Brainstorming visual companion server exits immediately on Windows (owner PID detection fails)

[smshd]

Description

The brainstorming visual companion server (start-server.sh + server.js) exits almost immediately on Windows with Git Bash. The server starts successfully and outputs server-started JSON, but by the time you try to connect in the browser, it's already shut down with "reason":"owner process exited".

Root Cause

start-server.sh line 94 resolves the owner PID via:

OWNER_PID="$(ps -o ppid= -p "$PPID" 2>/dev/null | tr -d ' ')"

On Linux/macOS, this correctly resolves to the Claude Code harness process, which stays alive for the session. On Windows/Git Bash, the grandparent process is another ephemeral process (mintty or cmd.exe wrapper) that exits shortly after the script returns. The server's lifecycle check in server.js (lines ~311-320) then finds the owner dead and shuts down.

Steps to Reproduce

1. Windows 11, Claude Code CLI, Git Bash
2. Run start-server.sh --project-dir /path/to/project
3. Script returns success with server-started JSON
4. Try to open the URL in browser -> ERR_CONNECTION_REFUSED
5. Check .server.log -> shows {"type":"server-stopped","reason":"owner process exited"}

Environment

• Windows 11
• Superpowers v5.0.2
• Git Bash (MINGW64)
• Node.js 20+

Workaround

Blanking out OWNER_PID before passing to the server works. The server's ownerAlive() function returns true if OWNER_PID is null, and the 30-minute idle timeout still prevents orphaned servers.

Suggested Fix

Either:

1. Skip owner PID monitoring on Windows (detect via OSTYPE or uname)
2. Use a different mechanism on Windows (e.g., a heartbeat file the harness touches periodically)
3. Fall back gracefully when ps -o ppid= returns an unreliable PID

[obra]

This should now be fixed. please reopen if not.