Problem

The memory index dirty flag was showing a false positive after a clean sync. When running clawdbot memory status, it would always report Dirty: yes even after a successful clawdbot memory sync, causing confusion about whether the index actually needed reindexing.

Root Cause

Each clawdbot memory command creates a fresh MemoryIndexManager instance. The issue occurred because:

1. Constructor unconditionally set this.dirty = this.sources.has("memory") (line 248)
2. sync() would set this.dirty = false after successful indexing
3. close() removes the manager from cache
4. Next command creates a new instance -> dirty resets to true again

The actual indexing logic was working correctly (using file hashes to skip unchanged files), but the dirty flag was purely a UX issue showing incorrect state.

Solution

Persist the dirty flag to the database metadata so it survives across manager instances:

1. Extended MemoryIndexMeta type to include optional dirty?: boolean field
2. Modified constructor to restore dirty state from metadata: this.dirty = meta?.dirty ?? this.sources.has("memory")
3. Added persistDirtyState() helper that updates the dirty flag in metadata
4. Persist after sync completes (when setting dirty = false)
5. Persist when watcher detects changes (when setting dirty = true)

Testing

Added comprehensive test suite in manager-dirty-flag.test.ts that verifies:

• After sync, dirty is set to false and persisted
• New manager instances restore the dirty state from metadata
• The flag doesn't incorrectly reset to true on each new instance

Impact

• User-facing: Status now accurately reflects whether the index needs reindexing
• Backward compatible: Existing indexes without the dirty field default to true (safe fallback)
• No data changes: Only affects the status reporting, not the actual indexing logic