[Upstream PR #209] Add interactive askUserQuestion and configurable swarm spawn mode

[quangdang46]

Mirrored from upstream 1jehuang/jcode — Pull Request #209 by @Zephyr709
Original state: open
Created: 2026-05-14T00:15:05Z · Updated: 2026-05-15T03:48:44Z
Diff: https://github.com/1jehuang/jcode/pull/209.diff
This issue is an auto-mirrored copy. Comments and edits here are local to quangdang46/jcode — do not expect them to propagate upstream.

---

Summary

This branch collects the current agent/tooling UX work:

• adds askUserQuestion tool support and an interactive TUI modal
• adds ToolSearch deferred-tool registry plumbing
• improves provider/model allowlist handling and local startup recovery
• adds configurable swarm spawn behavior via spawn_mode = visible | headless | auto
• documents background/non-focus-stealing agent orchestration guidance
• activates the macOS desktop window on launch
• includes rustfmt cleanup for touched provider/TUI files

Notable new swarm behavior

swarm spawn and spawn-backed assignment actions can now request:

• visible: launch a terminal window
• headless: create an in-process/headless worker, avoiding Ghostty focus stealing
• auto: preserve historical visible-then-headless fallback behavior

Validation

• cargo fmt --check
• git diff --check
• cargo check -p jcode-protocol
• cargo check -p jcode
• cargo check -p jcode-desktop
• cargo test -p jcode-protocol test_comm_spawn_roundtrip_with_optional_nonce -- --nocapture
• cargo test -p jcode-protocol test_comm_assign_next_roundtrip -- --nocapture
• cargo test -p jcode --lib assign_next_prefers_worker_with_dependency_context -- --nocapture
• cargo test -p jcode --lib assign_next_prefers_worker_with_matching_subsystem_metadata -- --nocapture

Follow-ups

• Add a config-level default for swarm spawn mode so users can prefer headless globally.
• Add a Jcode ingestion bridge for the local agent telemetry warehouse.

[quangdang46]

Mirrored upstream PR #209: askUserQuestion + configurable swarm spawn mode. 89 files / 10001 lines — too large for a port; needs to be designed as separate local features. Closing as not planned for now.

[quangdang46]

Upstream PR 1jehuang#209 adds askUserQuestion + configurable swarm spawn mode in 89 files / 10001 lines. The change cuts across the swarm protocol, agent prompting, message types, and TUI input — deferring a port that large risks regressing the recently-stabilized swarm/communicate stack here without a dedicated regression matrix.

Tracked locally — will be picked up as a sequence of smaller deliberate PRs (interactive ask-user-question first, then swarm spawn-mode config). Closing this mirror as not planned for the bulk port. Reopen if you want me to start on the askUserQuestion piece in isolation.

[quangdang46]

Implementation Plan

Scope (mega-issue: 89 files in upstream PR — break into 3 PRs)

Sub-PR A: askUserQuestion tool

A new built-in tool that pauses the agent and surfaces a question to the user, awaiting their typed response. Answer is fed back to the model as a tool result.

• New src/tool/ask_user.rs.
• TUI: when tool is called, show prompt + capture line input.
• Headless mode: read from stdin or fail clean with a hint.

Sub-PR B: configurable swarm spawn mode

• [swarm] config: mode = "auto" | "manual", max_workers, default_role.
• Slash command /spawn <role> to manually create a worker.
• Currently swarm spawning is implicit via the swarm tool — add explicit user control.

Sub-PR C: askUserQuestion for swarm coordination

Workers can call askUserQuestion to ask the coordinator (other agent) for direction. Routed via the existing inter-agent messaging.

Estimated scope: XL total, 3 PRs each M/L. PR-A is the smallest standalone win — ship first.

[quangdang46]

Deferred — interactive askUserQuestion + configurable swarm spawn are 89-file changes. Will be revisited when there's a specific user blocking on the swarm UX. The slash command /q and ambient mode cover some of this surface today.