevents closing and hides the window, leaving no tray entry to restore it. Earlier Linux and settings-visibility review findings are fixed in the current diff.
Recommendation. Keep open: work remains. Gate close and minimize hiding on a working tray and test registration failure through the real window handlers.
Confidence high. Release: Not applicable. PR readiness: Needs work or a decision.
Limits. No Windows tray, shutdown, or registration-failure runtime was executed. Required Check, Test, Release Smoke, Mobile Native Static Analysis checks are absent on the collected head.
Request. Restore confirmed file contents and show an error when autosave fails.
Audit finding. Main leaves failed edits pending and has no rollback callback or save-error toast. This diff restores the last confirmed text and discards the failed latest edit, while preserving newer edits during a write. That is a data-retention choice as well as an error-display fix, and overlaps the save-coordinator changes in PR #8630 and PR #8632.
Recommendation. Keep open: decision needed. Decide whether failed edits should be retained for retry before merging automatic rollback.
Confidence high. Release: Not applicable. PR readiness: Needs work or a decision.
Request. Keep desktop windows and restore stopped backends when update installation fails.
Audit finding. Main destroys windows before calling the updater, so the reported failure remains. The diff handles an Effect failure from quitAndInstall, but the Electron wrapper only catches thrown exceptions and electron-updater normally reports install failures by emitting error and returning. In that real path the new code still destroys windows, which the injected-failure test does not cover.
Recommendation. Keep open: work remains. Handle the updater error event as an install failure before destroying windows and test that event path.
Confidence high. Release: Not applicable. PR readiness: Needs work or a decision.
Limits. The installed electron-updater BaseUpdater implementation was read locally; no packaged installer failure was run. Required CI has no result on the reviewed head: Test, Check, Mobile Native Static Analysis, Release Smoke.
Request. Settle pending OpenCode requests during teardown and expose unknown permissions in both clients.
Audit finding. Main still leaves pending request maps unresolved during stop and maps unknown permissions to unknown. This PR covers interruption, session errors, replacement, shutdown, and generic approval controls beyond the narrower teardown proposal. Its old request handlers and interrupt implementation must be combined with current child routing and acknowledged recursive cancellation, and it does not repair historical pending rows.
Recommendation. Keep open: work remains. Rebase request settlement onto the current child-request and cancellation flow, retaining web and mobile generic approvals.
Confidence high. Release: Not applicable. PR readiness: Needs work or a decision.
Limits. Historical pending approval rows are explicitly outside this PR. Canonical latest-commit checks do not include Check, Test, Mobile Native Static Analysis, Release Smoke. GitHub reports a merge conflict with current main.
Request. Prevent two servers from opening the same T3 data directory.
Audit finding. Main has no server-directory ownership lock. This patch uses exclusive file creation, but its stale-lock loop updates mtime without reading or comparing it and then unlinks after three scheduler yields, so the claimed heartbeat protection is not implemented. A successor can claim the path between the final read and remove, and the tests do not exercise that race.
Recommendation. Keep open: work remains. Replace the stale-lock reclaim algorithm and add a controlled concurrent-successor test before review.
Confidence high. Release: Not applicable. PR readiness: Needs work or a decision.
Request. Show the current thread provider quota near the composer through an opt-in setting.
Audit finding. Main has neither the underlying quota snapshot nor the composer quota control. This PR is explicitly stacked on the quota PR and adds a distinct current-provider meter, so it is not a duplicate of the Usage page proposal. The inspected setting is off by default and the meter waits for a started turn, but the full shared probe dependency remains unmerged and its Check job is failing.
Recommendation. Keep open: work remains. Review the composer-only delta after the underlying quota PR is corrected and selected.
Confidence medium. Release: Not applicable. PR readiness: Needs work or a decision.
Check: GitHub check. Head CI: Test success, Check failure, Mobile Native Static Analysis skipped, Release Smoke success.
Limits. The 64-file stacked diff was not fully reviewed. The unique composer meter, visibility logic, setting default, and shared PTY dependency were inspected.
Request. Render private GitHub attachment images through signed server asset URLs.
Audit finding. Main still loads GitHub images directly and cannot authenticate private attachments through the server. The diff fixes older-server and load-error fallback, but its redirect resolver leaves response bodies unread and the no-environment image branch drops expansion and copy behavior. Those current review findings match the latest code, so it needs changes before merge.
Recommendation. Keep open: work remains. Consume or cancel proxy responses and preserve expansion and copy behavior for direct fallback images.
Confidence high. Release: Not applicable. PR readiness: Needs work or a decision.
Limits. Per-repository authorization is disputed in review. This audit does not settle that access-policy question. GitHub reports merge conflicts with the target branch. Latest head has failing checks: Macroscope - UI Consistency. Failure logs were not triaged in this audit.
Request. Add a clipboard paste action to the mobile terminal toolbar.
Audit finding. Main has no toolbar paste action, so this feature is not already handled. Both unresolved review findings are supported by the diff: the encoder leaves unsafe control bytes intact, and the asynchronous clipboard result calls a captured writeInput after a terminal or route change. The current web Ghostty encoder explicitly sanitizes those bytes, unlike the proposed mobile helper.
Recommendation. Keep open: work remains. Fix control-byte sanitization and cancel stale clipboard reads, with focused tests for both cases.
Confidence high. Release: Not applicable. PR readiness: Needs work or a decision.
Comment: Discussion comment. Review discussion checked against the source.
Check: GitHub check. Head 8a2f764: Check, Test, and Release Smoke pass. Mobile Native Static Analysis is skipped. The unrelated Vercel marketing preview requires authorization.
Limits. The Android toolbar and native bracketed-paste behavior were not tested.
Request. Read OpenCode usage across standard, channel, and per-provider-instance databases.
Audit finding. Main has no OpenCode SQLite usage collector. This proposal includes both database schemas, reasoning normalization, source identity, instance overrides, and web/mobile display, so it is more than a presentation-only change. The reader uses synchronous SQLite queries and a one-second busy wait on the server thread, and the service integration predates the latest incremental scan refactor.
Recommendation. Keep open: work remains. Move database queries and lock waits off the server event loop, then rebase the collector onto the current scan service.
Confidence high. Release: Not applicable. PR readiness: Needs work or a decision.
Pr: PR #7221. The overlapping proposal already uses a worker for SQLite reads.
Limits. Canonical latest-commit checks do not include Check, Test, Mobile Native Static Analysis, Release Smoke. GitHub reports a merge conflict with current main.
Request. Limit V2 startup and shutdown recovery reads to threads with recoverable runtime state.
Audit finding. The reviewed feature commit replaces a full shell-history walk with an explicit SQL recovery-candidate query and a memory-store predicate. Pinned main still uses the earlier orchestration engine and does not contain ProviderRuntimeRecoveryService. The V2 base and this head are not ancestors of pinned main, so recent main memory fixes do not close this branch-specific change.
Recommendation. Keep open: work remains. Review the six-file recovery delta against the current V2 branch and its recovery-state coverage.
Confidence medium. Release: Not applicable. PR readiness: Needs work or a decision.
Observed GitHub metadata. GitHub state OPEN. Author saphid. Updated 2026-09-01T05:36:36Z. Draft yes. Target t3code/codex-turn-mapping. Head a81d00a4541433723049ef4901478f33427b01fd. Branch codex/fix-v2-recovery-scan. Size +189920 / -99932, 926 files, 316 commits. Mergeability CONFLICTING. Merge state DIRTY. Target is not main. Metadata snapshot 2026-09-01T11:57:41.491617+00:00.
Commit: Commit a81d00a45414. Reviewed the owned recovery query and service delta, separate from its large base diff.
Limits. The 926-file feature-branch base diff was not fully reviewed. The memory-store implementation does not cover outbox-only recovery candidates. GitHub reports merge conflicts at the reviewed head.
Request. Refresh Windows PATH for existing provider instances, Claude session starts, and installer detection.
Audit finding. Main still keeps custom provider environments and Claude executable resolution from adapter creation, so the broader Windows refresh fix remains needed. The PR addresses those lifetimes, but exact-head ProviderRegistry tests still use the removed maintenanceCapabilities property in two typed fake instances. Its refresh effect also does not pin the host references used when the registry was created. Fix those leftovers and align its lazy maintenance API with the installer PR.
Recommendation. Keep open: work remains. Migrate the two stale registry fakes and pin the refresh effect to its captured host references.
Confidence high. Release: Not applicable. PR readiness: Needs work or a decision.
Request. Apply hidden preview viewport changes natively and prevent stale timeout rollback.
Audit finding. Main resize only saves a viewport snapshot and has no server compare-and-set token. The reviewed server and contract changes add a versioned rollback guard, while the broader PR adds CDP viewport intent handling. The body still requires the desktop status-schema prerequisite and a manual Electron pass.
Recommendation. Keep open: work remains. Integrate the status-schema prerequisite and verify hidden resize plus concurrent rollback in packaged Electron.
Confidence medium. Release: Not applicable. PR readiness: Needs work or a decision.
Limits. Native viewport application and guest replacement were not reproduced. The 133,936-byte diff was not fully read. Review covered the named feature paths and current main, not every file or test. Current review findings and general discussion were sampled, but historical resolved review threads were not read in full.
Request. Count usage from all configured Claude and Codex homes without double-counting overlap across environments.
Audit finding. Main still scans only legacy provider homes and then merges per-provider totals using per-directory claims. The PR fixes instance enumeration and now honors the effective environment, but its overlap rule drops an entire environment contribution, including unique homes when neither set contains the other. The tests explicitly accept that lost coverage, so it does not fully replace per-home attribution work.
Recommendation. Keep open: work remains. Separate home enumeration from cross-environment attribution and decide how to retain unique-home usage.
Confidence high. Release: Not applicable. PR readiness: Needs work or a decision.
Limits. Required checks have no results on this head: Test, Check, Mobile Native Static Analysis, Release Smoke. An older related reference could not be retrieved from GitHub. It is excluded from verified dependencies and fix evidence.
Request. Release snapshot and debugger work after timeout so later preview automation can recover.
Audit finding. Main still runs capturePage inside the serialized control session without an end-to-end snapshot deadline. The reviewed IPC and host helper add one remaining-time budget and a tagged timeout result. The larger background-automation branch changes the same capture and teardown code, so these fixes need an explicit integration choice.
Recommendation. Keep open: work remains. Choose one snapshot recovery implementation and verify a hung capture followed by successful evaluate and snapshot calls.
Confidence medium. Release: Not applicable. PR readiness: Needs work or a decision.
Pr: PR #8981. The open background branch also changes bounded capture and recovery.
Limits. The 125,666-byte diff was not fully read. Review covered the named feature paths and current main, not every file or test. GitHub reports merge conflicts with the current base.
Request. Confirm preview key delivery and keep native keys away from stale guests or host shortcuts.
Audit finding. Main still focuses the guest and sends CDP key packets without delivery receipts or document quarantine. The reviewed key contract changes input to named keys and printable ASCII, with Unicode text left to preview_type. The full manager change is large and must be reconciled with composer focus and annotation focus fixes.
Recommendation. Keep open: work remains. Verify the native receipt path on each desktop OS before integrating the overlapping focus fixes.
Confidence medium. Release: Not applicable. PR readiness: Needs work or a decision.
Pr: PR #4118. Composer DOM-focus restoration is a separate open patch.
Limits. No native keyboard fixture was run. The 211,882-byte diff was not fully read. Review covered the named feature paths and current main, not every file or test. Current review findings and general discussion were sampled, but historical resolved review threads were not read in full. GitHub reports merge conflicts with the current base.
Request. Reject hidden preview clicks and report whether native mouse input was sent.
Audit finding. Main sends mousePressed and mouseReleased without a visible-presentation guard or explicit dispatch receipt. The reviewed broker change requires a new host capability and a dispatch marker, which is a useful rollout boundary rather than a landed fix. The large desktop and host implementation still needs complete review and native verification.
Recommendation. Keep open: work remains. Run the guarded-click fixture against hidden, replaced, and old-host targets before integration.
Confidence medium. Release: Not applicable. PR readiness: Needs work or a decision.
Limits. The PR body states that its native browser fixture was not run. The 131,300-byte diff was not fully read. Review covered the named feature paths and current main, not every file or test. Current review findings and general discussion were sampled, but historical resolved review threads were not read in full. GitHub reports merge conflicts with the current base.
Request. Choose a shell when creating a web or desktop terminal.
Audit finding. Main has no shell override in the terminal contract or shell-list RPC. The patch adds both picker locations, but only open carries the chosen shell, leaving the reviewed attach-first and evicted-session cases able to use or restart into the default shell. Git Bash also receives no login arguments, and the Settings default promised by the body is absent from the final diff. Earlier default-reporting and explicit-restart findings are corrected in the code even though some review threads remain open.
Recommendation. Keep open: work remains. Finish selected-shell persistence across attach and restore, Git Bash startup, and the stated default-shell scope before merging.
Confidence high. Release: Not applicable. PR readiness: Needs work or a decision.
Limits. No shell-picker or Windows Git Bash runtime was executed. Required Check, Test, Release Smoke, Mobile Native Static Analysis checks are absent on the collected head.
Request. Redesign the provider editor and make long model lists easier to manage.
Audit finding. The prerequisite settings cleanup is merged, but main still has the capped model list and the old boolean-only Fast mode label. The inspected remaining diff adds filtering, grouped ordering, visibility switches, and a separate configuration grid, so it is not fully supplied by the earlier cleanup. Model order persistence and the larger editor layout still need review after restacking.
Recommendation. Keep open: partial fix. Rebase onto the merged settings cleanup and review only the remaining editor redesign.
Confidence medium. Release: In stable source. PR readiness: Needs work or a decision.
Request. Choose any registered Git worktree, including detached checkouts, when starting a web or desktop thread.
Audit finding. Main still has only the previous-worktree shortcut and no worktree-list RPC. This diff adds the broader picker, but the exclusion check resolves cwd rather than the checkout root and compares it with only normalized listed paths. Detached worktrees at the same commit also have identical pathless web labels. Both details need correction before the shared RPC and mobile follow-up proceed.
Recommendation. Keep open: work remains. Resolve the checkout root and listed paths consistently, and distinguish detached worktrees in the web picker.
Confidence high. Release: Not applicable. PR readiness: Needs work or a decision.
Pr: PR #8568. Native mobile follow-up depends on this worktree contract and RPC.
Comment: Discussion comment. The current exclusion check does not use the same canonical form on both sides.
Limits. Latest head has no results for required checks: Test, Check, Mobile Native Static Analysis, Release Smoke. GitHub reports merge conflicts with the target branch.
Request. Show Claude next-prompt suggestions in the empty web composer.
Audit finding. Main explicitly drops prompt_suggestion events and has no suggestion activity or composer state. The diff adds the event and persistence path, then enables a Tab-to-accept prompt on web and desktop while hiding suggestions on mobile and touch-sized web views. Default enablement, the lack of a T3 setting, and the mobile gap need a product decision before merge.
Recommendation. Keep open: decision needed. Decide default enablement and mobile behavior for provider prompt suggestions.
Confidence high. Release: Not applicable. PR readiness: Needs work or a decision.
Commit: Commit 6f34ad3e87eb. The earlier SDK coverage change deliberately left suggestions without a client UI.
Limits. The web composer interaction was not run in this read-only audit. Latest-head required checks are not reported: Test, Check, Mobile Native Static Analysis, Release Smoke. GitHub reports merge conflicts against the target branch. An older related reference could not be retrieved from GitHub. It is excluded from verified dependencies and fix evidence.
Request. Put account-wide Connect removal in Connections and prevent automatic host resume from restoring removed access.
Audit finding. Merged account-page deregistration does not provide Connections rows or prevent automatic restoration of revoked links. This diff adds those controls and intent checks, but the mobile Live Activity preference path still calls the new explicit relink path and can restore a removed environment without a link action. Its account-membership guard also discards cached removal state during refresh, so the final review still has material gaps.
Recommendation. Keep open: partial fix. Separate preference updates from explicit relinking and preserve known account membership during refresh.
Confidence high. Release: In stable source. PR readiness: Needs work or a decision.
Comment: Discussion comment. Confirmed preference updates can restore a revoked link.
Limits. The author used isolated relay and account fixtures, not real-account revocation or live Cloudflare cleanup. Required current-head checks are absent: Test, Check, Mobile Native Static Analysis, Release Smoke. GitHub reports merge conflicts with the current base. Current review checks still fail: Macroscope - Effect Service Conventions.
Request. Support projects rooted below a Git repository in checkpoints and review diffs.
Audit finding. Main still tests for a .git entry at the exact project directory before checkpointing and rejects registered review roots outside the server root. This diff fixes those paths and generates untracked-file patches with repository-relative names. PR #8559 and PR #8874 overlap parts of the change but do not cover the full nested-workspace behavior.
Recommendation. Keep open: work remains. Consolidate the overlapping review and checkpoint fixes while retaining the nested untracked-file tests.
Confidence high. Release: Not applicable. PR readiness: Needs work or a decision.
Request. Discover picker skills in the selected project or worktree instead of the server process workspace.
Audit finding. Main still reads picker skills from provider health snapshots and has no provider.listSkills RPC. This patch resolves the project or thread workspace, uses a five-second per-instance cache, and updates web and mobile, but includes the entire unmerged explicit-invocation prerequisite. The unresolved plan-follow-up gap belongs to that shared stack, so this is not a duplicate or a landed fix.
Recommendation. Keep open: work remains. Finish the explicit-invocation prerequisite, then review the workspace-discovery-only diff.
Confidence high. Release: Not applicable. PR readiness: Needs work or a decision.
Request. Render completed Mermaid fences as diagrams while preserving source copy and plan previews.
Audit finding. Main still renders these fences as code and truncates collapsed plans by line count without tracking fences. This patch adds lazy SVG rendering, a source toggle, bounded caching, and fence-aware previews across the shared web renderer. Another open Mermaid implementation uses serialized rendering, so the two designs should be consolidated before enabling diagrams by default.
Recommendation. Keep open: decision needed. Choose one Mermaid implementation with the earlier proposal and verify its render cost and copy behavior.
Confidence high. Release: Not applicable. PR readiness: Needs work or a decision.
Request. Name providers in a toast when more than one provider update is available.
Audit finding. Main still uses a generic install description for one-click updates, while manual-only updates list provider names. The earlier toast cleanup changed progress display, not this text. This resubmission remains a copy decision after the earlier proposal was closed, not a fixed defect.
Recommendation. Keep open: decision needed. Decide whether to use the provider-name list in one-click update toasts.
Confidence high. Release: Not applicable. PR readiness: Needs work or a decision.
Request. Open provider settings on Models instead of Configuration by default.
Audit finding. The provider editor still defaults to Configuration. The later settings cleanup deliberately moved Configuration first to match that default, so this is an unselected UI choice, not a landed fix. The single-line PR still changes the requested behavior and keeps unknown drivers on Configuration.
Recommendation. Keep open: decision needed. Decide the default provider editor tab after the merged settings cleanup.
Confidence high. Release: Not applicable. PR readiness: Needs work or a decision.
Request. Insert local filesystem paths for dropped non-image files in desktop-managed threads.
Audit finding. Main already uploads general file drops through the attachment flow. On the latest PR head, path insertion never runs because videoMimeType returns null for non-videos but the filter compares with undefined. Correcting that filter still leaves the proposed browser/remote upload behavior and WSL path mapping to resolve.
Recommendation. Keep open: decision needed. Fix the attachment classifier while retaining browser and remote uploads and translating or rejecting WSL paths.
Confidence high. Release: Not applicable. PR readiness: Needs work or a decision.
Comment: Discussion comment. The current filter sends every file through attachment upload, leaving path insertion unreachable.
Limits. Latest head has no results for required checks: Test, Check, Mobile Native Static Analysis, Release Smoke. Latest head has failing checks: Macroscope - UI Consistency. Failure logs were not triaged in this audit.
Request. Show review diffs for registered checkout projects outside the server root without substituting another repository.
Audit finding. Main still limits review roots to the server cwd or worktrees directory and then retries rejected web requests at the server cwd. This diff adds an exact active-project existence query and removes that fallback. PR #8527 overlaps the authorization change but also changes checkpoint and untracked-file behavior.
Recommendation. Keep open: work remains. Use one shared review-root fix with PR #8527 and retain the metadata-free authorization query.
Confidence high. Release: Not applicable. PR readiness: Needs work or a decision.
Request. Upgrade the web and Android terminal engine and migrate its changed C and WASM APIs.
Audit finding. Main still uses Ghostty 9f62873 and the old constructors, mode reads, and WASM allocators. This patch upgrades to 76e568b and gates unused features, while PR #8767 proposes a different upgrade with explicit byte and line limits. This diff sets only the line option and keeps a second Android revision constant, so the final upgrade should reconcile those choices rather than treating the PRs as equivalent. iOS is separate.
Recommendation. Keep open: work remains. Choose one Ghostty upgrade patch that includes the agreed scrollback limits and one canonical revision for both builders.
Confidence high. Release: Not applicable. PR readiness: Needs work or a decision.
Request. Choose existing registered worktrees in the native mobile new-task flow.
Audit finding. Main still has no existing-worktree picker in the mobile new-task flow. The diff preserves a detached branch during creation, but its unchanged branch picker still replaces null with the project current/default branch. It also inherits the parent PR #8517 current-checkout exclusion problem, so the stack needs both fixes before native verification.
Recommendation. Keep open: work remains. Preserve detached selection in the branch picker after correcting the parent worktree listing.
Confidence high. Release: Not applicable. PR readiness: Needs work or a decision.
Limits. No native mobile screenshots or integrated verification evidence is supplied in the PR body. Latest head has no results for required checks: Test, Check, Mobile Native Static Analysis, Release Smoke. GitHub reports merge conflicts with the target branch.
Request. Let a user schedule automatic continuation after a provider usage limit resets.
Audit finding. Main has no durable usage-limit resume state, timer commands, or resume action. The patch adds provider error classification, persistent schedules, retry backoff, lifecycle cancellation, and hidden continuation turns, which are separate from recent recovery and settling fixes. The author presents it as a working idea, and its automatic work and web-only control need a product decision before full orchestration review.
Recommendation. Keep open: decision needed. Decide whether to ship user-armed automatic continuations and then complete the lifecycle and recovery review.
Confidence high. Release: Not applicable. PR readiness: Needs work or a decision.
Limits. The schedule contracts, retry classifier, decider, and reactor paths were inspected, but the full 53-file diff and provider tests were not reviewed line by line. The PR intentionally leaves mobile without schedule controls. The full historical review discussion was not read; the body, current thread status, and available maintainer direction were checked. Canonical latest-commit metadata has no results for these required checks: Test, Check, Mobile Native Static Analysis, Release Smoke.
Request. Transfer compatible Codex sessions between provider instances without overlapping native writers or changing thread identity.
Audit finding. ProviderService still starts the target instance before stopping stale sessions, and stale-stop failures are logged rather than propagated. Codex resume can also fall back to thread/start, so main does not guarantee same-thread, single-writer handoff. The PR changes release proof, locks, cursor persistence and stale-owner filtering across shared session paths.
Recommendation. Keep open: work remains. Complete an independent lifecycle review of handoff, failed cleanup and out-of-band approval responses.
Confidence high. Release: Not applicable. PR readiness: Needs work or a decision.
Limits. The handoff and lock design was sampled, but the 168 KB nine-file lifecycle diff was not fully reviewed. Latest-head required check results are absent for Test, Check, Mobile Native Static Analysis, Release Smoke.
Request. Add OMP as a built-in ACP provider and document Claude use with Z.AI.
Audit finding. Main has no OMP driver, and a generic ACP integration remains a separate open proposal. The inspected branch adds a dedicated OMP adapter plus an unrelated Claude Z.AI guide. The author confirms that canceled-turn markers remain retained after a cleanup attempt caused double settlement, so that known lifecycle concern is not fixed merely because its review thread says resolved. Keep the provider decision separate from the docs and finish the adapter review before merge.
Recommendation. Keep open: decision needed. Split the Z.AI guide from the OMP provider and decide the provider integration path.
Confidence medium. Release: Not applicable. PR readiness: Needs work or a decision.
Source: docs/user/providers-claude.md. The current Claude guide documents custom endpoint setup but not this Z.AI walkthrough.
Comment: Discussion comment. The author confirms the cancellation-marker concern remains after reverting the attempted fix.
Limits. The 24-file provider diff was not fully reviewed. Driver wiring, ACP authentication and model selection, and the separate Z.AI guide were inspected. Required checks have no results on this head: Test, Check, Mobile Native Static Analysis, Release Smoke.
Request. Reuse an opt-in administrative development credential across worktree ports without sharing databases.
Audit finding. Main has no reusable development credential. The diff adds a separate long-lived admin token, preserves ordinary credential precedence, and changes pairing and environment identity checks, so the auth design still needs a decision. After the pinned snapshot, merged PR #8367 moved initial session configuration to the subscription snapshot, so this PR must move its environment-ID guard to that new path. Initial config still has no equality check against the prepared environment ID. The PR adds that check to a unary bootstrap block that the new merge removes.
Recommendation. Keep open: decision needed. Port the environment-ID guard to the initial subscription snapshot before completing the reusable-credential review. Move the identity guard before readiness in the owned snapshot path. Keep the development-credential design decision.
Confidence high. Release: Not applicable. PR readiness: Needs work or a decision.
Commit: Commit b883fc066ea5. Verified post-snapshot main change accepts initial configuration from subscribeServerConfig without the proposed environment-ID guard.
Latest main change. Initial config still has no equality check against the prepared environment ID. The PR adds that check to a unary bootstrap block that the new merge removes. Move the identity guard before readiness in the owned snapshot path. Keep the development-credential design decision.
Limits. No real browser or Tailscale multi-worktree verification was run. GitHub reports merge conflicts with the target branch.
Request. Give remote Stop immediate feedback and prevent a turn from remaining stuck after interruption.
Audit finding. Main still has no pending state on the main turn Stop path. The diff only sets and resets an isStoppingTurn value that no rendered control reads, and it leaves the stated server fallback, missing-turn-ID handling, and mobile work unimplemented. Most changed files instead sanitize OpenCode terminal escapes and shared model strings, which is a separate concern.
Recommendation. Keep open: work remains. Split the OpenCode parsing change out and complete the visible Stop behavior with focused interruption tests.
Confidence high. Release: Not applicable. PR readiness: Needs work or a decision.
Issue: Issue #8618. The reported remote Stop behavior needs more than an unused state variable.
Limits. Required current-head checks are absent: Test, Check, Mobile Native Static Analysis, Release Smoke. GitHub reports merge conflicts with the current base.
Request. Reveal aligned SwiftUI message timestamps with a deliberate leftward swipe.
Audit finding. The inspected SwiftUI base has no transcript-wide timestamp reveal, and the standalone SwiftUI client is not on main. This patch handles gesture ownership, streaming identity, bottom anchoring, and VoiceOver metadata, including the four reviewed fixes. The author removed broken media links and explicitly says current-head motion evidence is still missing.
Recommendation. Keep open: work remains. Capture current-head timestamp-swipe evidence and review it as part of the SwiftUI prerequisite stack.
Confidence high. Release: Not applicable. PR readiness: Needs work or a decision.
Observed GitHub metadata. GitHub state OPEN. Author saphid. Updated 2026-08-30T02:27:21Z. Draft no. Target t3code/rebuild-mobile-app-swift. Head bfb058405a28d78e87eec3923475720180744507. Branch swiftui/pr-112-timestamps. Size +946 / -15, 2 files, 4 commits. Mergeability MERGEABLE. Merge state CLEAN. Target is not main. Metadata snapshot 2026-09-01T11:57:41.491617+00:00.
Request. Explain enabled but unreachable environments in the experimental SwiftUI new-task picker.
Audit finding. The four-file patch adds unreachable-environment notices, retry state, and routing when no reachable project exists. It targets the still-open SwiftUI rebuild, and pinned main has only the React Native mobile app, not apps/swift-ios. The change therefore remains a stack follow-up and cannot be treated as a fix for the shipped mobile client.
Recommendation. Keep open: work remains. Integrate and verify this patch against the current SwiftUI rebuild parent.
Confidence high. Release: Not applicable. PR readiness: Needs work or a decision.
Observed GitHub metadata. GitHub state OPEN. Author saphid. Updated 2026-08-29T13:19:36Z. Draft no. Target t3code/rebuild-mobile-app-swift. Head 77c3bf8aabf5b4187f73a4c8816dff69eccbe589. Branch swiftui/pr-131-unreachable-environments. Size +449 / -52, 4 files, 4 commits. Mergeability MERGEABLE. Merge state CLEAN. Target is not main. Metadata snapshot 2026-09-01T11:57:41.491617+00:00.
Request. Add individual path, branch, and thread ID copy actions to SwiftUI thread menus.
Audit finding. The SwiftUI base row menu has no Copy submenu, while main does not yet contain the standalone SwiftUI client. The patch uses existing row context, preserves raw copied values, and handles missing path and environment metadata. The visible menu is narrower than its helper, which also builds unused project, environment, and URL actions.
Recommendation. Keep open: work remains. Review the three visible copy actions within the SwiftUI stack and remove unused helper scope.
Confidence high. Release: Not applicable. PR readiness: Needs work or a decision.
Observed GitHub metadata. GitHub state OPEN. Author saphid. Updated 2026-08-30T02:23:31Z. Draft no. Target t3code/rebuild-mobile-app-swift. Head 2cc5a7e28c9d093b44c255495f20304621f61f6b. Branch swiftui/pr-140-thread-copy-actions. Size +508 / -4, 5 files, 6 commits. Mergeability MERGEABLE. Merge state CLEAN. Target is not main. Metadata snapshot 2026-09-01T11:57:41.491617+00:00.
Request. Track SwiftUI title regeneration and suppress duplicate or stale title updates.
Audit finding. The SwiftUI base already exposes Regenerate title but only dispatches and refreshes without pending-request tracking. This patch adds receipts, shell reconciliation, recovery, and disabled in-progress actions, so the existing menu is not complete coverage. Stale dispatch handling still needs review because finishDispatch matches only thread ID while shell updates can replace the pending request.
Recommendation. Keep open: work remains. Guard dispatch outcomes by request identity and verify overlapping old and new requests before integrating the SwiftUI stack.
Confidence high. Release: Not applicable. PR readiness: Needs work or a decision.
Observed GitHub metadata. GitHub state OPEN. Author saphid. Updated 2026-08-29T13:57:06Z. Draft no. Target t3code/rebuild-mobile-app-swift. Head ba8a1d2f5fab7108ced1cde7e35d99681e724ffd. Branch swiftui/pr-141-title-regeneration. Size +1336 / -36, 13 files, 7 commits. Mergeability MERGEABLE. Merge state CLEAN. Target is not main. Metadata snapshot 2026-09-01T11:57:41.491617+00:00.
Commit: Commit b67837984e56. The inspected SwiftUI stack base has no timestamp reveal or Copy submenu and only a basic title-regeneration call.
Pr: PR #5178. The standalone SwiftUI client is still an open prerequisite.
Pr: PR #8623. The submitted FeatureTitleRegenerationTracker.finishDispatch accepts threadID and a receipt but no requestID.
Check: GitHub check. Current-head Test, Check, and Release Smoke succeed. Mobile Native Static Analysis is skipped.
Limits. The target SwiftUI tree exists only on an unmerged stack, not pinned main. The PR states that its media is not from the exact current commit. The overlapping-dispatch concern is based on source review, not a native runtime reproduction.
Request. Display provider reasoning summaries in collapsible web timeline rows.
Audit finding. Main ingestion still accepts only assistant_text as assistant message content, so reasoning deltas do not reach the timeline. The inspected PR adds a reasoning channel, requests provider summaries, and excludes those messages from answer semantics, but mobile deliberately hides them. Older clients strip the channel and treat reasoning as ordinary answers, making version-skew behavior part of the required review.
Recommendation. Keep open: work remains. Review reasoning and answer separation across current and older clients before integration.
Confidence medium. Release: Not applicable. PR readiness: Needs work or a decision.
Limits. The 34-file reasoning diff was not fully reviewed. Channel contracts, provider summary requests, timeline rendering, and current ingestion were inspected. Codex and OpenCode were not manually verified by the author, and mobile has no reasoning UI. Required checks have no results on this head: Test, Check, Mobile Native Static Analysis, Release Smoke. An older related reference could not be retrieved from GitHub. It is excluded from verified dependencies and fix evidence.
Request. Reject WebSocket-only turn bootstrap data at the HTTP contract boundary.
Audit finding. Main advertises the full client command union on HTTP but does not execute bootstrap there. This patch addresses the maintainer's old handler-layer objection by using a narrower HTTP schema, but that change-request review has not been renewed. The separate open HTTP parity proposal still requires a choice between supporting bootstrap and rejecting it.
Recommendation. Keep open: decision needed. Choose schema rejection or shared HTTP bootstrap support before merging either proposal.
Confidence high. Release: Not applicable. PR readiness: Needs work or a decision.
Request. Suppress phone alerts while a client is focused but continue publishing activity state.
Audit finding. Main publishes state without a notify flag, and the relay still derives alerts from every attention transition. The diff signs the new flag, preserves old-publisher defaults, and tests silent Live Activity updates and skipped APNs pushes. Its policy suppresses alerts for all threads when any visible client is focused, so it needs a product decision about global focus rather than per-thread attention.
Recommendation. Keep open: decision needed. Approve the global-focus notification policy and integrate the signed server-to-relay change.
Confidence high. Release: Not applicable. PR readiness: Needs work or a decision.
Request. Add a default-on native Acrylic background option to supported Windows desktop builds.
Audit finding. Main uses solid native window backgrounds and only renderer-level glass opacity. The patch adds a Windows 11 build gate, persisted Appearance control, native state reporting, and explicit translucent app containers. the reviewed auxiliary-window, opacity-stacking, and popup issues are corrected in the final diff. It changes default window creation across Windows and remains in conflict, so it needs a product decision and an integrated supported/unsupported-Windows check, not closure as already fixed.
Recommendation. Keep open: decision needed. Decide whether default-on Windows Acrylic is accepted before rebasing and verifying the native fallback.
Confidence high. Release: Not applicable. PR readiness: Needs work or a decision.
Limits. The manual Windows 11 report was not repeated, and no unsupported-Windows or overlay rendering check was run. Required Check, Test, Release Smoke, Mobile Native Static Analysis checks are absent on the collected head. The current head conflicts with main, and post-rebase checks are not available.
Request. Preserve execute permissions for resource-monitor files in published CLI packages.
Audit finding. Main has no publishConfig.executableFiles and the release manifest builder copies an explicit field list, so both parts of the packaging gap remain. This patch adds the four shipped paths and a real local pack-mode test. Its author marks it as a CI-only mirror and explicitly says not to merge it on GitHub.
Recommendation. Keep open: decision needed. Ask the author to identify the upstream landing path for the packaging change.
Confidence high. Release: Not applicable. PR readiness: Needs work or a decision.
Limits. The separate canonical repository was not inspected. Required CI checks are absent at the reviewed head: Test, Check, Mobile Native Static Analysis, Release Smoke.
Request. Scope Codex model catalogs to the provider instance and avoid invented custom-model capabilities.
Audit finding. Main still appends custom models with the first catalog model's capabilities and retains previous model metadata. The PR addresses provider scoping, but its unchanged previousBySlug merge can restore old capabilities when a new custom entry reports null. Its body explicitly identifies this as a CI-only mirror and forbids a GitHub merge.
Recommendation. Keep open: decision needed. Ask the author for the intended upstream change after fixing stale capability restoration, without merging this CI mirror.
Confidence high. Release: Not applicable. PR readiness: Needs work or a decision.
Limits. The external review system named in the PR was not accessed. Latest-head required check results are absent for Test, Check, Mobile Native Static Analysis, Release Smoke.
Request. Show OpenCode context use, cumulative tokens, and session cost in the meter.
Audit finding. Main does not emit OpenCode token usage or render session cost in the meter. This proposal includes model-limit retry and cumulative cost, but defines usedTokens as input plus cache while the other meter proposal also counts output and reasoning. It adds the cost row only on web, and metadata calls can block the sequential event handler for up to five seconds.
Recommendation. Keep open: work remains. Agree on context-token semantics, move catalog loading off event handling, and decide mobile cost display before consolidation.
Confidence high. Release: Not applicable. PR readiness: Needs work or a decision.