Bug: Cursor sub-agent processes shown as separate sessions instead of grouped under parent

[1Ckpwee]

Description

When using Cursor, a single session may spawn multiple sub-agent processes internally (e.g. for parallel file search, code analysis, etc.). CodeIsland detects each sub-agent process as an independent session, resulting in 10+ separate cards for what is logically one Cursor session.

Screenshot

In this example, one Cursor session working on astra-mono spawned several sub-agents. CodeIsland shows each as a separate session card:

• astra-mono #7490 — Cursor, 5h, thinking
• astra-mono #7529 — Cursor, 5h, thinking
• astra-mono #7560 — Cursor, 5h, thinking
• astra-mono #7630 — Cursor, 5h, thinking
• astra-mono #7710 — Cursor, 5h, thinking
• ... (and more)

All of these are from the same Cursor session, but they appear as fully independent sessions in the island.

Expected Behavior

Sub-agent processes from the same parent Cursor session should be grouped together — either:

1. Collapsed under the parent session with a count badge (e.g. astra-mono — Cursor (+5 sub-agents))
2. Hidden entirely, showing only the main session card
3. At minimum, visually nested under the parent to indicate they are child processes, not independent sessions

Why This Matters

• The island becomes unusable when one session generates 10+ cards — you can't find the actual session you're interacting with
• It gives a false impression of workload (looks like 10 active sessions, but it's just one doing parallel work)
• This is different from Feature request: Visual indicator to distinguish subagent sessions from main agent sessions #141 (which addressed visual indicators for Claude Code subagents). The issue here is at the session detection layer — Cursor child processes need to be recognized as belonging to the same parent session rather than being registered as independent sessions

Environment

• macOS
• Cursor (latest)
• CodeIsland (latest)

[wxtsky]

Fixed in 7d475b4 — landing in the next release.

Root cause: bridge's session_id fallback used <source>-ppid-<getppid()>. Cursor IDE spawns N parallel sub-agent subprocesses per logical session (file search, code analysis, etc.), and each sub-agent runs its own hook subprocess with a different immediate ppid → each became a separate session card.

Fix: new CLIProcessResolver.resolvedSessionPID walks the process ancestry and picks the root-most same-source binary instead of the nearest one. All sub-agents spawned by the same root cursor-agent collapse onto a single session_id → one card per Cursor session as expected.

Trade-off worth noting: this picks option 2 from your list (single card, sub-agents hidden) rather than 1 (collapsed with (+5) badge). If you'd find the explicit sub-agent count valuable later, please open a follow-up — that'd reuse the SubagentStart/Stop pattern from #141.