Windows desktop app takes ~1-2 minutes to become usable while startup launches multiple CLI processes

This was generated by AI during triage.

Environment

• Windows 11
• GitHub Copilot desktop app v1.0.25 (updater reported latest)
• App-spawned Copilot CLI v1.0.71
• Existing sessions and multiple projects

Actual behavior

After starting the desktop app, the reporter observes approximately 1-2 minutes before existing sessions become usable.

One captured startup/resume sequence took 91.9 seconds:

Segment	Duration
App start -> resume request	74.577s
Resume request -> completion	17.402s

The entire 91.9 seconds is not resume time, and logs do not prove the pre-request interval was entirely blocking.

Expected behavior

The app should become usable promptly after startup.

Observations

• Startup booted eight CLI processes: seven for catalog warming and one for the requested resume.
• Five CLI boot intervals overlapped at peak.
• No CLI spawn failures or retries occurred in this startup window.
• Across 22 samples from the full run, spawn_ms was 6,598-21,343ms (median 12,196ms).
• Seven catalog creations reported create_session_rpc_ms of 11,507-27,413ms.
• Cold session_create_ms samples were 16,005-31,896ms; warm-process samples were 100-343ms.

These are separate metrics measured on this host.

Sanitized log shapes

github_app::session::manager::cli_pool: spawned new CLI process for pool ... pid=Some(<redacted>) spawn_ms=14516

github_app::session::core: CLI session created ... create_session_rpc_ms=<duration>

github_app::session::manager::lifecycle: session::create completed ... retain_cli_ms=<duration> model_fetch_ms=<duration> session_create_ms=26391 total_ms=<duration>

Unconfirmed factors

The profile contained 1,145 sessions and 1.37GB of session state. A clean-profile control was not performed because the investigation was read-only.

MCP configuration, Defender, orphaned processes, authentication warnings, and an invalid hook remain unproven as causes.

Related issues

• https://github.com/github/copilot-cli/issues/4165 - Windows cold resume hangs on CLI v1.0.72-0; this report concerns desktop startup.
• https://github.com/github/copilot-cli/issues/3652 - WSL startup spends 40-80 seconds in listSessions.
• https://github.com/github/copilot-cli/issues/3429 - Linux v1.0.49 cold boot regression caused by credential lookup.

Questions

1. Is catalog warming expected to create one CLI process per project or directory?
2. Should these CLI processes be reused or pooled?
3. Which diagnostics expose the internal phases of create_session_rpc_ms?
4. Is a clean-profile reproduction required?

Additional context

This is a read-only investigation from one host. It establishes timing and correlation, not the internal root cause.

No source files, tests, or entry points are identified. Start with the captured startup/resume logs and the spawn_ms, create_session_rpc_ms, and session_create_ms measurements; determine whether catalog warming and multiple CLI processes explain the delay, and establish a reproducible completion criterion.