fix: skip stale check for ad-hoc instances

[bloodcarter]

Summary

• Ad-hoc instances created via hcom start --as <name> get falsely marked as stale and cleaned up after ~10 seconds
• Root cause: HEARTBEAT_THRESHOLD_NO_TCP (10s) applies to ad-hoc instances which have no heartbeat mechanism (no PTY thread, no TCP endpoint)
• Fix: skip heartbeat-based stale check for tool="adhoc" instances, same as remote instances

The problem

When a script creates an ad-hoc identity for event polling:

hcom start --as controller
# ... poll events with hcom events --from agent --last 1 ...

The identity is killed by stale cleanup within seconds because:

1. Ad-hoc has tcp_mode=0 and no notify endpoint → has_tcp=false
2. Threshold becomes HEARTBEAT_THRESHOLD_NO_TCP (10 seconds)
3. No PTY delivery thread sends heartbeats for ad-hoc
4. After 10s with no heartbeat, status becomes inactive:stale:listening
5. Cleanup removes it

The fix

3-line change in get_instance_status(): check data.tool == "adhoc" and skip the heartbeat timeout, same as is_remote instances.

Test plan

• hcom start --as test && sleep 200 && hcom list test — survives 200s (was dying at 10s)
• All 1243 existing tests pass
• Normal PTY agents still get stale-checked correctly

🤖 Generated with Claude Code

[bloodcarter]

Closing — the root cause is deeper than the stale check. Ad-hoc identities created from within Claude Code tool calls are killed by the hooks lifecycle, not the stale checker. The instances.rs fix only helps for pure adhoc (outside Claude), which is a rarer case. Will investigate the hooks lifecycle separately.