Environment

• OS: WSL2 (Windows 11), Ubuntu
• hermes-agent: v0.14.0 (latest main branch)
• SQLite: 3.37.2
• Filesystem: ext4 on WSL2 virtual disk (VHDX)

Problem

kanban.db gets corrupted when multiple gateway processes run simultaneously.

Error messages:

• sqlite3.DatabaseError: database disk image is malformed
• sqlite3.OperationalError: disk I/O error
• ERROR kanban dispatcher: board default database is not a valid SQLite database

Root Cause

1. Multiple gateway processes (12+) open the same ~/.hermes/kanban.db file simultaneously
2. Each process holds 2-6 file descriptors to the database file
3. WAL mode's -shm (shared memory) file has synchronization issues on WSL2's 9p filesystem
4. Even with flock-based serialization in write_txn(), corruption still occurs
5. Corruption happens at the filesystem level, not SQLite protocol level

Attempted Fixes

1. Added flock (LOCK_EX/LOCK_UN) in write_txn() — works in isolated tests but corruption still in production
2. Added PRAGMA busy_timeout=5000 — reduces but doesn't eliminate corruption
3. Tried tmpfs for kanban.db — works but data lost on WSL shutdown
4. Current workaround: watchdog script auto-rebuilds database on corruption

Reproduction

1. Start 12+ gateway processes (multiple profiles)
2. Dispatch 5+ concurrent kanban tasks
3. Within minutes, database corruption occurs

Request

Is there a recommended configuration or fix for running multiple gateway processes with a shared kanban.db on WSL2?

Related: microsoft/WSL#2395 (sqlite write locks aren't respected in WSL)