Summary

CLI startup time regressed ~4x on Windows between v2026.2.19 and v2026.2.26.
The delay is in node process startup itself (loading dist/index.js), not in network or gateway connectivity.

Last known good version

2026.2.19

First known bad version

2026.2.26

Regression window and trigger

Started after upgrading from v2026.2.19 to v2026.2.26. No config changes were made.

Steps to reproduce

1. Install openclaw v2026.2.26 on Windows (npm install -g openclaw)
2. Run: Measure-Command { node %APPDATA%\npm\node_modules\openclaw\dist\index.js --version }
3. Observe ~10-14s startup time

Expected behavior

CLI starts in ~3s (as in v2026.2.19 on Ubuntu with same Node v22.22.0)

Actual behavior

node dist/index.js --version takes ~10.8s on Windows v2026.2.26
openclaw gateway status takes ~14s total
Occasional spikes up to 128s observed

Operating system

Windows Server 2022 (NT 10.0.20348.0, x64)

Install method

npm global

Logs, screenshots, and evidence

# Windows v2026.2.26
Measure-Command { node openclaw\dist\index.js --version }
→ TotalSeconds: 10.82

# Ubuntu v2026.2.19 (same Node v22.22.0)
time node openclaw/dist/index.js --version
→ real 0m2.860s

# Plain node baseline (same Windows machine)
Measure-Command { node -e "require('fs'); console.log('ok')" }
→ TotalMilliseconds: 124ms

dist file count: 2571 (v2026.2.26) vs 2138 (v2026.2.19) — +433 files
Process Monitor confirms zero network activity during the delay — pure file I/O

Impact and severity

Affected: All Windows users
Severity: Medium — CLI is usable but every command has 10-15s overhead
Frequency: 100% reproducible on Windows Server 2022

Additional information

Temporary workaround: rolling back to v2026.2.19-2 resolves the issue.
Related issue: #4453 (similar performance regression in Discord MessageListener)
Root cause hypothesis: v2026.2.26 dist contains 433 more files than v2026.2.19.
Windows NTFS + Node module loader has significantly higher per-file overhead than Linux ext4,
causing ~4x slower cold start when loading a large module graph.