cc-explorer can't inspect or fork a live current-session subagent (shows dispatch_only, masked id)

[coryking]

Problem

Subagents spawned by the current, still-running session are not resolvable through cc-explorer while that session is live. They show up as opaque stubs, so you can't inspect them or fork/attach to them — which is exactly when you'd most want to (e.g. "the agent I just spawned found X; let me go talk to it").

Repro (observed 2026-06-16, session d455a0da, project coryking-plugins)

1. From a session, spawn one or more subagents (here: background Agent spawns, which in this harness become agent-teams teammates).
2. list_session_agents(session=<current session>) → every row comes back with:
   • agent_id: "--------" (masked)
   • source: "dispatch_only"
   • status: "unknown", null tokens/tools/duration
     Only the description and date are real.
3. Because the id is masked, get_agent_detail and convert_session have nothing to target — you can't read what the agent did, and you can't fork it.
4. The session's on-disk …/<session>/subagents/ dir is empty — the live teammates' transcripts aren't there (they live in the agent-teams runtime), so there's no file-path fallback to recover an id either.

Net effect: the convert→attach→interrogate flow (the headline cc-explorer capability) cannot be pointed at an agent the current session just produced. You can only reach agents from past sessions that have been flushed to the corpus.

Why it matters

• Blocks dogfooding the interrogation flow on fresh work ("fork the research agent I just ran and ask it a follow-up").
• The masked-id / dispatch_only rows are also just confusing — they look like a bug (everything null) rather than "not indexed yet."

Breadcrumbs / open questions

• Is this purely a "current-session transcripts aren't flushed/indexed until the session ends" limitation, or could list_session_agents resolve live teammates another way?
• Agent-teams teammate transcripts don't land in <session>/subagents/; where do they live, and can cc-explorer reach them?
• Relptes to convert_session: copying a recently-resumed/possibly-live source can snapshot a stale or torn tail, silently #35 (copying a live/recently-resumed source) but distinct: this is about a live agent being unresolvable at all, not about snapshotting a torn tail.
• Possible smaller win: have list_session_agents label these rows honestly (e.g. status: "live/unindexed") instead of unknown + masked id, so it's clear they're not-yet-readable rather than broken.

Not prescribing a fix — flagging the gap + repro.