Summary

• slugify extracted markdown link paths the same way import/sync slugify page paths
• let sync-time extraction find files by slugified path instead of assuming slug + '.md' exists on disk
• accept a positional brain dir for gbrain extract in addition to --dir, so gbrain extract links /path/to/brain works as users expect
• add regression tests for uppercase filenames / slugified paths and positional-dir CLI behavior

Why this matters

There were two sharp edges in extraction:

1. extraction used raw repo-relative paths like Folder/INDEX while import stored slugified paths like folder/index, so links/timeline for uppercased or otherwise-normalized filenames silently failed to materialize;
2. gbrain extract links /path/to/brain silently ignored the positional path and scanned . unless the caller knew to pass --dir.

Together those make the graph look much flatter than the corpus actually is.

Test plan

• bun test test/extract.test.ts
• bun test test/extract.test.ts test/slug-validation.test.ts test/sync.test.ts
• bun test