Repository navigation
Codex fork 0.156.0
Codex fork 0.156.0
This is a main-based fork release following the corrected 0.155.1 runtime, not a byte-for-byte rebuild of OpenAI's 0.156.0 tag.
Comparison baseline
- Previous fork runtime: 0.155.1, corrected source
1a82148f5d2cb846dadd976c2297260a8edc76cf. - Previous upstream-tracking base:
4981b6d01effb84d0a79faa5e41f5e06c4fc5ce7. - New upstream-tracking base:
cdd1d9e1bbc0cccc80eaa03e6ad0de176e11e5a6, corresponding to OpenAImain@7db578fca663b3e0026996e9c50104784d420744. - Reviewed runtime and SDK release source:
e8bab0d47f7c74642f4f98032f8f306bee1404ff. - New release tags:
rust-v0.156.0andsdk/go/v0.156.0. Earlier version tags remain unchanged.
The upstream section below covers all 194 non-merge commits between these pinned bases. OpenAI's official 0.156.0 changelog uses a different baseline and includes features already present in the previous fork release; those are not claimed as new here.
Upstream changes since the fork's 0.155.1 base
Terminal UI and transcript
- Fullscreen transcript mode is enabled by default in this pinned main snapshot;
/tuiselects the terminal UI mode for the next launch. - Added transcript search, compact browsing and prompt navigation, per-activity detail controls, selection/copying, plain-click links, and right-click copying for transcript/composer selections.
- Added mouse selection and editing to the fullscreen composer, preserving drafts until sessions are ready and retaining streamed answers when subagents finish.
- Improved transcript ordering, persisted activity details, viewport anchoring, wrapping, list spacing, selection/autoscroll behavior, and full URL destinations at different terminal widths.
- Added status filters, simpler navigation, and improved task metadata/layout in the agent command center; read-only agent sessions handle Escape navigation correctly.
- Added independent controls for visual effects and Mermaid/math/table rendering. Mermaid supports stadium nodes; terminal math supports aligned equations and optimization notation.
- Refined warning visibility, quota notices, picker consistency, clock preferences, keyboard hints, reduced-motion voice UI, usage-dashboard navigation, and layout.
- The welcome animation is limited to onboarding; completion hints and recap spacing are kept close to the transcript tail.
Agent lifecycle and collaboration
- Added persistent local agent message boards: channel posts/subscriptions, pagination, latest-post indexes, collaboration tools, and thread-deletion cleanup.
- Added attributed message previews, automatic subscription to newly posted channels, preservation of explicit unsubscribes, and suppression of notifications back to the post author.
- Consolidated lifecycle operations in
AgentControl/LocalAgentControl, with captured spawn/send requests and local runtime state separated from controller handles. - Agent inspection can read unloaded-agent metadata without restoring runtimes; status subscriptions stream agent snapshots, and close operations return snapshots.
- Addressed queued-message/eviction races, overlapped spawn persistence, cleaned up cancelled children, and skipped unnecessary stored-title reads for ephemeral forks.
- Conditional turn interruption preserves pending input. Subagents can request MCP elicitation input; delivered messages survive post-tool hook failures.
- Guardian uses thread context by default, supports extra policy configuration, retains ordered assistant context, and tolerates compatible compaction-hash differences for synchronous reviews.
Context, Code Mode, and runtime state
- Compaction checkpoints persist resume metadata; retained remote-compaction history excludes descendant channel posts.
- Code Mode shares stored JSON values across cells with
Arc, preserves empty-cell metadata under recorder pressure, and recovers executed-tool metadata under capacity pressure. - Yielded skill calls retain originating turn metadata; model catalogs can override Code Mode tool messages.
- MCP request attribution accumulates across requests and history; transport workers retain request trace context.
- Extension tool history is materialized lazily. Cloud skill catalogs refresh at turn startup and can be reused until invalidated.
- Thread creator identity and item lifecycle timestamps are persisted; plugin recommendations move into developer context.
- Added GPT-6 Sol/Luna model catalog entries, including Amazon Bedrock catalogs. Catalog metadata and service-tier availability were refreshed.
invalid_promptis preserved as its own error classification; image-generation failures retain request IDs.
API, plugins, authentication, and networking
- Added explicit gateway OAuth read/login/cancel operations, authentication status notifications, and the initialize capability for explicit gateway sign-in.
- Hardened gateway credential persistence and error redaction; MCP OAuth authorization endpoints are restricted to HTTP(S).
- MCP resource reads can target specific apps/accounts; MCP server status exposes HTTP origins.
- Plugin summaries expose hosted extensions such as entrypoints, icons, quick actions, search providers, and settings.
- Standalone web-search redirects and realtime WebSocket connections honor configured proxy routes; the network proxy accepts caller-provided MITM CAs.
- File blob upload timeout increased from one minute to five minutes; attachment uploads carry thread IDs.
- Temporary structured threads use read-only permissions; Unix local MCP processes are restricted to stdio descriptors.
- Environment updates are serialized; removed pending attachments are cancelled. Foreign working directories and required Windows environment variables are preserved.
- Daemon socket paths are masked through ancestor bind mounts. macOS filesystem helpers avoid fork; local child-process spawning moves into the shared PTY utilities.
Background server, diagnostics, and builds
- Automatic daemon startup is enabled by default; the TUI offers explicit recovery for incompatible background servers and supports
/importwith remote/local daemon sessions. - Improved
codex doctordiagnostics for SQLite databases and path/URL check status without exposing configuration values. - Added rollout compression diagnostics and bounded report reason tags.
- Reduced unused dependency features and avoided building the host SQLite driver for SQLx macros.
- Removed upstream's
rust-release-prepareworkflow; clarified Bazel lock verification failures. - Added or updated focused regression tests across message boards, lifecycle ordering, snapshots, queues, and credential handling.
Fork reconciliation and Go SDK
All 129 fork commits were replayed without dropping a commit. Runtime adaptations follow the new AgentControl, SpawnAgentOptions, captured send requests, and core-owned submission types instead of restoring retired controller wrappers. Per-turn spawn accounting, legacy parent-completion watching, upstream cancelled-spawn cleanup, and the new Guardian/MCP persistence metadata are preserved together.
The fork-only agent status tool rename is rolled back: both collaboration APIs again advertise and dispatch the upstream wait_agent name. This avoids rejection of check_agent_status inside the reserved collaboration namespace. Wait behavior, timeout controls, namespace configuration, and historical trace recognition remain unchanged.
The Go SDK now exposes the actual stable and experimental app-server additions:
- Explicit gateway OAuth read/login/cancel calls and status notifications.
ClientConfig.ExplicitGatewayOAuthis opt-in; its exact Rust wire field isexplicitGatewayOauth, including default-false and omission behavior. Managed login captures the authorization URL before waiting for the blocking RPC and handles RPC completion/error without a notification. Cancellation stops an unsent login locally or follows the already-sent login with a connection-scoped cancel request. Only one managed login is active per client; failed cancellation retains ownership, and an old handle cannot cancel a subsequent one. - MCP HTTP origins and explicit resource targets, including required-nullable
linkIdfor selected-account versus no-auth reads. - Plugin extension entrypoints, quick actions, search/settings metadata, tagged unions, flattened fields, and raw JSON. Variant-specific defaults no longer leak into other union variants.
- Nullable item start/completion timestamps with legacy omitted-field support.
Existing SDK methods and definitions remain present. Stable and experimental protocol/schema/manifest digests were regenerated; use the matching runtime and SDK together. The module path and explicit compatibility-override policy remain unchanged. No earlier SDK tag is moved.
Gateway login is registered on the app-server before background dispatch, so an early cancellation cannot acknowledge success and then leave an already-accepted login starting afterward. Dropped queued requests release their admission slot; connection ownership and notification opt-out checks remain enforced.
Existing fork capabilities retained
These are cumulative fork differences, not all-new 0.156.0 features. The targeted preservation checks are described below.
- Independent
mcp_max_linesoverride for direct MCP results and the whole emitted Code Mode result, including exec/wait/notify; ordinary direct tools retain the general line limit. - Original byte/token policies retained through saved metadata, model switches, and replay; raw nested JavaScript values and typed/encrypted MCP results preserved.
- Custom providers and typed/fallback context-window/auto-compaction overrides; configurable cross-provider plaintext messages and inherited dynamic tools.
- Cumulative per-turn spawn budgets, race-safe publication accounting, quiet status waits, and parent follow-up completion notifications.
- Compaction headroom, retained-history bounds, cache/prefill consistency, no-progress failure reporting, and at least three capacity retries.
- Tolerant malformed usage metadata handling and the expanded 2,097,152-character user-input limit.
- Typed Go SDK resource clients, request/event routing, strict runtime compatibility, preserved filtered completion backlog, realtime closure, image paths, and JSON Schema forwarding.
- Direct-only Code Mode routing recovery, explicit tool omissions, consistent instructions, and warnings for unobserved nested results.
Downloadable packages
Exactly five system archives, each containing bin/ with unsuffixed executable names:
- Linux GNU/glibc x86_64 and ARM64:
codex,codex-app-server,codex-code-mode-host,codex-responses-api-proxy,bwrap. - macOS x86_64 and ARM64:
codex,codex-app-server,codex-code-mode-host,codex-responses-api-proxy. - Windows x86_64:
codex.exe,codex-app-server.exe,codex-code-mode-host.exe,codex-responses-api-proxy.exe,codex-command-runner.exe,codex-windows-sandbox-setup.exe.
Linux packages use GNU/glibc, not musl. Optional native voice/audio resources and provisioned Windows sandbox-service packaging are outside this bin-only layout. A feature's presence upstream is not a claim that these optional resources are bundled here.
Upgrade note: automatic daemon startup is now enabled by default. Replacing binaries on disk does not update an already-running app-server; an older daemon can still cause configuration or protocol-handshake errors. Use a matching daemon version, or launch the CLI with --no-daemon to bypass it. The tool-name correction does not update or restart existing daemons.
SHA256 of the five release archives:
42f1cffd6413990e9e419dc70ac71c6f51b3d8636f7751101befad9cb08e201d codex-package-aarch64-apple-darwin.tar.gz
54b59176f1fac3847b3a87d945e4237181f615527ae4fb3b0f6796ddcd878b1e codex-package-aarch64-unknown-linux-gnu.tar.gz
d812366cc95480135bcd72c093afe0b8dfdd99afcf4efae208b9bf5428207b58 codex-package-x86_64-apple-darwin.tar.gz
79cbe3b3f0f545875f9256c6de2df1bf4682306b2d33769a60585fcf29f75a89 codex-package-x86_64-pc-windows-msvc.zip
9f6b2bddc494caf498e5d8f5717f680085497a0668d1697b7b0219a7126ce45d codex-package-x86_64-unknown-linux-gnu.tar.gz
Validation and limitations
- All 129 fork commits were replayed one-to-one: 102 patch-identical and 27 adapted, with no dropped or unmatched commits. The pinned upstream base is an ancestor of the new source.
- Core validation: 119 focused runtime checks, six reconciled scenario snapshots, nine subagent-elicitation replacement checks, and 87 Code Mode runtime tests passed. Six completion-watcher checks passed after fixing notification delivery when a child disappears before status subscription. Four critical direct-MCP/Code Mode/originating-budget model-input regressions were also executed explicitly without network skips.
- Adjacent Rust suites: 354 protocol tests and 92 history/output-truncation/Code Mode protocol tests passed. App-server protocol: 350 passed, one skipped. An app-server/Linux targeted selection passed 58 checks; three daemon namespace/socket checks were separately confirmed to execute without environment early-return guards.
- The server-side OAuth correction passed an all-target app-server compile check, ten Gateway OAuth checks, and seven connection-gate/turn-admission checks. The deterministic pre-handler-cancellation regression verifies that no authorization URL or OAuth HTTP request is produced.
- The later agent-tool naming correction passed 100 focused core checks, 262 analytics/prompts/rollout-trace/features checks, and 23 targeted core/app-server/exec checks. Six model-context snapshots were reviewed and updated for the canonical tool name and resulting hashes. A rebuilt CLI and fresh strict SDK handshake passed; installed user binaries were not modified.
- The full Go SDK race-enabled suite passed after the Gateway lifecycle corrections. Stable/experimental manifest and generator checks passed. A freshly built debug CLI reported
codex-cli 0.156.0; a fresh strict SDK/runtime handshake passed with the real-runtime guard and without a compatibility override. - App-server and config schema generation, Bazel lock verification, 52 GitHub script tests, scoped Clippy/fix, and repository formatting passed. The 0.156.0 version stamp changed only the version fields of local workspace packages; dependency changes inherited from upstream remain part of the new base.
- Full Rust core, workspace, and app-server suites were not run. Focused checks and compile checks are not presented as full-suite coverage. Tests used isolated/mock environments; no real OAuth browser login or paid provider call was exercised.
- Independent review covered the rebase and SDK changes; all three original findings were corrected and closed through targeted rechecks. The subsequent user-requested agent-tool naming correction passed a separate targeted review. Its single test-order expectation finding was corrected, the affected test passed, and the reviewer closed it on the exact final source above. No review blockers remain in the agreed scope; the broad rebase review was not repeated for this isolated correction.
- All five targets passed in manual run 35837840982, attempt 1, completed 2026-09-23 at 11:42:53 UTC. Every runner checked out the exact reviewed source and passed the packaged
codex-cli 0.156.0version assertion. - All five downloaded artifact sizes and SHA256 digests matched GitHub metadata. Outer ZIP CRC, release archive integrity, exact executable layout, ELF/Mach-O/PE architectures, and Unix 0755 permissions were verified. The downloaded Linux x86_64 CLI also passed a version smoke check in an isolated home. Only artifacts from this corrected-source run are included.
Source and full upstream changelog
- Previous fork release 0.155.1.
- Pinned upstream-base comparison.
- Official upstream 0.156.0 release notes.
Rebase rewrites older fork commits. The pinned-base comparison and the list below separate new upstream changes from those rewritten cumulative patches.
All 194 upstream non-merge commits since the previous pinned base
7db578fca6Preserve user prompt URL destinations across terminal widths (openai#47360)722dae7afdColor paths and URLs incodex doctorby check status (openai#47358)3ebfe7b300Reserve spacing after recaps at the transcript tail (openai#47355)559264d92ePreserveinvalid_promptas a distinct error classification (openai#47353)8edfca892dReuse cloud skill catalogs across turns until invalidated (openai#47350)9124e23e2cMove plugin recommendations into developer context (openai#47349)5121aeff13Overlap subagent spawn persistence and clean up cancelled children (openai#47348)df30941072Add GPT-6 Sol and Luna to Amazon Bedrock catalogs (openai#47347)d93909a939Skip stored title lookups for ephemeral forks (openai#47342)04fc75adbeAdd conditional turn interruption that preserves pending input (openai#47340)1b1b58f5a5Move the activity inspection hint into keyboard shortcut help (openai#47338)49e95cc73fAdd GPT-6 Sol and Luna to the model catalog (openai#47332)0a73d55b80Avoid redundant spacing above the transcript composer (openai#47331)3b7a8520efIdentify failed SQLite databases incodex doctoroutput (openai#47330)00d92ad3eeStandardize TUI menu description wording and punctuation (openai#47329)6cfe299849Preserve image generation request IDs on failure (openai#47327)2c2a42e65dRestrict MCP OAuth authorization endpoints to HTTP(S) (openai#47326)8a577399c7Keep report reason tags within Sentry limits (openai#47325)286d4ecf44Persist resume metadata in compaction checkpoints (openai#47323)5a11c45606Render TUI Markdown lists with Unicode bullets and checkboxes (openai#47322)13a966fc65Honor the system clock preference across more TUI timestamps (openai#47321)2c087a0cc2Fix Escape navigation in read-only agent sessions (openai#47320)a3275210bdMatch agent status symbols to the selected row style (openai#47319)d6093d3228Offer explicit recovery for incompatible background servers (openai#47318)7e6c4c20fdEnable/importin remote and local daemon sessions (openai#47317)44b857c00eClean up legacy Guardian thread metadata in SQLite (openai#47303)4f6c106cddUse compact display metadata for Guardian review threads (openai#47302)5c07856e25Separate local agent runtime state from controller handles (openai#47301)c6c09fe29fConsolidate agent operations in theAgentControlimplementation (openai#47292)d9ac1f02edUpdate the Guardian extra-policy snapshot for retained instructions (openai#47287)89a7298affAdd regression coverage for message board unsubscribe persistence (openai#47281)279ba89415Enable Guardian thread context by default (openai#47275)99784bd09bAllow Guardian sync review across compaction hash differences (openai#47272)16bb1b9d7eImplementAgentControlforLocalAgentControl(openai#47271)ac7634b9f7Stream agent snapshots from local status subscriptions (openai#47267)a1f40f3f13Expose hosted plugin extensions in app-server summaries (openai#47263)d3e5840932Subscribe authors to message-board channels created by posting (openai#47259)3d5cdc51d0Test spawn settings reporting after child removal (openai#47258)0160eb3497Preserve explicit message-board unsubscribes when posting (openai#47257)fdd89e78acReturn agent snapshots fromclose_agent(openai#47252)c117207a6fDrop descendant channel posts from retained remote compaction v2 history (openai#47249)3b50349e1aAdd explicit app and account targets to MCP resource reads (openai#47248)c46aa58a7bCentralize agent resume handling inLocalAgentControl(openai#47244)acc20df49fPrevent V2 agent eviction from racing with queued messages (openai#47240)b58b05e062Suppress agent message board notifications to the post author (openai#47236)f908e5a968Route multi-agent spawning throughSpawnRequest(openai#47235)94174e44cbPreserve MCP request trace context across transport workers (openai#47212)064e701b0fAdd explicit gateway OAuth sign-in to app-server (openai#47207)6ea62c4396Resolve proxy routes for standalone web search redirects (openai#47204)b781733f95Separate thread preparation from execution persistence (openai#47199)639d2478ccRender aligned equations and optimization notation in TUI math (openai#47191)d1092865f8Allow forking conversations locked by another app in the TUI (openai#47185)75ec81c862Enable automatic daemon startup by default (openai#47179)5106a5234fEnable the fullscreen transcript by default (openai#47178)a37f260dc6Preserve empty cell metadata under pending tool-call pressure (openai#47174)f4389de279Add explicit gateway login control and authentication status (openai#47170)30daed37adApply workspace routing to Guardian v2 classifier requests (openai#47162)b23324986dExpose HTTP origins in MCP server status (openai#47159)9f22ba8039Harden gateway OAuth credential persistence and error redaction (openai#47158)bf7c87a7b7Ignore stale thread-close notifications after resuming a thread (openai#47155)c11ed24c2aUpdate code mode catalog snapshot hashes (openai#47149)517f51a6f7Extract exec-server CLI startup into a dedicated module (openai#47143)b045757052Honor system proxy settings for standalone web search (openai#47142)720a21ae66Prevent horizontal transcript selection from triggering autoscroll (openai#47137)e4dba902abSupport caller-provided MITM CAs in the network proxy (openai#47132)cf6754e682Remove theultrafastservice tier fromgpt-5.6-sol(openai#47130)69759baefdPreserve foreign working directories in extension tool environments (openai#47129)26cb4d73e2Add extra policy configuration for Guardian reviews (openai#47125)171dd0d39cIncrease OpenAI file blob upload timeout from 60 seconds to 5 minutes (openai#47122)4a8cfe6bb7Pass thread IDs to attachment uploads (openai#47121)5aa4939be4Support model-catalog overrides for Code Mode tool messages (openai#47118)3fd5160cd6Honor the configured product SKU in remote plugin requests (openai#47116)772abc9425Preserve and expose thread item lifecycle timestamps (openai#47114)23ae65bb06Persist thread creator identity in rollouts and SQLite (openai#47113)bf91058b81Lowercase the retry hint for conversations open in another app (openai#47112)023d81b057Preserve required Windows runtime variables for filesystem helpers (openai#47108)6a601c00daPreserve originating turn metadata for yielded skill calls (openai#47106)142360dac8Honor configured proxies for realtime WebSocket connections (openai#47101)8e7f50cba9Preserve assistant answers in bounded TUI task responses (openai#47100)e8cd975a55Keep TUI hints immediately above the composer (openai#47096)9c77996cd1Remove therust-release-prepareworkflow (openai#47095)afe4249cfeRestrict Unix local MCP servers to stdio descriptors (openai#47094)b4b30dd5ebCentralize JSON-RPC response serialization through payloads (openai#47089)09bb3d6942Disable unused default dependency features (openai#47088)c37b7307dbFormat analytics credit usage with consistent decimal precision (openai#47086)deb0a08f24Update model catalog descriptions and GPT-5.6-Sol priority (openai#47085)3627bb91b2Avoid building a host SQLite driver for SQLx macros (openai#47084)429afbcaa9Extract per-thread enrichment into a dedicated async helper (openai#47083)2fe4a2b4d8Show the voice toggle in the TUI shortcut overlay (openai#47082)ac98537678Track cumulative MCP attribution across requests and thread history (openai#47081)7cec3e110cMask daemon socket paths exposed through ancestor bind mounts (openai#47079)62ea6d41e8Serialize environment updates and cancel removed pending attachments (openai#47077)e51aacad60Include attributed previews in agent message board notifications (openai#47075)f3037cafd3Refresh host-supplied cloud skills at turn startup (openai#47074)20cdcb0156Wait for thread idle in Guardian context budget tests (openai#47073)f07aaf920bHighlight selected newlines in the transcript view (openai#47068)a4c8fc7a6bDeduplicate deprecation notices in the TUI transcript (openai#47065)51d221a9b7Stop redundant message-board reads when limits reach one (openai#47064)fe6455a438Add a partial index for agent message board root posts (openai#47063)d583e73c4dFetch local message board thread summaries in one query (openai#47055)0e69d3bd9bBatch message-board subscriber reads into one SQLite result (openai#47050)f484f7228dShare SQLite pools across local agent message-board handles (openai#47049)40eeb6e8a8Recover corrupt message-board storage during thread deletion (openai#47042)3483294596Preserve delivered messages through post-tool hook failures (openai#47039)87bc50f9d4Cache OS discovery for user agents and telemetry (openai#47038)dd6eff3954Share stored JSON payloads across code mode cells withArc(openai#47036)3ead673946Defer extension tool history materialization until first access (openai#47035)f585f13a12Preserve ordered assistant context in Guardian reviews (openai#47030)73129716c2Delete persisted message boards with their root threads (openai#47029)8a62cc4dc1Align message-board tools with the multi-agent namespace (openai#47028)23a6b705baAdd diagnostics for rollout compression metadata failures (openai#47026)6149914a0eWire agent message boards into persistent multi-agent runtimes (openai#47017)7d99ee82d7Add agent inspection without restoring unloaded runtimes (openai#47010)ebc05da3bdDecouple tool restrictions from session identity withToolPolicy(openai#46999)3938a27c1dRoute agent delivery through captured send requests (openai#46994)f747d23d4bClarify Bazel lockfile verification failures (openai#46993)48f897e4ceExpand agent message board tool integration coverage (openai#46989)695ab0a0b4Add collaboration tools for the agent message board (openai#46985)ffbd30aeb7Canonicalize the workspace trust key in the doctor config test (openai#46981)4d160ea71dIndex channel posts by timestamp for latest-message lookups (openai#46979)8f2c15c398Implement paginated queries for the local agent message board (openai#46978)293de27177Add SQLite storage for local agent message boards (openai#46966)0ad18769adAvoid exposing config values incodex doctorerrors (openai#46962)abbdde95b5Add shared agent message-board extension interfaces (openai#46959)a6fdb11edaPreserve metadata in native user verification requests (openai#46952)774d642582Populate the agents overview from background tool responses (openai#46946)a86631502dAdd independent TUI rendering toggles for Mermaid, math, and tables (openai#46938)57567b8d9eUse shared, keymap-aware tips in the fullscreen composer (openai#46931)05f39d7346Remove theErrorprefix from TUIconfig.tomlload errors (openai#46929)d1e3f9dfe3Fix realtime V3 transcript reconciliation for handoffs (openai#46922)75db67bbc6Enforce current provider requirements for the model catalog (openai#46917)bb054a2b03Keep quota warnings visible in the TUI (openai#46912)ee2149a9b4Preserve transcript position when opening settings pickers (openai#46910)88ef37c649Identify local background servers in/status(openai#46905)eaff9a5f14Hide Back to bottom when the current transcript tail is visible (openai#46902)4ca8e7afffMake transcript list spacing uniform after streaming (openai#46899)d99213289fHonor the effective terminal color level in activity charts (openai#46897)2db0ab7af1Add right-click copying for transcript and composer selections (openai#46895)9fdff739eaEnable plain clicks on transcript links and style bare URLs (openai#46884)2f34d236f5Add/tuito choose the terminal UI mode for the next launch (openai#46883)93b44c3a73Streamline agent command-center shortcuts and layout (openai#46882)282d655defPreserve voice playback across pauses and packet bursts (openai#46880)8b79d8a5f5Count TUI launches by fullscreen transcript configuration (openai#46879)c45ea25ffbAllow subagents to request MCP elicitation input (openai#46877)e29eceb751Preserve streamed answers when subagents finish (openai#46867)b48232f975Enable mouse navigation in the TUI usage view (openai#46866)d6d43270bdImprove usage report layouts and preserve reading positions (openai#46864)46f7d99d0eStabilize usage dashboard navigation and add keyboard help (openai#46863)5e5cadad6bRefresh analytics on identity changes and compact usage menus (openai#46862)d2d196bd6aRemove analytics TUI snapshots (openai#46861)ad538e1c2dLimit the welcome logo animation to onboarding (openai#46859)e39839509aAdd mouse selection and editing to the fullscreen composer (openai#46858)f3f07753caExtract shared text selection helpers for the TUI (openai#46857)2426ed7684Support stadium nodes in Mermaid terminal diagrams (openai#46856)7c1e485c87Increase the default terminal probe timeout from 100 ms to 250 ms (openai#46855)a2de8fedccMove fullscreen transcript control to TUI configuration (openai#46849)79ef00ded1Honor the system clock preference in TUI completion timestamps (openai#46845)15312c326cSelect read-only permissions for temporary structured threads (openai#46844)724cdfbd32Simplify agent command center hints and back navigation (openai#46840)eef77aa211Add status filter tabs to the agent command center (openai#46839)569888ac65Refine the agent command center layout and metadata editing (openai#46838)d4cb75357fImprove agent command center rows and page navigation (openai#46837)b05b3e180bRespect reduced motion in the voice UI (openai#46835)bba48e5ce8Add independent controls for TUI visual effects (openai#46832)5c5308fc9aAdd an animated Codex logo to fresh conversations and onboarding (openai#46752)6b0a1a8325Add a warning footer and dedicated warnings viewer to the TUI (openai#46751)9ed1b74699Preserve startup drafts and submit them when the session is ready (openai#46750)de6ae71ad4Stabilize transcript and composer interactions in the TUI (openai#46749)551844b3efAdd compact transcript browsing and prompt navigation to the TUI (openai#46739)df7f717c85Add transcript search and per-activity detail controls (openai#46734)f1feda2299Integrate the interactive transcript into the alternate-screen TUI (openai#46733)d7d9f2b9b5Add selection and copying to the transcript viewer (openai#46732)6ba6b73657Render dynamic tool activity and preserve TUI history ordering (openai#46731)392f56a611Cancel pending transcript Home jumps on subsequent navigation (openai#46722)a63ae005d4Anchor transcript scrolling to entries and bound viewport rendering (openai#46721)30ab0ee65bCache transcript layouts across measurement and rendering (openai#46720)1ac4b6973bExtract the TUI transcript overlay into its own module (openai#46719)daed4b9755Split shell snapshot credential tests into focused cases (openai#46715)50a90f3408Skip polling delays in the external queue lifecycle test (openai#46714)ed0ee704b4Recover executed tool call metadata under recorder capacity pressure (openai#46712)f43faf132dAlign persisted TUI activity groups and reasoning with live output (openai#46711)4abcb8d1daRestore rich tool details in persisted TUI transcripts (openai#46710)321c50fc2cAdd compact activity rendering and preserve transcript source text (openai#46709)fcaa035438Preserve logical source text and styles across TUI wrapping (openai#46708)5ee2bdf1e0Unify TUI picker styling and improve compact session layouts (openai#46697)c9d13e8c75Unify TUI prompts with shared picker styling and layouts (openai#46695)e28fa3e7a4Unify TUI completion popup styling and preserve result identity (openai#46694)018687bd8cUnify TUI picker styling and improve compact layouts (openai#46692)79d2144ec6Standardize TUI picker presentation and fix wrapping boundaries (openai#46691)132c2be239Improve TUI contrast, keyboard hints, and picker layouts (openai#46680)a5290028a2Extend server version notices to prerelease and local clients (openai#46673)595cc91e8cAvoid fork when spawning macOS filesystem helpers (openai#46661)3801fc8deaMake local child process launch settings explicit (openai#46660)d086e2752dMove local child-process spawning intocodex-utils-pty(openai#46659)