refactor: followups (reserved slugs, moderation batches, http api split)

[steipete]

Followups after #298/#313.

• reservedSlugs: add index + query helpers (no take+sort)
• moderation: move user-ban/unban skill+embedding batching into skills; add scheduler helper
• httpApiV1: split handlers into modules; keep exports stable + __handlers for tests

Gate:

• bunx convex codegen
• bun run lint
• bun run build
• bun run test

Greptile Overview

Greptile Summary

Refactored moderation batch operations and HTTP API structure with improved query performance.

• Moved user ban/unban skill batching from users.ts to skills.ts for better domain separation
• Split monolithic httpApiV1.ts (~1400 lines) into domain-specific modules (skillsV1.ts, soulsV1.ts, usersV1.ts, etc.) under httpApiV1/ directory, maintaining stable exports via __handlers for tests
• Added by_slug_active_deletedAt compound index to reservedSlugs table to eliminate need for in-memory filtering and sorting - queries now use index ordering directly
• Introduced scheduleNextBatchIfNeeded helper in lib/batching.ts to consolidate batch continuation logic
• Refactored reserved slug helpers to use new index and eliminate pickLatestActiveReservation helper function

Confidence Score: 4/5

• This PR is safe to merge with one logic concern to verify
• Score reflects clean refactoring with improved modularity and query performance. The code structure improvements are well-executed, with proper domain separation and stable exports maintained. One potential logic change in moderation status handling needs verification (treating undefined as 'active' may affect legacy skills), but overall changes follow established patterns and maintain existing functionality.
• Pay close attention to convex/skills.ts:2170 for the moderation status logic change

_{Last reviewed commit: 61b41f9}

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
clawhub	Ready	Preview, Comment	Feb 15, 2026 2:25am

[cursor]

PR Summary

Medium Risk
Large refactor of HTTP handler code and data-access patterns; behavior should be unchanged but routing/parsing and reserved-slug query semantics could regress without thorough endpoint and migration testing.

Overview
Refactors the Convex httpApiV1 surface into modular handlers by extracting the large monolithic httpApiV1.ts implementation into httpApiV1/*V1.ts files plus a shared.ts helper (response helpers, path parsing, multipart publish parsing, tag resolution, soft-delete error mapping), while keeping the exported httpAction endpoints and __handlers stable.

Optimizes reserved-slug handling by adding a new reservedSlugs index (by_slug_active_deletedAt) and rewriting lookups to query only active reservations in descending deletedAt order (removing the previous take+sort approach), with tests updated accordingly.

Moves ban/unban skill moderation batching into skills.ts (including embedding visibility updates), introduces lib/batching.ts (scheduleNextBatchIfNeeded) for consistent cursor-based rescheduling, and updates users.ts to call the new internal.skills.* batch mutations.

^{Written by Cursor Bugbot for commit 61b41f9. This will update automatically on new commits. Configure here.}

[greptile-apps]

_{14 files reviewed, 1 comment}

_{Edit Code Review Agent Settings | Greptile}

[greptile-apps]

logic change from previous code - old version checked skill.moderationStatus === 'active' (strict equality), which would be false for undefined values. New code treats undefined as 'active' via the nullish coalescing operator, causing skills with undefined moderationStatus to have their moderation fields overwritten during ban operations. This could be intentional for normalizing legacy data, but if not:

Suggested change

	const shouldMarkModeration = (skill.moderationStatus ?? 'active') === 'active'
	const shouldMarkModeration = skill.moderationStatus === 'active'

Prompt To Fix With AI

This is a comment left during a code review.
Path: convex/skills.ts
Line: 2170:2170

Comment:
logic change from previous code - old version checked `skill.moderationStatus === 'active'` (strict equality), which would be `false` for `undefined` values. New code treats `undefined` as `'active'` via the nullish coalescing operator, causing skills with `undefined` moderationStatus to have their moderation fields overwritten during ban operations. This could be intentional for normalizing legacy data, but if not:

```suggestion
      const shouldMarkModeration = skill.moderationStatus === 'active'
```

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

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 61b41f9fda

ℹ️ About Codex in GitHub

Your team has set up Codex to 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 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Await scheduler enqueue for continuation batches

scheduleNextBatchIfNeeded now fires scheduler.runAfter without awaiting or handling rejection, so a scheduler error (e.g., transient failure or invalid function reference during deploy) will let applyBanToOwnedSkillsBatchInternal / restoreOwnedSkillsForUnbanBatchInternal return { ok: true } even though the next page was never enqueued, leaving large owner skill sets only partially processed. Before this refactor the equivalent calls in users.ts were awaited, which surfaced enqueue failures instead of silently truncating the batch workflow.

Useful? React with 👍 / 👎.