[QUALITY-569] Stage 2 Client: OrchestrationConfig on Plan Card + Auto-Launch + Disabled Card by cephalonaut · Pull Request #9927 · warpdotdev/warp

cephalonaut · 2026-05-02T18:37:29Z

Description

Adds the client-side orchestration config UI on plan cards, auto-launch for matching run_agents tool calls, and per-conversation config scoping. This builds on top of the Stage 1 confirmation card and shared controls.

Demo: https://www.loom.com/share/18e96159e91641e3b6a78924ea1b9f76

What

Plan card config block — An inline config block on plan cards with:

Approval toggle (pill switch) controlling whether orchestration is enabled
"View details" expand/collapse for Cloud/Local picker, harness, host, environment, and model dropdowns
Vertical full-width layout matching the Figma mock
Dirty-sync transport: config changes piggybacked onto the next outbound request as OrchestrationConfigUpdate

Auto-launch — When the user approves a config and the agent sends a matching run_agents tool call, the confirmation card is skipped and agents spawn immediately. Deferred to stream completion via ActionBlockedOnUserConfirmation to handle streaming timing.

Per-conversation scoping — Orchestration config moved from the AIDocumentModel singleton to per-conversation AIConversation storage (following the todo_lists pattern), preventing config from one conversation leaking into another.

Reactive config hydration — OrchestrationConfigSnapshot messages are processed inline in apply_client_action() as they arrive, rather than scanning all messages on every streaming update. Scanning is only used for conversation restore.

Architecture

crates/ai/src/agent/orchestration_config.rs — Rust-native types, matches_active_config() with 15 unit tests
app/src/ai/blocklist/inline_action/orchestration_controls.rs — Shared controls (trait-based generic pickers) used by both the confirmation card editor and the plan card config block
app/src/ai/document/orchestration_config_block.rs — Plan card config block view
Config block and document view subscribe directly to BlocklistAIHistoryEvent::OrchestrationConfigUpdated (no passthrough relay)

Testing

40 unit tests for run_agents_card_view (21 new covering auto-launch/auto-deny decision logic)
15 unit tests for orchestration_config (matching, proto round-trips)
37 orchestration integration tests passing
Manual validation against the validation checklist covering plan card UI, dirty sync, per-conversation scoping, auto-launch, confirmation card, and edge cases

Server API dependencies

Relies on OrchestrationConfigSnapshot and OrchestrationConfigUpdate proto messages (already deployed)

Agent Mode

Warp Agent Mode - This PR was created via Warp's AI Agent Mode

Conversation | Plan

Co-Authored-By: Oz oz-agent@warp.dev

oz-for-oss · 2026-05-04T17:25:44Z

@cephalonaut

I'm starting a first review of this pull request.

You can view the conversation on Warp.

I completed the review and no human review was requested for this pull request.

Comment /oz-review on this pull request to retrigger a review (up to 3 times on the same pull request).

Powered by Oz

oz-for-oss

Overview

This PR adds client-side orchestration configuration on plan cards, dirty-sync transport for config updates, and auto-launch/deny behavior for matching run_agents actions.

Concerns

The pending orchestration update is stored as a global document-model singleton and is piggybacked onto the next outbound request without checking the request's conversation, so edits in one conversation can be sent with another conversation's request.

Verdict

Found: 0 critical, 1 important, 1 suggestion

Request changes

Comment /oz-review on this pull request to retrigger a review (up to 3 times on the same pull request).

Powered by Oz

oz-for-oss · 2026-05-04T17:47:39Z


-        if let Err(e) = self.send_request_input(
+        // Piggyback any pending orchestration config update.
+        if let Some(dirty_event) = AIDocumentModel::as_ref(ctx)


⚠️ [IMPORTANT] dirty_orchestration_event is a singleton, so the next request in any conversation can carry another conversation's config update; store the source conversation_id in the dirty event and only piggyback it when it matches this request's conversation.

Keyed by conversation ID for now

oz-for-oss · 2026-05-04T17:47:39Z

+        };
+        let model_handle = oc::new_standard_picker_dropdown(&colors, ctx);
+        model_handle.update(ctx, |d, c| d.set_use_overlay_layer(true, c));
+        oc::populate_model_picker(&model_handle, &display_model_id, ctx);


💡 [SUGGESTION] The plan-card model picker is populated once and never refreshed when LLMPreferences finishes loading; mirror the LLMPreferencesEvent::UpdatedAvailableLLMs subscription added to RunAgentsCardView so the dropdown isn't stuck empty or stale.

…ypes - Add OrchestrationConfig, OrchestrationExecutionMode, OrchestrationConfigStatus types in crates/ai/src/agent/orchestration_config.rs with matches_active_config() pure function and proto round-trip conversions (15 unit tests). - Extract shared orchestration controls into app/src/ai/blocklist/inline_action/orchestration_controls.rs: OrchestrationEditState, OrchestrationControlAction trait, generic picker creation/population/sync/render helpers (~720 lines). - Refactor RunAgentsCardView to delegate to shared module (1394→850 lines). All 19 existing card view tests pass unchanged. Co-Authored-By: Oz <oz-agent@warp.dev>

- Add conversation-level orchestration config fields: orchestration_config, orchestration_status, orchestration_plan_id, dirty_orchestration_event - Add DirtyOrchestrationEvent struct for dirty-sync transport - Add OrchestrationConfigUpdated event variant - Add accessors: active_orchestration_config(), orchestration_status(), dirty_orchestration_event(), clear_dirty_orchestration_event() - Add mutators: set_orchestration_config(), set_orchestration_status() - Add hydrate_orchestration_config_from_snapshot() for server snapshot ingestion - Handle new event variant in AIDocumentView and PlanAndTodoListView matches Co-Authored-By: Oz <oz-agent@warp.dev>

- Create OrchestrationConfigBlockView in app/src/ai/document/orchestration_config_block.rs with 'Use orchestration' toggle, Cloud/Local mode selector, and all picker dropdowns (harness, host, environment, base model) using shared controls. - Approved state shows full controls; disapproved state hides them. - Embed config block in AIDocumentView render above the editor content, conditionally shown when AIDocumentModel has an active orchestration config. - Subscribe to OrchestrationConfigUpdated events for re-rendering. - Field edits and toggle changes route through AIDocumentModel.set_orchestration_config() which sets the dirty flag for the dirty-sync transport (Phase 4). Co-Authored-By: Oz <oz-agent@warp.dev>

- Add AIAgentInput::OrchestrationConfigUpdate variant for piggybacking user config edits on outbound requests. - Hook BlocklistAIController.send_query() to collect the dirty event from AIDocumentModel before send_request_input, and clear it after. - Extend convert_input_to_user_input() to convert OrchestrationConfigUpdate to the proto OrchestrationConfigUpdate UserInput wire format. - Handle new variant in all exhaustive match statements across the codebase (redaction, persistence, telemetry, SDK driver, view_impl, etc.). Co-Authored-By: Oz <oz-agent@warp.dev>

Add auto-launch path: when the active orchestration config is approved and matches the request, the card immediately dispatches execute_run_agents() and renders in spawning state from the start (no confirmation UI). Add denied card: at construction, detect existing Denied results in history and render a static disabled card with Cancelled styling. Thread active orchestration config from AIDocumentModel through AIBlock::ensure_run_agents_card_view() to RunAgentsCardView::new(). Co-Authored-By: Oz <oz-agent@warp.dev>

Co-Authored-By: Oz <oz-agent@warp.dev>

The Flex::column() wrapping the editor gave unbounded height to its children. The editor's Scrollable requires a finite height constraint. Fix by making the editor an Expanded child so it fills remaining space after the config block with a bounded constraint. Co-Authored-By: Oz <oz-agent@warp.dev>

- Remove unused imports OrchestrationConfig and ModelHandle in orchestration_config_block.rs - Fix unused closure parameter ctx -> _ in controller.rs - Replace .clone() with copy dereference for UiComponentStyles in orchestration_controls.rs Co-Authored-By: Oz <oz-agent@warp.dev>

AIDocumentModel was missing the subscription to BlocklistAIHistoryModel that triggers scanning for OrchestrationConfigSnapshot messages in the conversation task list. Without this, the server-set config never populated the model and the plan card config block never appeared. - Subscribe to AppendedExchange, UpdatedStreamingExchange, and RestoredConversations events - Scan all tasks in the conversation for the OrchestrationConfigSnapshot message variant - Clone the snapshot to release the history borrow before calling hydrate_orchestration_config_from_snapshot with &mut ctx Co-Authored-By: Oz <oz-agent@warp.dev>

Auto-launch was triggering unintentionally because AIDocumentModel is a singleton that carries orchestration config across conversations. A config set in one conversation would cause run_agents calls in unrelated conversations to auto-launch without showing the confirmation card. Disable auto-launch entirely (hardcode false) until the active config is properly scoped to the owning conversation. The confirmation card flow continues to work normally. Co-Authored-By: Oz <oz-agent@warp.dev>

…rsation AIConversation The orchestration config (OrchestrationConfig, OrchestrationConfigStatus, orchestration_plan_id) was stored as singleton fields on AIDocumentModel, a global singleton shared across all conversations. This caused stale config from one conversation to leak into unrelated conversations, triggering incorrect auto-launches of the RunAgents card. Fix: move these three fields to AIConversation so each conversation owns its own config. Update all readers (OrchestrationConfigBlockView, AIDocumentView, ensure_run_agents_card_view in block.rs) to read from the conversation via BlocklistAIHistoryModel instead of the singleton. The dirty_orchestration_event field stays on AIDocumentModel since it is about outbound sync, not conversation state. Re-enable auto-launch in RunAgentsCardView (previously hardcoded to false as a workaround) with an additional guard against empty configs. Co-Authored-By: Oz <oz-agent@warp.dev>

- Add 1px horizontal divider between description and controls section - Initialize pickers in constructor when config is already approved - Use accent_overlay_2 for active Cloud/Local segment on plan card - Replace text On/Off toggle with pill-shaped switch widget (36x18) Co-Authored-By: Oz <oz-agent@warp.dev>

The model picker dropdown on the RunAgentsCardView confirmation card was not populating because LLMPreferences loads model choices asynchronously from the server. The picker was populated once in ensure_pickers() but never refreshed when new models arrived. Subscribe to LLMPreferencesEvent::UpdatedAvailableLLMs to repopulate the model picker when available models change, matching the pattern used by ModelSelector, AISettingsPage, and other model picker consumers in the codebase. Also removes the unused active_config parameter from RunAgentsCardView::new() (auto-launch is currently disabled) and cleans up the associated dead code at the call site. Co-Authored-By: Oz <oz-agent@warp.dev>

…irty-sync - Move orchestration config from AIDocumentModel singleton to per-conversation AIConversation storage (follows todo_lists pattern) - Plan card UI: pill toggle, View details link, vertical pickers, full-width layout - Confirmation card: LLMPreferences subscription for async model loading - Model picker: use set_selected_by_name instead of set_selected_by_index - Picker dropdowns: remove 190px max width cap, overlay layer for plan card - Dirty-sync: preserve plan_id on user edits, default empty worker_host to warp - Auto-launch: deferred to stream completion via ActionBlockedOnUserConfirmation - matches_active_config: clean up after debug log removal Co-Authored-By: Oz <oz-agent@warp.dev>

- Replace scan-all-messages with reactive OrchestrationConfigSnapshot handling in apply_client_action(), keeping scan only for restore path (now finds last snapshot, not first) - New BlocklistAIHistoryEvent::OrchestrationConfigUpdated for reactive path - Only clear dirty_orchestration_event on successful send (N2) - Add OrchestrationConfigUpdate telemetry variant instead of reusing PassiveSuggestionResult (N1) - Move pill toggle construction inside Hoverable closure (N6) - Simplify matches_active_config match to tail expression (T2) - Remove redundant function-scoped import (T3) - Fix _app parameter name in AIDocumentView::render (T1) Co-Authored-By: Oz <oz-agent@warp.dev>

Extract should_auto_launch() and compute_is_denied() as testable pure functions. Tests cover: happy paths, guard conditions (already launched, denied, spawning, editor open, empty configs), config matching (model, harness, execution mode, field inheritance), and denied-flag computation (history result, disapproved config, combined states). Co-Authored-By: Oz <oz-agent@warp.dev>

- Config block view and document view subscribe directly to BlocklistAIHistoryEvent::OrchestrationConfigUpdated, removing the AIDocumentModel passthrough relay - Remove AIDocumentModelEvent::OrchestrationConfigUpdated variant (no subscribers remain) - Replace FQN paths with top-level imports in agent/mod.rs - Remove process-internal 'Stage 1'/'Stage 2' references from comments Co-Authored-By: Oz <oz-agent@warp.dev>

…g Figma mock Co-Authored-By: Oz <oz-agent@warp.dev>

- Replace orchestration_event_streamer.rs with master's version, add OrchestrationConfigUpdated to ignore list - Add active_config parameter back to RunAgentsCardView::new() and pass it from block.rs - Fix output.rs: use singular open_skill_button_handle/read_from_skill_button_handle (not HashMap) - Add missing orchestration_config imports to run_agents_card_view.rs Co-Authored-By: Oz <oz-agent@warp.dev>

- Refactor dirty_orchestration_event from a single Option to a HashMap<AIConversationId, DirtyOrchestrationEvent> so switching conversations doesn't clobber pending config updates. - Controller takes and clears only the matching conversation's entry; re-inserts on send failure to avoid silent loss. - Remove conversation_id from DirtyOrchestrationEvent (the HashMap key provides the scoping). - Subscribe OrchestrationConfigBlockView to LLMPreferencesEvent::UpdatedAvailableLLMs so the model picker refreshes when LLM preferences load asynchronously. Co-Authored-By: Oz <oz-agent@warp.dev>

cephalonaut · 2026-05-04T20:22:18Z

Advait - this is dependent on https://github.com/warpdotdev/warp-server/pull/10854 which Suraj is still reviewing

cla-bot Bot added the cla-signed label May 2, 2026

cephalonaut marked this pull request as ready for review May 4, 2026 17:25

oz-for-oss Bot reviewed May 4, 2026

View reviewed changes

cephalonaut force-pushed the matthew/orchestration-tool-2 branch 2 times, most recently from 6c7be3a to 4ed836b Compare May 4, 2026 19:20

cephalonaut force-pushed the matthew/orchestration-tool-2-stage-2 branch from 0a50692 to 708a42e Compare May 4, 2026 19:40

cephalonaut force-pushed the matthew/orchestration-tool-2 branch from 4ed836b to ff84142 Compare May 4, 2026 19:55

cephalonaut force-pushed the matthew/orchestration-tool-2-stage-2 branch from 708a42e to 502c6e7 Compare May 4, 2026 19:58

cephalonaut changed the title ~~Stage 2 Client: OrchestrationConfig on Plan Card + Auto-Launch + Disabled Card~~ [QUALITY-569] Stage 2 Client: OrchestrationConfig on Plan Card + Auto-Launch + Disabled Card May 4, 2026

cephalonaut force-pushed the matthew/orchestration-tool-2-stage-2 branch from 6266d02 to 79cbf6b Compare May 4, 2026 20:16

Base automatically changed from matthew/orchestration-tool-2 to master May 4, 2026 20:16

cephalonaut and others added 18 commits May 4, 2026 16:17

cargo fmt

75d8ff8

Co-Authored-By: Oz <oz-agent@warp.dev>

Fix picker colors: neutral_4 border, composited text contrast matchin…

2491524

…g Figma mock Co-Authored-By: Oz <oz-agent@warp.dev>

cephalonaut and others added 2 commits May 4, 2026 16:17

cephalonaut force-pushed the matthew/orchestration-tool-2-stage-2 branch from 79cbf6b to b563051 Compare May 4, 2026 20:18

cargo fmt

67d2d62

cephalonaut requested a review from advait-m May 4, 2026 20:21

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

[QUALITY-569] Stage 2 Client: OrchestrationConfig on Plan Card + Auto-Launch + Disabled Card#9927

[QUALITY-569] Stage 2 Client: OrchestrationConfig on Plan Card + Auto-Launch + Disabled Card#9927
cephalonaut wants to merge 21 commits intomasterfrom
matthew/orchestration-tool-2-stage-2

cephalonaut commented May 2, 2026 •

edited

Loading

Uh oh!

oz-for-oss Bot commented May 4, 2026 •

edited

Loading

Uh oh!

oz-for-oss Bot left a comment

Uh oh!

oz-for-oss Bot May 4, 2026

Uh oh!

cephalonaut May 4, 2026

Uh oh!

oz-for-oss Bot May 4, 2026

Uh oh!

cephalonaut May 4, 2026

Uh oh!

cephalonaut commented May 4, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

Conversation

cephalonaut commented May 2, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Description

What

Architecture

Testing

Server API dependencies

Agent Mode

Uh oh!

oz-for-oss Bot commented May 4, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

oz-for-oss Bot left a comment

Choose a reason for hiding this comment

Overview

Concerns

Verdict

Uh oh!

oz-for-oss Bot May 4, 2026

Choose a reason for hiding this comment

Uh oh!

cephalonaut May 4, 2026

Choose a reason for hiding this comment

Uh oh!

oz-for-oss Bot May 4, 2026

Choose a reason for hiding this comment

Uh oh!

cephalonaut May 4, 2026

Choose a reason for hiding this comment

Uh oh!

cephalonaut commented May 4, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

cephalonaut commented May 2, 2026 •

edited

Loading

oz-for-oss Bot commented May 4, 2026 •

edited

Loading