Investigate terminal_create >1s for Claude/system PTY spawn #10
Description
Evidence (server-debug.jsonl)
- 2026-02-03T00:46:55.060Z erminal_create duration 1982.81ms, mode claude, shell system, sessionId 088ebe22-11a4-4a8d-a70e-e65a8d80e564
- 2026-02-03T00:46:54.782Z erminal_create duration 1705.72ms, mode claude, shell system, sessionId d724b38b-33bf-47ba-9e2b-b82f9e6eec0b
- 2026-02-03T00:46:54.236Z erminal_create duration 1688.27ms, mode claude, shell system
- 2026-02-03T00:46:54.062Z erminal_create duration 1521.63ms, mode claude, shell system
Why this matters
1s terminal creation is a visible delay when opening or resuming sessions, especially during session repair flows.
Investigation goals
- Instrument TerminalRegistry.create to separate: session-repair wait, PTY spawn, and initial attach/send.
- Check if sessionRepairService.waitForSession is still impacting erminal_create after caching fixes.
- Validate if
ode-pty spawn on Windows is the bottleneck (e.g., shell startup or cwd switching).
Acceptance criteria
- Identify the slow step and its percent contribution.
- Reduce worst-case erminal_create to <200ms or clearly document why it’s not feasible.
- Add/adjust perf spans so we can see the breakdown in logs.