[#573] Separate install wizard from server launch

[realproject7]

Summary

• Split npx quadwork init flow into two distinct phases: install (init) and launch (start)
• cmdInit() now exits after prerequisites + config instead of starting the server — ends with "run npx quadwork start"
• Smart default: npx quadwork with no subcommand runs start for returning users (config exists), init for fresh installs
• Eliminates the AC startup race condition (25-50s install no longer blocks agent registration)

Changes

• bin/quadwork.js: Removed server startup from cmdInit() (−87 lines), replaced with setup-complete banner
• bin/quadwork.js: Added undefined case to command switch for smart default dispatch

Test plan

• npx quadwork init on fresh install runs wizard, installs AC, does NOT start server
• Wizard ends with "Setup complete! Run npx quadwork start" banner
• npx quadwork start launches server + AC + agents normally
• npx quadwork with existing config → runs start behavior
• npx quadwork without config → runs init wizard
• npx quadwork init with existing config → re-runs wizard (force)

Fixes #573

🤖 Generated with Claude Code

[realproject7]

RE2 review — PR #574 (#573)

Checked the diff against the issue's acceptance criteria:

Criterion	Status
npx quadwork on fresh install → wizard only, no server	✅ Server launch + SIGINT handler + browser-open removed from cmdInit
Wizard completes all installation before finishing	✅ AC clone/venv/pip still run in earlier steps; only the launch was removed
Wizard ends with clear "run npx quadwork start" instruction	✅ New banner says "Setup complete!" with npx quadwork start next step
npx quadwork start launches server + AC + agents	✅ cmdStart is unchanged
npx quadwork with existing config → start behavior	✅ New case undefined checks cfg.projects.length > 0 → cmdStart()
npx quadwork init always runs wizard	✅ case "init" unchanged — always calls cmdInit()

readConfig fallback is safe — returns { projects: [] } when no config exists, so the smart default correctly routes fresh installs to cmdInit().

default case change is correct — previously if (command) process.exit(1) let undefined fall through with exit 0 (print help). Now undefined has its own case, so default can unconditionally process.exit(1) for unknown commands.

No issues found. Clean separation, net -55 lines, no behavioral regressions for existing init/start/stop paths.

[project7-interns]

Verdict: REQUEST CHANGES

Summary

This change removes the init-time server launch, but it does not add any alternate path for a fresh install to reach the dashboard and create its first project.

Findings

• [high] Fresh-install onboarding is now blocked after npx quadwork init.
  • File: bin/quadwork.js:1301, bin/quadwork.js:1346, bin/quadwork.js:1563, bin/quadwork.js:2158
  • Suggestion: preserve a path from fresh install into the dashboard setup flow, and make the no-subcommand default key off config existence / installation state rather than projects.length > 0.
  • Details: cmdInit() now writes only the global config + port and exits with “run `npx quadwork start`”. But cmdStart() still treats the zero-project state as the place where the user must create a project from the setup page, and the new smart default routes npx quadwork back into init whenever projects.length === 0. On a fresh install that means the user can no longer reach /setup automatically, and after completing init they are sent to a start path that was not the default selected for that same state.

Decision

Requesting changes because the PR strands first-time users before project creation, which violates the issue acceptance criteria for fresh installs.

[project7-interns]

Verdict: APPROVE

Summary

The follow-up fixes the onboarding regression from the first revision. A true fresh install still runs init, while a post-init zero-project state now routes npx quadwork into start so the operator can reach the dashboard setup flow.

Findings

• None.

Decision

Approving because the split install/start flow now matches the issue intent without stranding first-time users.