feat(index): resolve each session's source repo at index time

[dacorvo]

What

Resolve the source repository a session belongs to (owner/name of its checkout's git remotes) once at index time and store it in a new repo column, rather than resolving it live when a reader needs it.

Why

This prepares the introduction of curated memories associated to git repos.

The curation process will attribute a session to a publication by its repo. Doing that resolution at read time — reading the transcript's cwd and running git against it — is fragile: it only works if the checkout still exists on this machine, at that path, right now. A deleted or moved checkout, a removed worktree, a temp dir, or a session synced from another machine all come back unattributable. It also re-pokes git on every read.

Resolving at index time captures the attribution when the checkout is present and fresh, and stores it — so it survives the checkout later being deleted or moved, and a reader just reads a column.

How

• src/repo.rs — the resolver: the transcript's recorded cwd (top-level cwd for Claude/pi, payload.cwd for Codex) → git -C <cwd> remote → each remote URL normalized to owner/name, space-joined (any remote, so a fork's upstream is included). Empty when unresolvable.
• repo column — appended last, the add_columns slot harness already uses (the column-order tripwire test pins it). The indexer resolves each unit once, cached per cwd, and stamps its chunks.
• Backfill (disposable [[example]]s, kept until every store carries repo) — additive via Lance add_columns: no data rewrite, no re-embed, no reindex.
  • backfill_repo — local ~/.funes/store.
  • backfill_remote_repo — a Hub store, resolving from the local transcripts (a personal mirror mirrors this machine); hf_dataset::add_column lands the new column's files in one head-guarded commit via the capture store, the same pattern as append/reindex/rename_column.
  • Both skip a store that already has repo and verify the column at the end.

Notes

• A store must be backfilled (or freshly indexed) before the new binary appends to it — the schema now has repo. Disable session-end auto-indexing, backfill, then resume.
• Follow-up (separate branch): curate's discovery switches from live git to a plain read of this column.

Tests

Resolver unit tests (URL normalization across ssh/https/HF, multi-harness cwd extraction, all-remotes listing), the column-order tripwire updated, full suite green.

🤖 Generated with Claude Code

[Copilot]

Pull request overview

This PR moves session-to-repository attribution from read time to index time by resolving git remotes for a session’s recorded working directory and persisting the result in a new repo column. This makes curation attribution robust even if the original checkout is later moved/deleted and avoids repeated live git lookups during reads.

Changes:

• Add src/repo.rs resolver to extract transcript cwd and normalize git remote -v URLs into owner/name identities.
• Extend the Lance schema and indexing pipeline to stamp each chunk with a per-session repo value (cached per cwd during indexing).
• Add one-off backfill examples for local and Hub stores, plus a remote hf_dataset::add_column helper to land add_columns output in a single guarded commit.

Reviewed changes

Copilot reviewed 9 out of 9 changed files in this pull request and generated 2 comments.

Show a summary per file

File	Description
src/repo.rs	Adds transcript cwd extraction + git remote normalization/resolution + unit tests.
src/lib.rs	Exposes the new repo module.
src/index.rs	Adds repo column to schema/batches and stamps chunk repo at index time with a per-cwd cache.
src/hf_dataset.rs	Adds add_column() to commit remote add_columns outputs in one guarded commit.
src/hello.rs	Updates test/fixture chunk construction for the new repo field.
src/chunk.rs	Adds repo field to Chunk and wires it through chunk (de)serialization paths.
scripts/backfill_repo.rs	Local-store one-off backfill example using Lance add_columns.
scripts/backfill_remote_repo.rs	Hub-store one-off backfill example using remote add_column commit pattern.
Cargo.toml	Registers the two new backfill examples.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.