## What changed
- Add an optional opaque `providerId` to `externalAgentConfig/import` for analytics attribution, independently of the `migrationSource` selector.
- Include the provider ID in completed-import and import-failure analytics events.
- Have TUI imports identify their selected migration source as the provider.
## Testing
- Cover provider attribution in app-server import completion and failure tests and analytics serialization tests.
GitOrigin-RevId: 0ee7313a43652e5305939c526bf6b903a79334ad
## What changed
- Add `maxSessionAgeDays` and `maxSessions` to the external-agent config detection request.
- Apply the requested limits when discovering sessions from supported migration sources.
- Preserve the existing defaults of a 30-day maximum age and 50 sessions when either option is omitted.
GitOrigin-RevId: 46376e94b64f1f61d8776e34d82cdd2cdeb36ca0
## What changed
- Render the first line of a completed hook warning in the hook header using
`says:`, with any remaining lines indented below it.
- Keep the existing hook header when no warning is present and avoid rendering
warning entries a second time in the output body.
- Update hook event snapshots and the multiline warning test for the new layout.
GitOrigin-RevId: 291b556daf3b6fb64a99b090a6df1a222a18a736
## What changed
- Remove the `spawn_agents_on_csv` and `report_agent_job_result` tools and their agent-job runtime and state models.
- Drop the legacy `agent_jobs` and `agent_job_items` tables during state database migration.
- Keep `features.enable_fanout` and `agents.job_max_runtime_seconds` accepted as no-op compatibility settings while omitting them from the generated configuration schema.
## Testing
- Verify upgrades remove both legacy agent-job tables.
- Verify the removed feature and configuration keys still parse without taking effect.
GitOrigin-RevId: 8cc3337da78c67162229f02f40a747f503542646
## What changed
- Add `additionalContextLimit` to command hook configuration for events that can emit `additionalContext`.
- Apply the limit independently to each hook's context before it is sent to the model. Unset values retain the 2,500-token default, while `0` disables spilling for that hook.
- Expose configured limits through app-server hook and requirements responses, and show them in the TUI hooks browser.
## Testing
- Cover JSON and TOML parsing, discovery and hashing, per-hook spilling behavior, app-server responses, and TUI rendering.
GitOrigin-RevId: 55ddc2dbbdf9067dceee3e1968c03617bbf6bcea
## Why
Pending Guardian review state could outlive an interrupted or completed turn,
leaving stale review details visible during a following goal turn.
## What changed
- Clear pending Guardian reviews during both turn completion and finalization.
- Restore an active MCP startup header after cleanup, falling back to `Working`
when no MCP server is starting.
## Testing
Added coverage for interrupted and completed turns, including goal continuation
and overlapping MCP startup status.
GitOrigin-RevId: a15ed81a5d6005e9294089f4424a8dedbc6e9cb5
## Why
The TUI already receives skill metadata from the app-server protocol, but converted it into the core-skills model before using it.
## What changed
- Use `codex_app_server_protocol::SkillMetadata` throughout skill toggles, mentions, and input submission.
- Remove the protocol-to-core conversion and the TUI's `codex-core-skills` dependency.
GitOrigin-RevId: 43d43d74bfb67fc21ef11a0ca0592aa44b89a122
## What changed
Add a borrowed `Line` conversion and use it when rendering and measuring
hyperlink-aware history and pager content. This preserves line styling and
alignment while avoiding clones of the underlying span strings.
GitOrigin-RevId: c9cd6ff5ec63cd29719ea2ae51acede7364f2bd8
## What changed
- Add one-shot composer and status-line transitions when the active reasoning effort changes to Max or Ultra.
- Keep a tier-specific prompt accent after the transition, including a distinct Ultra prompt glyph.
- Preserve drafts and footer priority, avoid replaying effects for restored or unchanged settings, and disable motion when animations or sufficient terminal color support are unavailable.
## Testing
- Add unit and snapshot coverage for animation frames, prompt accents, status-line transitions, narrow layouts, terminal palettes, restored sessions, and reduced-motion behavior.
GitOrigin-RevId: 6da1d501d228d0cf1144b3d211d5d1304a06f447
## Why
Large command streams, including commands that never emit a newline, can make an active TUI cell grow without bound.
## What changed
- Retain live output in full up to 1 MiB, then preserve the first and last 50 completed lines plus the current partial line.
- Bound individual long lines while preserving UTF-8 boundaries, visible head and tail content, and ANSI rendering around omission markers.
- Show omitted line and byte counts in previews and transcripts. Final command output still replaces the bounded live preview when execution completes.
## Testing
Added unit and snapshot coverage for chunk boundaries, split CRLF sequences, newline-free and UTF-8 output, truncated ANSI sequences, retained-output limits, and preview and transcript rendering.
GitOrigin-RevId: 3cf3cc862db686f20ac09b1f6e7f9dc6d264e6eb
## What changed
- Encapsulate aggregated command output behind APIs for total and retained line
counts, reversible preview iteration, and full transcript iteration.
- Build truncated command previews from the retained head and tail lines while
preserving the logical omitted-line count.
- Cover head/tail rendering for long streamed agent and user-shell output.
GitOrigin-RevId: 54b23a19e4639e24c5e4778082edbbc6484ecf80
## Why
During TUI startup, `hooks/list` holds the global config queue while an
uncached `model/list` request may also take time. Waiting for
`configRequirements/read` before submitting the model request prevents these
operations from overlapping.
## What changed
Submit `model/list` and `configRequirements/read` together during bootstrap so
the model fetch can overlap the config requests. Preserve the request-specific
error context and update the session state after both requests succeed.
GitOrigin-RevId: 69a866e6d2fd4798b914dd000456194f49fcfef6
## Why
Sizing, rendering, and cursor placement can query the same chat widget layout
multiple times in one frame, repeatedly measuring active transcript cells.
## What changed
- Build one chat widget renderable tree per frame and reuse it for sizing,
rendering, and cursor placement.
- Cache each flex child's desired height by width for the lifetime of that tree.
- Reuse the bottom pane's renderable directly instead of forwarding each
renderable operation through a wrapper.
## Testing
- Verify flex layouts measure a child once across frame passes and remeasure it
when the width changes.
- Verify a chat widget frame measures its active transcript cell once.
GitOrigin-RevId: 5ad1a6711f4011c699b5d002b13dc3319cb4db8e
## What changed
- Move lifecycle and approval payloads into either the interrupt queue or the
immediate handler instead of cloning them for both branches.
- Borrow MCP elicitation request metadata while constructing the corresponding
form, and consume apply-patch changes after collecting notification paths.
## Testing
- Add a history replay test confirming deferred MCP start and completion events
remain FIFO after streaming finishes.
GitOrigin-RevId: 84ed34f6ff173463a1da016dc140405392c0ba45
## What changed
- Record when committed streaming content contains an inline visualization directive and keep using the canonical full-source renderer afterward.
- Restore directive state during full recomputes and clear it when the streaming renderer is reset.
- Cover directive tracking with and without visualization context, including a raw-to-rich render mode switch.
GitOrigin-RevId: 17786851d6dad6805240b79072fcbd0bb441874a
## What changed
- Remove unused APIs and test helpers across the Rust workspace.
- Route Linux sandbox test setup through the shared exec request path.
- Drop dependencies that became unnecessary after the cleanup.
GitOrigin-RevId: 48fd9b2b1575bad7add0819786b4d8ec88d8486f
## What changed
- Skip loaded-subagent backfills for fresh and forked threads, which cannot have pre-existing descendants.
- Keep the backfill after resuming a thread so agent navigation is populated immediately.
- Reuse thread status collected during backfill when opening the agent picker, avoiding duplicate `thread/read` requests while preserving liveness from live event channels.
## Testing
- Add a WebSocket-backed session lifecycle test that records app-server requests across fork, fresh-session, resume, and agent-picker flows.
GitOrigin-RevId: 604742b3cb495da12c97908d7a0ffa4ebc6a6da3
## Why
The transcript overlay cached every committed cell's height. Cells whose content changes after insertion, such as refreshed status output or newly available visualizations, could outgrow that cached height and be clipped.
## What changed
- Let history cells declare whether their transcript height is stable, and remeasure dynamic cells on later overlay renders.
- Preserve cached heights for stable cells when appending a committed cell instead of rebuilding all renderables.
## Testing
- Cover status output growing after a rate-limit refresh.
- Cover a visualization link replacing its unavailable placeholder.
- Verify that appending a cell preserves cached measurements for stable cells.
GitOrigin-RevId: 03e7166cb94d7353c9b89bff8abb3b7ccb84a6bb
## What changed
- Consume and sort `DiffSummary` entries directly when building renderables.
- Borrow paths and `FileChange` values in the shared row representation used by line-based summaries.
- Share line-count calculation between both rendering paths.
GitOrigin-RevId: b02668074def7529ff39445e9970a1fec209f02b
## Why
Finalized Markdown can be rendered repeatedly at the same width while measuring and displaying transcript history.
## What changed
- Cache the latest rendered lines for finalized agent messages and proposed plans.
- Invalidate cached lines when the width, syntax theme, terminal colors, or color level changes.
- Keep visualization directives uncached because resolving their local files depends on current filesystem state.
## Testing
Add tests for cache reuse, render-state invalidation, raw Markdown rendering, and visualization cache bypass.
GitOrigin-RevId: ba08140041a229b8cb9d297de87dc7bb39ebfe5d
## Why
Raw response items, realtime audio, and other notifications that the TUI does
not consume during replay can carry large payloads. Retaining them in each
thread's replay buffer wastes memory and can evict useful buffered events.
## What changed
- Skip replay buffering for notification types that are handled before routing
or ignored during replay, while still updating turn and approval state.
- Borrow notifications when recording events for an active thread so routing
does not require an extra clone.
## Testing
Add coverage that repeatedly sends large MCP progress and realtime audio
notifications and verifies that the replay buffer, pending approval, and active
turn state remain intact.
GitOrigin-RevId: 5b3414f1db7f3ae1feea4fe6c0275e4a7a1a4174
## Why
App-server command output deltas populate an exec cell before the command
finishes. Treating the presence of output as completion can therefore mark a
streaming command inactive too early and mishandle it when the turn is
interrupted.
## What changed
- Use the recorded duration, which is set when a command ends, as the exec
call's completion marker.
- Preserve streamed output when marking unfinished calls as failed and render
status lines only after completion.
- Keep a single aggregated output representation and render truncated output
without collecting every line first.
- Append grouped exploration calls in place and coalesce adjacent reads while
rendering without cloning the calls.
## Testing
- Add snapshots covering active streamed output, transcript output, and the
resulting history entry after interruption.
- Add coverage for truncating newline-dense output.
GitOrigin-RevId: 420ac1e5fe1b9fcdbdaa5a8998f3e43a5b720254
## Why
Providing an inline visualization context forced every streamed Markdown update
to rerender the full response, even when the source contained no visualization
directives.
## What changed
- Preserve the stable rendered prefix when visualization context is available
but no directive is present.
- Continue using canonical full rendering when the source contains a
visualization directive.
## Testing
Added a regression test that verifies directive-free streams match full
rendering while advancing the stable source boundary.
GitOrigin-RevId: cd70d2ba052db822d474eba8619d04cedf4c0573
## What changed
- Allocate Markdown table widths in bulk while preserving the existing column
shrink priorities and balancing behavior.
- Reuse flattened styled-line data during adaptive wrapping and detect URLs
across span boundaries.
- Remap wrapped terminal hyperlinks with a single forward scan.
## Testing
- Verify bulk table shrinking matches the previous one-cell-at-a-time results,
including a token-heavy column with a million-character width.
- Cover URLs split across styled spans and multiple links on indented Unicode
lines.
GitOrigin-RevId: 24a6fd6fdace6ca948e613d67c130a3a7f8d1a4e
## Why
The MCP image output cell only renders a fixed placeholder, so it does not need
to keep the decoded image alive for the lifetime of the history cell.
## What changed
Decode image content to validate it, then return a marker cell without storing
the resulting `DynamicImage`. The rendered history output remains unchanged.
GitOrigin-RevId: b2c12d08e8ba281d2bb993a8524d338c26d52b96
## What changed
Make the hyperlink-aware history insertion helper accept a slice and borrow
pending history batches when flushing them. This avoids cloning each batch's
entire `Vec<HyperlinkLine>` before insertion.
GitOrigin-RevId: 0bc86c33e2f047c62de60b7642899c4aaa4bb6b8
## Why
The `thread/started` notification for a newly forked side conversation can arrive after the fork response. Selecting the side thread in that window could incorrectly report that it was unavailable.
## What changed
- Seed agent navigation from the side-fork response before selecting the new thread.
- Skip redundant liveness and parent-title reads for side threads that already have local state, while preserving liveness checks for uncached agent threads.
## Testing
- Cover side-thread selection before `thread/started` is delivered.
- Verify uncached threads are still checked and regular forks still resolve their parent title.
GitOrigin-RevId: 1f9fb0586db8094bf5a2624bc4e1de06c9a783a1
## What changed
- Fork side conversations with `exclude_turns` so their initial UI starts empty instead of displaying the parent thread's turns.
- Keep regular fork responses unchanged, including for ephemeral threads.
## Testing
- Add coverage that compares regular and side-conversation forks of the same saved thread.
GitOrigin-RevId: 662607e9f588c20a2e06656deefa5b70bde19cd2
## What changed
- Return committed byte ranges from `MarkdownStreamCollector` so the streaming controller can render and scan the collector's source directly.
- Transfer the complete source out of the collector during finalization, preserving the final unterminated line while removing the controller's duplicate source buffer.
- Cover finalization of an empty collector and update streaming tests to validate the collector-backed source.
GitOrigin-RevId: 0aca912c53080267a2d12ad397e548641171942a
## What changed
Make `thread_to_transcript_cells` consume its `Thread` so transcript rendering can
move owned turn items, message content, IDs, and plan text into history cells
instead of cloning them.
GitOrigin-RevId: 80200ba8928a3ea0e91cbff8a13dd00e21432426
## What changed
- Redraw assistant and plan stream tails only when a completed line changes the visible tail.
- Cache the first reasoning header and update the status indicator only when its visible state changes, while restoring the cached header if the indicator is recreated.
- Keep regular commit ticks able to clear stale stream tails.
## Testing
- Add coverage for unchanged partial deltas, changed newline-delimited tails, reasoning-header redraws, recreated status indicators, stale plan tails, and history-line draw scheduling.
GitOrigin-RevId: 9016175f97ef2db2edc2fd5ba13a4d731dc2d98c
## What changed
Update normal-mode reasoning shortcuts to emit only an
`UpdateReasoningEffort` event instead of reapplying the current model alongside
the new effort.
Extend the shortcut tests to verify that both increasing and decreasing
reasoning effort leave the model unchanged.
GitOrigin-RevId: cf5206a3a8e0a0bf66ecc2d0bfcc1ba44c98fead
## Why
Streaming TUI responses repeatedly rendered the entire accumulated Markdown source for each committed delta, including blocks whose output could no longer change.
## What changed
- Retain rendered output for completed top-level Markdown blocks and rerender only the final mutable block as new source arrives.
- Track block boundaries during the existing Markdown parser pass and map them through table-fence normalization.
- Preserve canonical full rendering when source-wide state may affect earlier output, including reference link definitions and inline visualizations, and when the width or render mode changes.
- Append raw-mode lines directly while preserving blank lines.
## Testing
Added coverage that compares incremental output with full rendering across paragraphs, headings, lists, block quotes, code fences, tables, HTML blocks, links, visualizations, render-mode changes, and normalized table fences.
GitOrigin-RevId: 748171c4ed58923329b59c48ea3c2afa4819ec88
## What changed
- Add `tui.resume_cwd` with `current` and `session` modes for resume and fork flows.
- Let users persist either mode directly from the working-directory prompt, while preserving one-time choices when no preference is configured.
- Apply the preference consistently during startup and in-app resume, honor explicit `--cd` overrides, and reject unsupported or unavailable directory choices with a clear error.
## Testing
- Cover configuration parsing and persistence, prompt behavior, startup and in-app flows, session metadata fallback, and remote workspace constraints.
GitOrigin-RevId: 7bf144c4b8c55bdb868cc8a71caf397c7740451a
## Why
Audio variants were available in the user-input protocol but were replaced with unsupported-input placeholders instead of being sent to the model.
## What changed
- Serialize audio data URLs as `input_audio` content and convert local `wav`, `mp3`, `m4a`, `webm`, and `ogg` files to labeled data URLs.
- Validate and canonicalize base64 audio before requests, enforce the 50 MiB input limit, and replace invalid, unsupported, or oversized audio with explanatory text.
- Preserve audio attachments when mapping response items back to user-message events and document the app-server input variants.
## Testing
- Add unit coverage for local-file conversion, data URL validation, size and format errors, event mapping, and attachment extraction.
- Add client tests that verify data URL and local audio payloads sent to the Responses API.
GitOrigin-RevId: f72cd6b6e43ab12757eeb47621b1d670594ac7d9
## Why
Assistant-authored inline visualizations need a terminal fallback that lets users open generated artifacts in a browser.
## What changed
- Recognize `::codex-inline-vis{file="..."}` directives in assistant Markdown and replace them with browser links during streaming, finalized history rendering, and local transcript previews.
- Materialize visualization fragments in sandboxed viewer documents with a content security policy, while restricting artifacts to HTML files within the thread's visualization directory and limiting their size.
- Show an explicit unavailable message for invalid or missing artifacts, hide incomplete streaming directives, and preserve directive text in code blocks and user Markdown.
- Add a trusted file-link path for generated viewers without enabling arbitrary `file:` links in Markdown.
## Testing
Add coverage for directive rewriting, streaming and replay, artifact validation, sandboxed viewer generation, trusted terminal hyperlinks, and literal preservation.
GitOrigin-RevId: a666b4ebc82bb5798ad45bd137fd815d538e1de2
## What changed
- Add remote and local audio variants to core and app-server user inputs, including generated JSON and TypeScript schemas.
- Preserve audio fields when converting app-server inputs and response content into thread items.
- Keep unsupported audio out of model requests, TUI rendering, text extraction, and truncated tool output, using explicit text placeholders where audio is submitted as a model input.
## Testing
- Cover app-server media conversion, unsupported-input placeholders, TUI omission, and tool-output truncation.
GitOrigin-RevId: 16152bf3af674575ce569764a8b472e032e7569d
## Why
Opening the agent picker stopped after adding status history whenever path-backed subagents were present, so those agents could not be selected from the picker.
## What changed
- Continue into the picker after rendering status for path-backed subagents.
- Skip duplicate liveness refreshes for path-backed threads and display non-primary agents by their canonical path when available.
## Testing
- Add a snapshot test that opens the picker for a path-backed agent and verifies that selecting it emits `SelectAgentThread` for the expected thread.
GitOrigin-RevId: 1668bb7817ce4e13c91d85c4d7ddab31c04c97b3
## Why
Opening the agent picker could clear a successful spawn hint before the new
sub-agent had emitted a turn event, making a running agent appear stopped.
## What changed
- Treat active turns and matching completion events as authoritative liveness
transitions while leaving an empty event store unchanged.
- Remember terminal liveness so delayed activity cannot revive a stopped agent;
only a new turn can mark it running again.
- Ignore `Interacted` activity when deriving agent liveness.
## Testing
Added coverage for empty stores, stale and current completion events, completed
snapshots, delayed activity, and subsequent turns.
GitOrigin-RevId: 462539d8134a5d3520dbd41edb44b3251f50ead5
## Why
Reverse search fetched persistent history one entry at a time, making a deep or unsuccessful search require a lookup for every older entry.
## What changed
- Keep the newest-entry probe lightweight, then scan older persistent history in bounded batches.
- Cache batch results by absolute offset and preserve continuation cursors across repeated searches.
- Ignore stale responses, retry failed batch reads a bounded number of times, and restore the draft without reporting a false “no match” state when history is unavailable.
## Testing
- Cover batch continuation, caching after query changes or cancellation, stale metadata, retry exhaustion, and thread routing.
- Verify that scanning 1,024 entries uses one single-entry lookup and eight batch lookups.
GitOrigin-RevId: efc791fcbd5290050af703dd31c5c21b33240e23
## Why
The Desktop app can use Codex or ChatGPT branding while retaining stable platform identities. CLI discovery and TUI handoff should not depend on a single display name or hardcoded executable path.
## What changed
- On macOS, search for both `ChatGPT.app` and `Codex.app`, and accept only bundles with the `com.openai.codex` identifier.
- On Windows, detect installs by their package app ID and resolve the `codex` protocol executable from the AppX manifest before handing off a TUI session.
- Use “Desktop app” consistently in CLI and TUI user-facing text.
## Testing
Add coverage for selecting a ChatGPT-named Codex bundle and rejecting the classic ChatGPT bundle.
GitOrigin-RevId: b84a56eb6e960712152e1a9b023ab530ec9a354a
## What changed
- Add nullable `mustShowInstallationInterstitial` metadata to app-server
`PluginSummary` responses.
- Preserve the remote service value through catalog, installed-plugin cache,
list, read, and share-list paths.
- Return `null` for local plugins and remote responses that omit the field so
clients can fail closed when the policy is unavailable.
## Testing
- Cover `true`, `false`, and missing values across remote summary paths and
app-server plugin endpoints.
GitOrigin-RevId: db79702ee2bfa800a9910d64d933bcabecb3ce70
## What changed
- Add the `SessionEnd` hook event to hook configuration, protocol schemas, analytics, and the hooks browser.
- Run matching hooks for root threads during shutdown, including app-server archive, delete, idle unload, and graceful shutdown. Flush the transcript first and provide the session ID, transcript path, working directory, and `reason: "other"`.
- Keep teardown bounded: hook output is advisory, the default timeout is one second, configured timeouts are capped at three seconds, and async hooks run synchronously with a warning.
## Testing
- Cover transcript availability, matcher selection, timeout normalization, ignored control output, subagent exclusion, archive/delete ordering, and graceful shutdown of multiple loaded threads.
GitOrigin-RevId: 5d19a658677a137caf836ed5042dcb43f5eb6d1a
## Why
The TUI import request used the selected migration source as the identifier for
the product initiating the import, conflating two distinct protocol fields.
## What changed
Set `source` to `cli` for imports initiated by the TUI while continuing to pass
the selected adapter through `migrationSource`.
GitOrigin-RevId: 7b00728464cb1d4e478ac4aa6b18f7254d2e5152
## What changed
Remove the unused `PairProgramming` and `Execute` footer indicator variants and their label and styling branches. Keep `Plan` as the only `CollaborationModeIndicator` rendered by the TUI.
GitOrigin-RevId: 5f3e6c8ff50a04d41dd5fc342cf1347a7e4f78fe
## What changed
Remove `word_wrap_lines_borrowed` and use the general `word_wrap_lines` API for
borrowed line slices. Drop the duplicate helper-specific tests while retaining
coverage of borrowed iterator wrapping through the general API.
GitOrigin-RevId: aa55a6180705a676bf82ea66c9ba0fe15d00a5fe
## Why
Selecting a usage-limit reset from the picker immediately started redemption,
leaving no opportunity to verify the choice.
## What changed
- Open a separate confirmation popup before redeeming a selected reset, with
`No, go back` selected by default.
- Show backend-provided reset titles and descriptions in the picker and
confirmation, with generic fallback copy when those details are absent.
- Ignore duplicate or stale confirmation and redemption events, while retaining
the idempotency key needed to retry a failed redemption.
## Testing
- Added coverage for confirmation, cancellation, backend copy, duplicate and
stale events, successful redemption, and retry behavior.
GitOrigin-RevId: 828945f99f06e8c73941fd864e4aa9ac52729f98
## Why
Multi-agent V2 spawned threads are controlled by their parent and reject direct
app-server input. The TUI should expose that ownership instead of presenting an
input flow that the server will reject.
## What changed
- Add the experimental `Thread.canAcceptDirectInput` capability for loaded
threads and populate it from the same policy used to validate `turn/start`
and `turn/steer`. Stored, unloaded threads report `null`.
- Carry the capability through TUI thread discovery, selection, startup, and
resume flows, with a compatibility fallback for older app servers.
- Present parent-owned threads as view-only, preserve drafts and queued input,
block agent-directed commands and settings shortcuts, and keep local
navigation and inspection commands available.
## Testing
- Cover V1 writable and V2 view-only spawned threads through selection and
resume.
- Cover composer submission paths, queued and initial input preservation,
allowed navigation commands, settings shortcuts, and the view-only UI.
GitOrigin-RevId: 495ad051249e5577fb0b0c1cc8c40a0998d04918
## What changed
Give command, permissions, patch, and MCP elicitation approvals dedicated
request structs, and update TUI routing and rendering to use the wrapped
payloads.
GitOrigin-RevId: ffee4058dd27505be2a11de3758ac12bb707b6b8
## What changed
- Apply `agents.default_subagent_model` and
`agents.default_subagent_reasoning_effort` when a spawn request does not
provide explicit values, including agent job workers.
- Allow full-history forks to use configured defaults or explicit model and
reasoning-effort overrides while retaining the parent conversation context.
- Preserve the selected model and reasoning effort when an agent role does not
override them.
## Testing
- Cover configured defaults, explicit override precedence, full-history forks,
role application, and agent job workers.
GitOrigin-RevId: b1efa8c9abeee82355088fceef57560433144595
charlesgong-openai