Code Map

/quack:index maps the files git tracks (or the folder, outside git) into documentation/INDEX.md:

## src/app/api/download-notify/
- `route.ts` (24)
  POST L6-23

## src/emails/
- `notify.ts` (90) — Brand colors, taken from src/app/globals.css …
  Notification L14-21 · formatWhen L22-28 · html L38-64 · sendNotification L71-89

Folder, then file with its line count and purpose (the file's first comment or heading), then its symbols with line ranges. Claude reads the map, picks the file and the range, and reads only that.

Big repos

Repos with more than 250 files get one sub-index per folder in documentation/index/<folder>.md. INDEX.md then lists the folders, so nobody loads the whole map at once.

Staying fresh

Once the map exists, Quack refreshes it at the start of every session. The output is deterministic and only changed pages are rewritten, so the diff is small. Commit documentation/INDEX.md (and documentation/index/) so the whole team shares it.

Where the idea comes from

The map follows PageIndex: reason over a tree first, then read only the branch you chose. No embeddings, no vector database.