[bug] inspect reports live: Dead for detached-run workers that are running #47

Summary

For a detached run, meshloop inspect reports each running worker as "live": "Dead" while the worker process is alive and making progress.

Version / OS

meshloop 0.3.0, origin/main @ 2769f08. Windows 11 native.

Observed

Detached run holoself-sprint-31-34-v2: the meshloop resume --graph … process (spawned by run --detach) → node ml-harness.mjs claude <prompt> → claude -p …, all present in Win32_Process. At the same time, inspect --json showed node 1 state: Running, live: "Dead", pane_id: null. The worker exited normally 4.4 min later, and the node reached AwaitingReview.

Cause

CliHarness::session_live (crates/meshloop-adapters/src/harness.rs:182) returns Live only if the attempt is in this process's in-memory running registry. inspect runs in a different process from the detached resume that owns the child, so the registry is always empty there and every attempt reads Dead.

cancel is not affected: it kills by the persisted attempt.pid (main.rs ~1066). But live is shown to users and agents as if it were authoritative.

Expected

When the caller doesn't own the child, liveness comes from the persisted attempt PID: OS process existence plus a start-time check against PID reuse. It reports Ambiguous or unknown when that can't be determined, never Dead.

Acceptance criteria

• Test: detached run with a long-running fixture worker → inspect from a separate process reports Live.
• Test: worker exited → Dead. PID reused by an unrelated process → not Live.

🤖 Generated with Claude Code