docs(interactive): ConPTY interactive sessions spec + plan

This commit is contained in:
mika kuns
2026-07-23 16:47:15 +02:00
parent 17235a6cde
commit d91ad2d635
2 changed files with 206 additions and 0 deletions
@@ -0,0 +1,88 @@
# Plan — ConPTY Interactive Sessions
Spec: `docs/superpowers/specs/2026-07-23-conpty-interactive-sessions-design.md`
Date: 2026-07-23
Execution: subagent-driven-development, sonnet model, TDD where meaningful,
build + test + commit per task. Stage files explicitly by path (never
`git add -A`). Terminal rendering is visual — flagged for the user's visual pass.
## Task 0 — Spike: embed a ConPTY terminal running `claude`
Not TDD; a throwaway proof. Add a temporary window/view that embeds each
candidate control and launches `claude` in a known worktree.
- Evaluate **SvcSystems.UI.Terminal** and **Iciclecreek.Avalonia.Terminal**.
- Acceptance: the real `claude` TUI renders correctly — colors, resize/reflow,
and a live permission prompt is usable; input reaches the CLI.
- Output: pick one library; note the control API (start with cwd/exe/args/env,
process-exited event, dispose/kill). Record the decision in the spec.
- Remove the throwaway harness before Task 1 (or keep as a manual dev sample,
not wired into the app).
**Stop for the user's visual verification of the spike before continuing.**
## Task 1 — Worker: interactive launch-spec endpoint
- Add a Worker service/hub method that, given a taskId, prepares the worktree
(session-skills seeding, agent files, MCP config, env — reuse the autonomous
run prep path) and returns a `LaunchSpec { cwd, exe, args, env }`.
- Reuse `WindowsTerminalLauncher.BuildResumeCommand` for exe/args.
- Guards mirror `ResumeTaskInTerminal` (not Running/Queued, persisted SessionId,
worktree Active/Kept). Never-run task → spec without `--resume` (fresh start).
- Tests (Worker.Tests, real SQLite/git): guard cases, spec contents for a
resumable task, fresh-start case. No real `claude` in tests.
## Task 2 — Worker: ad-hoc launch-spec
- Method to build a `LaunchSpec` for a free session in a given directory:
MCP config + env set up, no task/session-skills seeding.
- Tests: env/MCP presence, arbitrary cwd.
## Task 3 — UI: terminal host control + view model
- Wrap the chosen library in an app control/view (e.g. `InteractiveTerminalView`
+ `InteractiveTerminalViewModel`) that starts from a `LaunchSpec` and exposes
running/exited state.
- `IWorkerClient`: add methods to fetch the task and ad-hoc launch specs; wire
the SignalR client + hub method.
- Update hand-rolled `IWorkerClient`/hub fakes in BOTH test projects.
- Tests: view model starts/stops lifecycle with a fake terminal backend;
fake worker returns a spec.
## Task 4 — Command Center: host interactive panes + entry points
- `MonitorPaneView`: autonomous panes keep the streamed log; interactive panes
host the terminal control.
- Entry points: "Open interactive session" from a task (task-based) and a
"New session" action (ad-hoc, pick directory).
- Layout toggle: focus (tabs) ↔ overview (grid); reuse/extend the existing
`UniformGrid` column logic for the grid mode.
- Tests: view-model level (pane kind selection, layout toggle state). Rendering
is a visual-pass item.
## Task 5 — Remove the streaming interactive stack
Only after Tasks 14 land and the terminal path works.
- Worker: delete `StreamingClaudeSession`, `InteractiveSessionService`,
interactive `WorkerHub` methods + broadcast events, DI registrations.
Verify `LiveSessionRegistry` / `IdleSessionReaper` usage first; remove only if
unreferenced.
- UI: remove composer bits on `TaskMonitorViewModel`, the composer/queued portion
of `SessionTerminalView`, `IWorkerClient` interactive methods.
- Update fakes and delete now-dead tests. Full build + all test projects green.
## Task 6 — Docs
- Update `docs/open.md` with visual-verification items (spike render, terminal
resize/focus, grid vs tabs).
- Update affected per-project `CLAUDE.md` (Worker interactive removal, UI new
terminal host).
## Verification gates
- After Task 0: user visual pass on the spike.
- After Task 4: user visual pass on Command Center (task + ad-hoc, tabs + grid,
permission prompt round-trip).
- Never claim the terminal UI works without the user running it.
@@ -0,0 +1,118 @@
# ConPTY Interactive Sessions — Design
Date: 2026-07-23
Status: Approved (design), implementation not started
## Problem
The current in-app interactive session path streams `stream-json` from a
`claude` process spawned **in the Worker** and renders it as a chat log with a
composer. It does not surface permission requests, AskUser questions, and other
TUI-native interactions well — the rendering is a partial reimplementation of
what the real Claude Code TUI already does. We want full fidelity for
interactive work without rebuilding the TUI.
## Decision
**Hybrid execution model:**
- **Autonomous queue tasks** (`Status=Queued`, picked by the queue): unchanged.
Headless `stream-json`, full orchestration (status flow, diff, review, merge).
- **Interactive sessions**: an **embedded ConPTY terminal** running the real
`claude` CLI, rendered in the **UI process**. Full TUI fidelity (permission
prompts, questions, colors, everything the standalone CLI does). Detached from
the review/merge/status machinery — these are a manual cockpit.
This is the third direction for interactive (external `wt` terminal → streaming
chat → embedded ConPTY). The streaming interactive stack is **removed**, not run
in parallel — accepted as discarded work in exchange for one interactive path
and full fidelity.
## Architecture
### Process location
ConPTY terminal controls render in-process and spawn their child (`claude`) as a
child of the host process. Therefore interactive sessions move **out of the
Worker and into the UI process**. They no longer flow over SignalR. This mirrors
the existing `ResumeTaskInTerminal` behavior (launch real `claude`), but embedded
instead of via external `wt.exe`.
### Worktree preparation (task-based sessions)
Interactive task sessions get the **same worktree preparation as autonomous
runs**: session-skills seeding, agent files, MCP config, environment. The Worker
performs the prep and returns a launch spec to the UI:
```
LaunchSpec {
cwd: string // worktree path
exe: string // resolved claude executable / shell
args: string[] // e.g. --resume <sessionId>
env: Dictionary<string,string>
}
```
The command construction reuses the existing logic in
`WindowsTerminalLauncher.BuildResumeCommand`. Guards mirror
`ResumeTaskInTerminal`: task not Running/Queued, a persisted SessionId exists,
worktree exists and is Active/Kept.
### Free / ad-hoc sessions
In addition to task-based sessions, the user can open an ad-hoc terminal in a
chosen directory (no task). These also get MCP config + env set up so the
`claudedo` tools are available, but no per-level session-skills seeding tied to a
task.
### Terminal host control
Embed a ConPTY-capable Avalonia terminal control in the UI. Candidates:
- **SvcSystems.UI.Terminal** (IvanJosipovic) — explicit ConPTY, full-screen TUI
render + resize reflow, selection/mouse reporting.
- **Iciclecreek.Avalonia.Terminal** — XTerm.NET + Porta.Pty, clean
`TerminalControl` embed API.
Selection is resolved by a **spike** (first plan step): the acceptance criterion
is that the real `claude` TUI renders correctly inside the control — colors,
resize/reflow, and a live permission prompt are usable.
### Command Center layout
- `MonitorPaneView` keeps the streamed log for autonomous tasks.
- Interactive panes host the terminal control instead of the log+composer.
- Layout is **toggleable**: focus mode (tabs, one session large) ↔ overview mode
(grid, several sessions at once).
## Removals
Worker:
- `StreamingClaudeSession`, `InteractiveSessionService`
- `WorkerHub` interactive methods: `OpenInteractiveTerminal`,
`SendInteractiveMessage`, `RemoveQueuedInteractiveMessage`,
`StopInteractiveSession`, `InterruptInteractiveSession`
- Broadcast events: `InteractiveSessionStarted/Ended`, `InteractiveQueueChanged`,
`InteractiveMessageSent`
- `IdleSessionReaper` and `LiveSessionRegistry` **iff** unused elsewhere
(verify during implementation — do not delete blindly).
UI:
- Composer on `TaskMonitorViewModel`: `ComposerDraft`, `SubmitComposerCommand`,
`InterruptInteractiveCommand`, `StopInteractiveCommand`, `QueuedMessages`,
`IsInteractiveLive`.
- The composer + queued-messages portion of `SessionTerminalView` (the log
portion stays for autonomous panes).
- `IWorkerClient` interactive methods.
## Open items (implementation time)
- Whether `LiveSessionRegistry` is referenced outside the interactive path.
- Exact env dictionary handed to the ConPTY control (parity with the autonomous
run environment).
- Session-id availability for a never-run task (no `--resume` → start fresh).
## Non-goals
- No screen-scraping of terminal output back into task status/diff/review.
- No change to the autonomous queue execution path.