Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

_{Comment @coderabbitai help to get the list of available commands and usage tips.}

Greptile Summary

This PR pipes GitHub's pushed_at field (last default-branch commit timestamp) through all backend layers — V18 migration, RepoResponse, Postgres upsert, Meilisearch ingest, all route mappers, and a new /internal/backfill-pushed-at endpoint for filling existing NULL rows. It also introduces TopicCodeMapper, a new object that resolves raw GitHub topics to 15 canonical codes, and exposes topicCodes on every RepoResponse path.

• pushed_at_gh is added as a nullable TIMESTAMPTZ column (V18), populated during ingest and backfill, and surfaced as pushedAt in RepoResponse from both the Postgres and Meilisearch search paths.
• TopicCodeMapper.resolve maps raw topics to a priority-ordered list of canonical codes (ai, privacy, security, etc.) computed at response time and never persisted.
• /internal/backfill-pushed-at reuses runBackfill and the backfillRunning gate; Gone/Archived repos receive a COALESCE(updated_at_gh, indexed_at) fallback stamp so they don't resurface in future backfill queries.

Confidence Score: 3/5

Not ready to merge without confirming the Python meili_sync.py is updated; the nightly sync will otherwise silently wipe pushed_at from every Meilisearch document it touches.

The Kotlin-side wiring is well-structured and follows existing patterns, but the nightly Python meili_sync.py run uses POST (full document replacement) and is not updated in this PR. Every nightly sync will overwrite pushed_at with null for all documents, making the Meili search path always return pushedAt = null regardless of what the backfill wrote. CLAUDE.md calls this out explicitly as a required step for any new RepoResponse field. There is also a lock-leak in the new backfill endpoint (shared with the existing one) where a DB error during candidate fetching leaves backfillRunning permanently set until restart.

MeilisearchClient.kt and InternalRoutes.kt need the most attention; the former depends on an out-of-repo Python change and the latter carries the lock-leak risk.

Important Files Changed

Sequence Diagram

sequenceDiagram
    participant GH as GitHub API
    participant GSC as GitHubSearchClient
    participant PG as Postgres (repos)
    participant Meili as Meilisearch
    participant SR as SearchRepository
    participant MeiliSync as meili_sync.py (Python)

    Note over GSC,PG: On-demand ingest path
    GH->>GSC: pushed_at (ISO-8601)
    GSC->>PG: upsert pushed_at_gh (OffsetDateTime)
    GSC->>Meili: "addDocuments (pushed_at = raw string)"

    Note over SR,PG: Postgres search path
    SR->>PG: SELECT pushed_at_gh
    PG-->>SR: JDBC string format

    Note over Meili: Meilisearch search path
    Meili-->>SearchRoutes: pushed_at (ISO-8601 string)

    Note over MeiliSync,Meili: Python sync (nightly)
    MeiliSync->>Meili: POST /documents (no pushed_at field)
    Meili-->>Meili: pushed_at wiped from all docs

    Note over InternalRoutes,PG: /backfill-pushed-at
    InternalRoutes->>PG: SELECT WHERE pushed_at_gh IS NULL
    InternalRoutes->>GH: refreshRepo(fullName)
    GH-->>InternalRoutes: Ok / Gone / Archived / TransientFailure
    InternalRoutes->>PG: UPDATE pushed_at_gh (real or fallback)

_{Reviews (1): Last reviewed commit: "Merge branch 'main' into fix/per-migrati..." | Re-trigger Greptile}