[Validation] Prove autonomy and monitor behavior on macOS, Linux, and Windows

[yanacuti1121]

Context

Local macOS evidence is not sufficient to claim production behavior across Linux and Windows.

Objective

Establish reproducible end-to-end evidence for system health, autonomy execution, supervision, signals, timeout, crash recovery, packaging, and service lifecycle on all supported operating systems.

Required matrix

• Ubuntu latest: native Rust tests, service lifecycle, cgroup/process evidence, crash recovery.
• macOS latest: launchd lifecycle, process groups, CPU/memory/disk/GPU evidence behavior.
• Windows latest: Task Scheduler/service lifecycle, Job Object behavior, paths and atomic replacement.
• Cross-language/CLI smoke where Bash or Python helpers remain part of an approved path.

Acceptance criteria

• Fresh package/install environment can initialize and run the monitor.
• Scheduler starts only after explicit service installation.
• Timeout never enters the critical operation after expiry.
• SIGTERM/stop performs clean shutdown and persists state.
• Forced crash releases kernel locks and allows recovery.
• Stable lock/snapshot files retain their expected inode/file identity where the platform exposes it.
• Existing 10x/10x shared-file race tests pass under the production locking protocol.
• CI logs preserve exact exit codes and platform diagnostics.
• Documentation distinguishes tested PASS from inspected-only portability claims.

Dependencies

• Executor, supervisor, and rollback issues.

Attribution

Roadmap owner: Vũ Văn Tâm. Initial issue structure: OpenAI Codex.