docs: lead decision 77, row 41 (S6) may change packages/bus and packages/cli; launcher in the trusted host accepted

Co-Authored-By: Claude Opus 5.5 <[email protected]>

docs/plans/2026-09-26_lead-decisions.md

@@ -1572,3 +1572,26 @@ which stay with him. Each item names who decided it and what happened.
       PM through the broker.
     - Stop: `systemctl --user stop mosaic-bus@mosaic-stack`. Data:
       `~/.mosaic-dev/bus/`.
+
+77. **Row 41 (S6) may change packages/bus and packages/cli (2026-10-09).**
+    Filbert's plan (agents/filbert/work/s6/PLAN.md) puts the session
+    launcher in the trusted bus host, because only the host holds the IPC
+    channel that binds a launch. That reaches past the brief's file list.
+    Sage accepts it.
+    - packages/bus gains IPC ops (identity, authorize, refuse, endLaunch,
+      credentialStatus) and `Broker.endLaunch`. packages/cli gains a host
+      launch socket and `mosaic bus start <business> --pm <harness>:<model>`.
+      No broker socket verb, no event kind, no schema change. If review
+      finds one, it comes back to Sage first.
+    - Darkwing's review on #1523 covers the new IPC ops as a trust
+      boundary, not only the harness and seat packages.
+    - The capability file (0600, in the run directory) is removed once the
+      runner reads it.
+    - The PID namespace per session narrows the decision 62 gap. Its limits
+      go in the package README as limits.
+    - A model in no `launch.max` family is refused. That's the fail-closed
+      reading and it stands.
+    - The live `mosaic-bus@mosaic-stack` unit runs `scripts/mosaic` from
+      this checkout, so it picks up S6 on its next restart. Sage restarts
+      it after S6 lands and records the restart. The recorded PM→coder run
+      stays on a scratch data root.