perf: denormalize latestVersionSummary into skillSearchDigest

[sethconvex]

Summary

• Adds latestVersionSummary to skillSearchDigest schema and syncs it via the existing trigger (SHARED_KEYS)
• buildPublicSkillEntries now reads summary directly from HydratableSkill instead of type-narrowing with 'latestVersionSummary' in skill
• Eliminates ~9MB of skillVersions reads per listPublicPageV2 call; old rows without the field fall back to fetching the full version doc

Test plan

• npx vitest run — all 714 tests pass
• npx convex dev --once — typecheck + push to dev
• Deploy to prod, run backfillLatestVersionSummary to populate existing digest rows
• Monitor insights — skillVersions should disappear from listPublicPageV2

🤖 Generated with Claude Code

[vercel]

Someone is attempting to deploy a commit to the Amantus Machina Team on Vercel.

A member of the Team first needs to authorize it.

[greptile-apps]

Greptile Summary

This PR denormalizes latestVersionSummary into the skillSearchDigest table so that listPublicPageV2 can serve version metadata directly from the lightweight digest rows, avoiding a separate ~6KB skillVersions doc read per skill. The schema is extended consistently across both tables, the SHARED_KEYS array and HydratableSkill type are updated to include the new field, and the buildPublicSkillEntries function is simplified to read the field directly with a clean fallback to ctx.db.get for old rows that haven't been backfilled yet.

• latestVersionSummary added to skillSearchDigest schema with a validator that mirrors the skills table exactly
• SHARED_KEYS in skillSearchDigest.ts extended so the field is synced by the existing trigger
• HydratableSkill type updated so the TypeScript compiler enforces parity between the two shapes
• buildPublicSkillEntries simplified from an 'in' skill type-narrowing pattern to a direct field access; rows without the field still fall back to fetching the full version doc
• Test helpers updated; the fallback path for old (un-backfilled) digest rows is no longer covered by the listPublicPageV2 tests — worth adding a case before the backfill runs in production

Confidence Score: 4/5

• Safe to merge; the fallback logic for un-backfilled rows is correct and the schema changes are consistent.
• The implementation is correct: schema validators match between skills and skillSearchDigest, the SHARED_KEYS constraint ensures compile-time parity, and the undefined fallback to ctx.db.get is preserved for old rows. Score is 4 rather than 5 solely because the listPublicPageV2 test suite no longer exercises the fallback code path, which will be the live path in production until backfillLatestVersionSummary completes.
• convex/skills.listPublicPageV2.test.ts — missing test coverage for the fallback path when latestVersionSummary is absent

Prompt To Fix All With AI

This is a comment left during a code review.
Path: convex/skills.listPublicPageV2.test.ts
Line: 336-341

Comment:
**Fallback path no longer covered by tests**

`makeSkill` now unconditionally includes `latestVersionSummary`, so every test in this file exercises the "use summary" fast path. The regression path — where `latestVersionSummary` is `undefined` and `buildPublicSkillEntries` must fall back to `ctx.db.get(skill.latestVersionId)` — is no longer tested. This matters because the PR description explicitly calls out that backfilling existing digest rows is a post-deploy step, meaning old rows without the field will hit this code path in production.

Consider adding a test with `latestVersionSummary: undefined` (or omitting it) that verifies `db.get` is called with the version ID and the result is surfaced correctly.

How can I resolve this? If you propose a fix, please make it concise.

_{Last reviewed commit: 576588b}

[greptile-apps]

Fallback path no longer covered by tests

makeSkill now unconditionally includes latestVersionSummary, so every test in this file exercises the "use summary" fast path. The regression path — where latestVersionSummary is undefined and buildPublicSkillEntries must fall back to ctx.db.get(skill.latestVersionId) — is no longer tested. This matters because the PR description explicitly calls out that backfilling existing digest rows is a post-deploy step, meaning old rows without the field will hit this code path in production.

Consider adding a test with latestVersionSummary: undefined (or omitting it) that verifies db.get is called with the version ID and the result is surfaced correctly.

Prompt To Fix With AI

This is a comment left during a code review.
Path: convex/skills.listPublicPageV2.test.ts
Line: 336-341

Comment:
**Fallback path no longer covered by tests**

`makeSkill` now unconditionally includes `latestVersionSummary`, so every test in this file exercises the "use summary" fast path. The regression path — where `latestVersionSummary` is `undefined` and `buildPublicSkillEntries` must fall back to `ctx.db.get(skill.latestVersionId)` — is no longer tested. This matters because the PR description explicitly calls out that backfilling existing digest rows is a post-deploy step, meaning old rows without the field will hit this code path in production.

Consider adding a test with `latestVersionSummary: undefined` (or omitting it) that verifies `db.get` is called with the version ID and the result is surfaced correctly.

How can I resolve this? If you propose a fix, please make it concise.

[sethconvex]

@codex review

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: b32a499774

ℹ️ About Codex in GitHub

Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".

[chatgpt-codex-connector]

Add a migration to populate digest latestVersionSummary

Switching listPublicPageV2 to trust skill.latestVersionSummary here means the optimization only works if old skillSearchDigest rows are populated, but this commit never backfills that table: the sync path only runs when a skills row mutates (convex/functions.ts trigger), and backfillLatestVersionSummaryInternal skips already-synced skill rows (convex/maintenance.ts), so most existing digests will keep the field absent and continue hitting ctx.db.get(skill.latestVersionId) on every page. In production this prevents the intended read reduction for historical data; add a one-time digest backfill (or force-touch skills) as part of this change.

Useful? React with 👍 / 👎.