What problem does it solve? Cross-agent collaboration requires watching an inbox for new messages, but each host (Codex, Claude, Antigravity) has a different wake model, and hand-rolled polling loops, cron jobs, or duplicate monitors cause wasted model turns and unsafe lifecycle states. This Skill arms exactly one durable, session-scoped inbox monitor using the correct host-native mechanism. ## Core Features & Use Cases - Host-native lifecycle selection: Uses Codex long-running exec sessions, Claude's persistent Monitor tool, or Antigravity's asynchronous run_command instead of a universal polling loop. - Strict session and runtime validation: Resolves strong session identifiers per host, verifies the canonical installed monitor runtime with ownership and symlink checks, and refuses placeholder or cross-session IDs. - Typed result contract: Returns precise outcomes such as armed, already_armed, degraded_no_event_wake, startup_failed, sandbox_blocked, or stopped instead of generic success. - Legacy Codex goal migration: Fail-closed transcript analysis and decision tables for safely detaching or stopping pre-4.5.3 goal-based monitors. - Use Case: A user says "start the inbox monitor" in a Claude session; the Skill validates the session ID, launches the canonical inbox-polling-monitor via the Monitor tool, verifies the five-line startup output, and reports armed with the retained task ID. ## Quick Start Ask the agent to start the inbox monitor for the current session so new cross-agent messages trigger a notification.
