fix(chat-store): the index seam must not widen the table's column types

[drewstone]

Follow-up to #346, caught by the first consumer that adopted it.

ChatExtraIndexes returned unknown[]. That forced a cast on the whole
extra-config array handed to sqliteTable, and the cast widened every column
in the emitted types to drizzle's cross-dialect union. agent-app's own
typecheck stayed clean; gtm's produced five errors on a single db.select:

Type 'PgColumn<{ name: "id"; tableName: TTableName; columnType: "SQLiteText"; … }>'
is not assignable to type '… | SQLiteColumn<…> | …'
  src/lib/.server/chat/gtm-chat-vertical.ts:766

The callback now returns drizzle's own SQLiteTableExtraConfigValue[], so the
spread stays inside the union sqliteTable expects and no cast is needed.

About the tests

Three new assertions pin that a table built WITH product indexes still inserts,
selects, and reports SQLite column types the same as one built without.

They did not catch this regression, and the comment says so. Reverting to
unknown[] + the cast produces 0 errors in this package — the widening is
invisible from inside it and only appears through the emitted .d.ts in a
consumer whose own db type meets the table. Writing that down beats leaving a
guard that reads stronger than it is; the real guard is a consumer's typecheck,
which is where the defect surfaced.

Verification

$ pnpm typecheck                      # 0 errors
$ pnpm docs:gen                       # regenerated, committed
$ pnpm test
  Test Files  231 passed | 6 skipped (237)
       Tests  3851 passed | 13 skipped (3864)

And the consumer that found it, on this build:

gtm-agent $ pnpm typecheck            # 0 errors  (was 5)

[tangletools]

✅ Auto-approved drewstone PR — f331bdfd

This PR was opened by the trusted drewstone account.
The full PR reviewer audit still runs separately and will publish findings if it detects issues.

_{tangletools · auto-approval · reason: drewstone_author · 2026-07-29T19:47:55Z}

[tangletools]

🟢 Value Audit — sound

Verdict	sound
Concerns	0 (none)
Heuristic	0.0s
Duplication	0.0s
Interrogation	370.9s (2 bridge agents)
Total	370.9s

💰 Value — sound

Tightens the index-seam return type from unknown[] to drizzle's own SQLiteTableExtraConfigValue[], letting the as never cast be dropped so the built tables keep their inferred column types — correct, minimal, proven by drizzle's own type signature.

• What it does: Two-commit follow-up to #346. ChatExtraIndexes (src/chat-store/schema.ts:101) and the internal extraIndexes helper (schema.ts:119) now return SQLiteTableExtraConfigValue[] instead of unknown[]. That type is exactly what drizzle's sqliteTable extra-config callback is declared to return (drizzle sqlite-core/table.d.ts:22,28), so the spread ...extraIndexes(...) stays inside the union th
• Goals it achieves: Restore correct emitted .d.ts column types for consumers of createChatTables. #346's unknown[] + as never widened every column to drizzle's cross-dialect union in the emitted types; this package's own typecheck stayed green, but a consumer whose db meets the table (gtm) hit 5 errors on a single db.select. The fix makes the seam type-check cleanly on its own, with no consumer-visible wi
• Assessment: Good change, in the grain of the codebase. It uses drizzle's own exported type for exactly the slot drizzle defines — no invented wrapper, no broader redesign. It targets the root cause (the as never cast that was forced by unknown[]) rather than patching symptoms. The new tests are unusually honest: their leading comment explicitly states they do NOT catch the regression (the widening is invi
• Better / existing approach: none — this is the right approach. Searched for SQLiteTableExtraConfigValue across the repo (3 hits, all this PR) and read the sibling factory createTeamTables (src/teams/drizzle/schema.ts:44) which has no index seam and no cast. The seam itself is justified (it's what #346 added so a product with one extra index can adopt the factory instead of hand-rolling a duplicate), so eliminating it is
• Model: opencode/zai-coding-plan/glm-5.2
• Bridge attempts: 2
• Bridge warning: opencode/kimi-for-coding/k2p7: bridge stream ended without value-audit content

🎯 Usefulness — sound

A minimal, correct type fix that swaps an invented unknown[]+as never escape hatch for drizzle's own canonical extra-config type, unblocking the seam's first consumer (gtm-agent) whose typecheck the old widening broke.

• Integration: The threadExtraIndexes/messageExtraIndexes seam lives on createChatTables (src/chat-store/schema.ts:85,88), exported via the /chat-store subpath and composed by createChatStore (src/chat-store/store.ts:2). It is reachable and its first real consumer is gtm-agent — PR #346 built this seam specifically to let gtm drop its hand-rolled duplicate message table (the PR body of #346 names exa
• Fit with existing patterns: Fits the grain perfectly and removes a competing workaround rather than adding one. The prior unknown[] return forced an as never cast on the whole config array, which fought drizzle's type inference and widened emitted columns to the cross-dialect union. The fix uses SQLiteTableExtraConfigValue — verified to be the exact type sqliteTable's array-config overload already expects (drizzle-or
• Real-world viability: Pure type-level change: a return-type annotation narrows from unknown[] to SQLiteTableExtraConfigValue[] and the as never cast is dropped. Zero runtime behavior change, and the runtime tests pin that insert/select/index behavior is identical with or without product indexes. The load-bearing proof is consumer-facing: the PR body shows gtm-agent's typecheck going 5 errors → 0 against the produ
• Model: opencode/zai-coding-plan/glm-5.2
• Bridge attempts: 1

No concerns — sound change, no better or existing approach found. ✅

---

What this audit checks

It judges the change on its merits — not whether it was tasked out in an issue. Unticketed, fast-moving work is fine; the question is whether the change is good and whether a better or existing approach should be used instead.

Pass	What it asks
Heuristic	Vague title? Whitespace-only or cruft-bearing diff? (content signals only)
Duplication	Do added function/class names already exist elsewhere in the repo?
Value Audit	What does it do? What goal does it achieve? Is it good? Better architecture or already-exists?
Usefulness Audit	Does it integrate and fit? Will it hold up in real use and actually get used?

Findings are concerns, not blocks — the human reviewer decides what to do with them.

_{value-audit · 20260729T195423Z}

[tangletools]

✅ No Blockers — f331bdfd

Review health 100/100 · Reviewer score 92/100 · Confidence 70/100 · 1 finding (1 low)

glm: Correctness 92 · Security 92 · Testing 92 · Architecture 92

Reviewer score is advisory once the run is complete and the verdict has no blockers.

Full multi-shot audit completed 2/2 planned shots over 2 changed files. Global verifier still owns final merge decision.

🟡 LOW Describe block name overpromises: tests do not guard the inference regression — tests/chat-store/chat-schema.test.ts

Empirically verified: reverting src/chat-store/schema.ts to base (unknown[] return + as never cast — the exact regression) leaves both tsc --noEmit and vitest run on THIS file fully green (19/19 pass, 0 type errors). The widening is only visible through the emitted .d.ts in a downstream consumer, never inside this package. The docblock at lines 263-277 admits this plainly ('these assertions did NOT catch the regression'), so it is honestly documented, not concealed. The residual risk is navigational: a future reader scanning describe names — or a grepper for 'inference' coverage — believes the regression class is guarded in-repo when the actual g

---

_{tangletools · 2026-07-29T19:56:55Z · trace}

[tangletools]

✅ Approved — 1 non-blocking finding — f331bdfd

Full multi-shot audit completed 2/2 planned shots over 2 changed files. Global verifier still owns final merge decision.

Full immutable report for this review: trace

Summary comment for this run: full summary

---

_{tangletools · 2026-07-29T19:56:55Z · immutable trace}