WIP: desktop UI (Tauri) + interactive chat-driven planning - #38
Draft
Lotus015 wants to merge 13 commits into
Draft
WIP: desktop UI (Tauri) + interactive chat-driven planning#38Lotus015 wants to merge 13 commits into
Lotus015 wants to merge 13 commits into
Conversation
Refs #37. Headless engine + Tauri client (retire TUI), local-first; chat-driven planning before execution (Mozaik participant reusing the Surgeon's DAG-mutation vocabulary); phased F0-F4, first milestone F0+F1. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Refs #37. Architect/planner runners spawn_and_capture stdout; BaroEvent stream only starts in orchestrate after prd.json exists. F1 needs a new long-lived conversational planner participant, not an extension of the capture path. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Refs #37. PlannerSession (Mozaik-flavoured) holds a draft PRD + transcript and mutates it per chat turn, reusing the Surgeon's pure applyReplan (add/remove/rewire) + a retier pass; rejects cycle-inducing mutations via buildDag. scripts/session.ts is ONE long-lived process: PLAN (stdin plan_message -> stdout plan_draft/plan_reply) then run_plan -> orchestrate() over the same event stream. Model call injectable: 5/5 deterministic tests with a mock; verified live end-to-end (sonnet planner split+retier+rewire). Note: scripts/ isn't in tsconfig include, so tsc doesn't check session.ts (caught a '*/'-in-comment bug only via tsx). Follow-up: typecheck scripts. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Closes #39. The StoryLifecycleForwarder hardcoded files_created/modified to 0 (StoryResultData carries no per-story file stats, and the run-total git diff only feeds the 'done' event). Now the forwarder attributes write/edit tool calls to their story via onExternalFunctionCall (the same signal the Sentry uses), tracking distinct paths classified by first-touch tool (Write/write_file -> created; Edit/MultiEdit/NotebookEdit/edit_file -> modified) and reports real counts. Backend-agnostic; no change to the story agents or StoryResultData. Verified e2e: a one-story run creating math.js + math.test.js now emits files_created:2, files_modified:0 (was 0/0). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Refs #37. New desktop/ Tauri app: Rust core (src-tauri/src/lib.rs) spawns the long-lived session.ts, forwards its stdout events as Tauri 'session-event's and writes plan_message/run_plan to its stdin. React 19 + Tailwind v4 + shadcn cockpit (src/App.tsx, src/protocol.ts): start a run, plan by chat with a live read-only DAG (per-story tier badges), RUN gate, then live story status + log. Frontend builds (tsc+vite); src-tauri cargo check passes against Tauri 2.11. Root workspace excludes desktop/src-tauri. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…ory_log feed Refs #37. session.ts now emits plan_status (planning/refining/idle) around the planner calls so the UI shows activity during the otherwise-silent ~10-20s planner turn. The desktop cockpit gains: a live activity strip ('Planning with sonnet…' / 'Running S4, S5' with a spinner), a typewriter reveal for planner messages, a 'thinking…' indicator, pulsing active DAG nodes showing each running story's latest log line ('where'), and a live story_log feed (the agents' real per-line stream, tool calls included) — previously dropped by the UI. Frontend builds (tsc+vite). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…t composer Refs #37. RUN now lives top-right of the DAG panel (planning phase only), where the plan it gates lives; the chat composer gets an always-present Send icon button (lucide) in RUN's old spot. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…unning Refs #37. Nodes were min-w only, so the card grew/shrank to fit the longest content — and the streaming 'last line' changes constantly, making cards resize. Now: fixed w-56, title line-clamp-2, and a reserved fixed-height truncated activity row so width and height stay stable as the run streams. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
… / effort Refs #37. The start screen was hardcoded to sonnet/no_git:false, so the 0.48.0 per-story tiering + endpoints were invisible in the app. Adds a 'Run options' section (shadcn select/switch/label): planner model, execution backend, effort, --tier-map, --openai-endpoint list (name=url, keys still from env), and a no-git switch — all passed through start_session (Rust already accepted them). Now a mixed-fleet run (e.g. MiniMax + Claude) is configurable from the UI. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Refs #37. Closes the 'message a running story' gap. orchestrate() gains an onOperatorReady hook so a caller can grab the Operator before stories run; session.ts no longer blocks/closes stdin on run_plan — it keeps reading and dispatches redirect/abort/shutdown to the live Operator (redirect → AgentTargetedMessage → the story agent's inbox). In the cockpit, during execution you click a running DAG node to target it (ring), and the composer sends a redirect there (placeholder shows the target); planning still sends plan_message. Orchestrator typecheck + desktop build clean. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…, brand accent Refs #37. Decomposes the monolithic App.tsx into a proper structure: - hooks/useSession.ts owns all runtime state + the Tauri event bridge - components/: TitleBar, StatusBar, StartView, RunOptions, ChatPanel, DagPanel, StoryNode, DoneSummary, Typed, Spinner - App.tsx is now a thin shell composing them IDE polish: rebranded to BDE (Baro Development Environment) — product name + window title; VS Code-style bottom StatusBar (activity · backend · planner · tokens · stories · elapsed); resizable chat|DAG panels (react-resizable-panels); a done-summary banner (result · counts · tokens · PR · New run). The baro brand orange (#ffb547, from baro-website) is wired as a Tailwind `baro` token and used subtly — brand dot, RUN button, running-node ring, spinner edge, PR links. Frontend builds clean (tsc + vite). Adds shadcn select/switch/label/resizable/tooltip. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Refs #37. Addresses run feedback: - The planner now speaks on the INITIAL plan, not just on refinements: seed asks the model for a conversational 'reply' (what it built, what's parallel, why those tiers, a nudge to refine or RUN), emitted as plan_reply. Previously the left pane was empty until you typed. - Status-bar timer only ticks while something is actually happening (planner thinking or stories executing); it freezes once the plan is drafted and waiting, and the clock restarts at RUN so it measures the run. (Was ticking during idle planning.) - Chat polished into proper conversation bubbles (planner left w/ baro dot + typewriter, you right); 'Output' header on the feed pane. session.ts transforms; planner tests 5/5; desktop builds clean. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Tracks #37.
Working branch for the new direction: retire the ratatui TUI, keep a headless engine, build a Tauri desktop UI, with interactive chat-driven planning before execution as the centerpiece.
This PR starts as a design spec (
docs/desktop-ui-and-interactive-planning.md) to elaborate on, then carries the F0 + F1 work:barostreams the full event stream (incl. planning) + local bridge.Draft until F0+F1 are testable. See #37 for the full rationale and phased plan.
🤖 Generated with Claude Code