Code indexing

Load tracked source files from a local Git checkout into gitcrawl's separate code-document index.

#Index a checkout

Mirror the GitHub repository first, then point gitcrawl at a checkout:

gitcrawl sync owner/repo
gitcrawl code index owner/repo --path /path/to/checkout

The repository identity comes from owner/repo; --path supplies the local Git worktree. The command records the current HEAD, whether tracked files are dirty, file counts, indexed bytes, and skip counts.

Defaults:

Only stage-0 regular blobs returned by git ls-files --stage are considered. Symlinks, submodules, paths that resolve outside the checkout, other non-regular files, tracked paths absent from the worktree, NUL-containing files, invalid UTF-8 files, and oversized files are skipped. Untracked files are ignored.

Each successful index replaces the previous code snapshot for that repository, so deleted and renamed paths do not linger in search.

Code indexing requires a normal local database. Portable-store runtime mirrors are refreshed from their published source and therefore reject code index; the source corpus is never added to the portable payload.

#Search source

gitcrawl search owner/repo --query "RefreshManifest" --scope code
gitcrawl search owner/repo --query "manifest cache" --scope all

Code search uses local SQLite FTS. --scope all interleaves thread results and code results within the requested limit.

#Boundaries

Code documents are deliberately separate from GitHub thread documents. They do not enter OpenAI embedding requests, semantic neighbors, or durable duplicate clusters. Portable-store and cloud publish flows also remain focused on GitHub archive data.