Plan Mode (HITL approval gate)

[bleleve]

Opening this as a discussion thread for the Plan Mode feature stack, a human-in-the-loop approval gate that runs the agent in plan-only mode first, asks for approval, then executes the approved plan. The stack was deployed in production at Fountain since 3 days.

All six PRs target main and are ready for review:

#	PR	Scope
1	#671	feat(control-plane): plan-mode HITL gate (storage + API + sandbox preamble)
2	#672	feat: deployment-wide default model and default plan model
3	#673	feat(web): plan approval UI and Home plan toggle
4	#674	feat(linear-bot): plan-mode label triggers and approve/reject commands
5	#675	feat(github-bot): plan-mode label triggers and approve/reject commands
6	#676	feat(slack-bot): plan-mode (App Home + approval modal + auto-classifier)

main
   └── PR #671  ← foundation (ready)
       └── PR #672 ── default model preferences (ready)
           ├── PR #673 ── web — plan approval UI (draft)
           ├── PR #674 ── linear-bot — label triggers (draft)
           ├── PR #675 ── github-bot — label triggers (draft)
           └── PR #676 ── slack-bot — App Home + approval modal + classifier (draft)

Happy to discuss and to walk through any piece 🙏

CodeRabbit has reviewed the stack, most threads are resolved, two intentional deferrals are documented on #672.