add TUI orchestration permission and configuration card

[harryalbert]

Description

Adds the TUI permission and configuration card for a blocked RunAgents request. When the request reaches the front of the confirmation queue, the card becomes the active interaction: it replaces the normal input/footer and warping row without mutating the input draft, cursor, selection, mode, or scroll state. Accept dispatches the edited request through the existing shared orchestration executor; Reject follows the existing action-cancellation path.

The acceptance card shows the proposed agents with deterministic theme-derived color/glyph identities and the resolved run-wide configuration. Ctrl+E opens a sequence of single-field configuration pages built on the reusable TuiOptionSelector from the PR below. Cloud runs expose Location, Harness, optional API key, Host, Environment, and Model; Local runs force the Warp harness and collapse the flow to Location and Model. Enter, Tab, arrows, Escape, Ctrl+C, numeric shortcuts, search, and mouse input all follow the TUI-specific interaction model while selections/defaults/validation come from the shared orchestration state and option snapshots.

The card re-resolves approved plan configuration and live harness/model/secret/worker catalogs, validates before launch, preserves the per-call computer-use setting, and falls back to the existing non-interactive tool-call presentation for streaming, spawning, restored, and terminal states.

This PR is intentionally mostly leaf-node TUI UI plus focus/layout event threading. It does not add a new orchestration executor or duplicate the orchestration domain model; it consumes the shared behavior and reusable editor/selector primitives established downstack.

Linked Issue

CODE-1822 — Orchestration

Visual verification

The end-to-end behavior and visual treatment are best reviewed in the walkthrough:

Watch the TUI orchestration card walkthrough on Loom

Because this change is mostly UI composition and event threading, please use the video as the primary verification artifact, then use the code paths below to review the state transitions behind it.

Review guide

Start here

• crates/warp_tui/src/orchestration_block/render.rs — acceptance/configuration presentation, metadata, identities, validation copy, and terminal-state fallback.
• crates/warp_tui/src/orchestration_block.rs — card modes, keybindings, dynamic page navigation, catalog subscriptions, focus ownership, accept/reject lifecycle, and request re-syncing.
• crates/warp_tui/src/orchestration_block/configuration.rs — the small adapter from each page to shared option snapshots, edit-state transitions, validation, and dispatch.
• crates/warp_tui/src/orchestration_block_tests.rs — focused coverage for page sequencing, plan inheritance, request reconstruction, navigation/focus, and exactly-once dispatch.

Integration threading

• crates/warp_tui/src/agent_block.rs — materializes a card for streamed RunAgents actions, routes Reject, and derives the front-of-queue blocking child.
• crates/warp_tui/src/terminal_session_view.rs and crates/warp_tui/src/transcript_view.rs — thread blocker changes upward, transfer focus, and derive input/footer visibility without storing a suppression flag or modifying input state.
• crates/warp_tui/src/orchestrated_agent_identity_styling.rs and crates/warp_tui/src/tui_builder.rs — deterministic agent identities and theme-derived orchestration styles.

Skim / supporting plumbing

• app/src/tui_export.rs, app/src/ai/blocklist/{mod.rs,action_model.rs}, and app/src/settings/init.rs expose existing shared models/executors or test seams to the TUI; they do not introduce new orchestration behavior.
• crates/warp_tui/src/{lib.rs,keybindings.rs,test_fixtures.rs} wires modules, bindings, and test construction.
• The one-line changes in existing test files are exhaustiveness/registration updates caused by the new view/event variants.
• specs/CODE-1822/TECH.md records the detailed implementation mapping and is useful as reference, but the runtime behavior is concentrated in the files above.

Testing

• Manually tested locally with ./script/run-tui
• Visually verified the end-to-end acceptance and configuration flow; see the Loom walkthrough above
• Added focused unit coverage for page sequencing, inherited configuration, request rebuilding, selector navigation, focus restoration, exactly-once decisions, keybinding registration, and deterministic agent identities

Screenshots / Videos

https://www.loom.com/share/552c213960c54c5c84fad85a726dd08e

Agent Mode

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

CHANGELOG-NONE

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

[harryalbert]

Warning

This pull request is not mergeable via GitHub because a downstack PR is open. Once all requirements are satisfied, merge this PR as a stack on Graphite.
Learn more

• add product and tech spec #13855
• Add TUI orchestration tab bar specs #13832
• add reusable TUI tab bar component #13831
• rich message component #13777
• Add local child agents to TUI #13776
• add TUI multi-session registry and focused-session root projection #13775
• add TUI orchestration permission and configuration card #13717 👈 (View in Graphite)
• add reusable TUI option selector over shared option snapshots #13716 : 1 other dependent PR (#13830 )
• add reusable TUI editor view #13817
• route TUI editor input by owning-view focus #13771
• add option snapshots and adapt GUI orchestration pickers onto them #13714
• extract orchestration edit state into a frontend-neutral module #13707
• master

This stack of pull requests is managed by Graphite. Learn more about stacking.

[moirahuang]

1. we can see how erica feels about this but if we only have 1 menu item i think we can auto select it

2. i had previously selected haiku. when i cancelled and spun up the orchestration cards again, the subagents were still called haiku. can we double check how we're saving the subagent names? i know there's some autogeneration involved so i'm not entirely sure if this is just the agent's choice

3. we should sort by value of the number, not just per character so 10 is after 6

4. nit: can we scroll if the input is too long instead of jumping to the start? and can we do this in a more general way, not just for the custom input

Screen Recording 2026-07-16 at 3.37.41 PM.mov (uploaded via Graphite)

5. can we keep the icon on the same row as the subagent name?

6. maybe it's just me but i didn't realize i should just type to search. i wonder if there's a visual indicator we can add here somehow (also maybe an erica question)

[moirahuang]

why is remote_computer_use_enabled stored outside of execution_mode but other fields are not?