feat(dashboard): add pagination, server-side search

[gora050]

Summary

• Added "Load More" for Skills list on /dashboard page via "Load More"
• Extended .gitignore

export-1775137075662.mp4

(AI Assisted)

Changes

• Add listDashboardPaginated query using Convex .paginate() for incremental loading of publisher skills
• Update dashboard UI with "Load More" button

Test plan

• Open /dashboard with a publisher that has many skills — initial load shows 50, "Load More" fetches more
• Verify "Load More" disappears when all skills are loaded

[vercel]

@gora050 is attempting to deploy a commit to the 0xBuns Team on Vercel.

A member of the Team first needs to authorize it.

[greptile-apps]

Greptile Summary

This PR adds server-side pagination to the Publisher Dashboard's skills list, replacing the old useQuery(api.skills.list, { limit: 100 }) call with usePaginatedQuery(api.skills.listDashboardPaginated, …, { initialNumItems: 50 }) and a "Load More" button. A companion toDashboardPage helper exposes pending.scan hidden skills to the owner, and a new test file validates the four main access paths.

Key changes

• New listDashboardPaginated Convex query with index-level soft-delete filtering (.filter((q) => q.eq(q.field("softDeletedAt"), undefined)) before .paginate()) — correctly avoids the post-paginate JS cursor mismatch.
• toDashboardPage helper maps non-public pending-scan skills into a pendingReview: true shape for owners.
• Dashboard UI shows a "Load More" button when skillsStatus === "CanLoadMore" and a spinner while loading.
• .gitignore extended with .agents, /skills (rooted), skills-lock.json, .claude.

Issues found

• The "Load More" button and the "No skills yet" empty state are in separate render branches without coordination. When all items in the first page are post-processed away (e.g. every skill is moderator-suspended), skills.length === 0 shows the empty state while skillsStatus === "CanLoadMore" still renders the button below it — a confusing UI state.
• toDashboardPage silently drops any owner skill whose moderationStatus is neither active nor exactly hidden/pending.scan, so suspended or otherwise moderated skills are invisible to the owner on their own dashboard.

Confidence Score: 4/5

Safe to merge with minor fixes — no data loss or security risk, but two edge-case correctness issues should be addressed before the feature is considered complete.

The core pagination implementation is solid: the soft-delete filter runs inside Convex before pagination, so cursor integrity is maintained. The two remaining findings are both P2 edge cases (confusing 'No skills yet' + 'Load More' co-appearance, and silently dropped moderated skills on the owner's own dashboard), but they affect correctness of the dashboard view in real (if rare) scenarios, warranting a score of 4 rather than 5.

src/routes/dashboard.tsx (empty-state / Load More coordination) and convex/skills.ts (toDashboardPage hidden-skill visibility for owners).

Prompt To Fix All With AI

This is a comment left during a code review.
Path: src/routes/dashboard.tsx
Line: 208-214

Comment:
**"Load More" can appear alongside the "No skills yet" empty state**

`skills.length === 0` is used to pick between the empty-state and the list, but `skillsStatus === "CanLoadMore"` is checked separately—outside that branch. If every item in the first page is filtered out by `toDashboardPage` (e.g. moderator-suspended skills for the owner, or non-public skills for a non-owner viewer), `paginatedSkills` will be empty while `usePaginatedQuery` still reports `CanLoadMore`. The result is the "No skills yet" message rendered above the "Load More" button simultaneously, which is confusing and implies there is nothing to load.

Guard the "Load More" / "LoadingMore" rows on `skills.length > 0`, or show them unconditionally outside the empty-state branch:

```suggestion
            {skills.length === 0 && skillsStatus !== "CanLoadMore" ? (
```

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

---

This is a comment left during a code review.
Path: convex/skills.ts
Line: 2263-2295

Comment:
**`toDashboardPage` silently drops owner-visible skills with non-`pending.scan` hidden status**

When `isOwnDashboard` is `true`, `toDashboardPage` shows a skill only if `toPublicSkill` succeeds (active status) OR `moderationReason === "pending.scan"`. Any other non-public state—e.g. `moderationStatus: "suspended"`, `"removed"`, or a future `"pending.human-review"` reason—will silently fall through to `return null` and be excluded from the owner's dashboard view.

This means an owner could lose visibility into why certain skills are absent. The original `api.skills.list` had the same blind spot, but `listDashboardPaginated` is explicitly described as a fuller owner view ("hidden skill versions remain visible here while checks are pending"), so the intent seems to be to expose more, not less.

Consider surfacing all non-deleted skills to their owner regardless of moderation state, or at minimum mapping every moderation status into a labelled entry rather than silently dropping it:

```ts
// inside isOwnDashboard branch — replace the isPending guard:
return {
  _id: skill._id,
  // ...fields...
  pendingReview: skill.moderationStatus !== "active",
  moderationStatus: skill.moderationStatus,
  moderationReason: skill.moderationReason,
};
```

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

_{Reviews (2): Last reviewed commit: "tests" | Re-trigger Greptile}

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 026b2929f9

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]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 79d04b45d4

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]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 89fcb1b600

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".

[gora050]

Originally this PR also had dashboard search (server-side full-text by name/description).

During review it became clear the whole search setup needs fixing — same issues exist in public /skills?q= (500-row cap on lexical fallback, zero-vector skills don’t show up in vector search).

So instead of hacking a dashboard-only fix, I cut this PR down to just pagination. Will handle search properly in a follow-up.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 9d9531413d

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]

Include linked-user legacy skills in publisher pagination

listDashboardPaginated only paginates by_owner_publisher, but the existing skills.list behavior for personal publishers explicitly merges linked-user legacy rows (ownerPublisherId unset) so pre-backfill skills stay visible. This new path drops that fallback, so requests using ownerPublisherId can silently miss legacy personal skills and loadMore will never recover them because they are on a different index path.

Useful? React with 👍 / 👎.

[chatgpt-codex-connector]

Avoid filtering soft-deletes after owner index scans

Both paginated branches call .filter(q => q.eq(q.field("softDeletedAt"), undefined)) after by_owner/by_owner_publisher, which forces Convex to read through deleted rows inside each owner partition before serving a page. For owners with many soft-deleted skills this can significantly inflate document reads and can hit read limits; this path should paginate on composite active-owner indexes instead of post-index filters.

Useful? React with 👍 / 👎.