chore: merge staging into main

[BunsDev]

Merge current staging into main.

Includes the latest design-foundation sweep and related staging changes.

Recent staging PRs include:

• Design system overhaul: unified tokens, component authority, standardized layout #1570 Design system overhaul: unified tokens, component authority, standardized layout

Co-authored-by: Nova [email protected]

[vercel]

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

Project	Deployment	Actions	Updated (UTC)
clawhub	Ready	Preview	Apr 7, 2026 1:26pm

[greptile-apps]

Greptile Summary

This is a large design-system and staging merge (~11 k additions) covering a unified token/component sweep, a new /users browse page, listPublic/getHoverStats Convex queries, and a seedDemoSkills internal mutation. Two correctness issues in the newly added Convex backend code should be addressed before this lands on main.

Confidence Score: 4/5

Two P1 correctness issues in new Convex backend code should be resolved before merging.

The getHoverStats query uses an unbounded paginate loop in a reactive context and will fail for users with many skills. The seedDemo digest stats mismatch produces inconsistent data from the first seed run. Both are in newly added code paths on the backend.

convex/users.ts (getHoverStats), convex/seedDemo.ts (skillSearchDigest stats)

Prompt To Fix All With AI

This is a comment left during a code review.
Path: convex/users.ts
Line: 467-497

Comment:
**Unbounded paginate loop in reactive query**

The `for (;;)` loop calls `.paginate()` repeatedly until `page.isDone`, pulling every `skills` document owned by the user into a single query transaction. For a prolific publisher this will easily exceed Convex's 8 MB read-bandwidth budget and cause the query to fail at runtime. Because this is a reactive `query` (not an action), every skill write for that user re-runs it for every subscriber watching the hover tooltip — there is no batch-splitting escape hatch available in a query context. Per the project's performance rules, hot read paths must be denormalized rather than issuing unbounded document scans inside a reactive query.

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/seedDemo.ts
Line: 326-332

Comment:
**Digest stats diverge from canonical skills stats**

The `skillSearchDigest` insert calls fresh `Math.random()` for `stats.versions` and `stats.comments` (lines 329–330) instead of reusing the `numVersions`/`numComments` variables already computed for the paired `skills` row (lines 242–243, used at 261–262). From the moment the seed runs, the digest will show different version/comment counts than the authoritative skills document, so any UI that sources counts from the digest will disagree with the underlying record.

```suggestion
          versions: numVersions,
          comments: numComments,
```

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

_{Reviews (1): Last reviewed commit: "feat: design foundation sweep (#1570) th..." | Re-trigger Greptile}

[greptile-apps]

Unbounded paginate loop in reactive query

The for (;;) loop calls .paginate() repeatedly until page.isDone, pulling every skills document owned by the user into a single query transaction. For a prolific publisher this will easily exceed Convex's 8 MB read-bandwidth budget and cause the query to fail at runtime. Because this is a reactive query (not an action), every skill write for that user re-runs it for every subscriber watching the hover tooltip — there is no batch-splitting escape hatch available in a query context. Per the project's performance rules, hot read paths must be denormalized rather than issuing unbounded document scans inside a reactive query.

Prompt To Fix With AI

This is a comment left during a code review.
Path: convex/users.ts
Line: 467-497

Comment:
**Unbounded paginate loop in reactive query**

The `for (;;)` loop calls `.paginate()` repeatedly until `page.isDone`, pulling every `skills` document owned by the user into a single query transaction. For a prolific publisher this will easily exceed Convex's 8 MB read-bandwidth budget and cause the query to fail at runtime. Because this is a reactive `query` (not an action), every skill write for that user re-runs it for every subscriber watching the hover tooltip — there is no batch-splitting escape hatch available in a query context. Per the project's performance rules, hot read paths must be denormalized rather than issuing unbounded document scans inside a reactive query.

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

[greptile-apps]

Digest stats diverge from canonical skills stats

The skillSearchDigest insert calls fresh Math.random() for stats.versions and stats.comments (lines 329–330) instead of reusing the numVersions/numComments variables already computed for the paired skills row (lines 242–243, used at 261–262). From the moment the seed runs, the digest will show different version/comment counts than the authoritative skills document, so any UI that sources counts from the digest will disagree with the underlying record.

Suggested change

	statsDownloads: s.downloads,
	statsStars: s.stars,
	statsInstallsCurrent: Math.floor(s.installs * 0.3),
	statsInstallsAllTime: s.installs,
	softDeletedAt: undefined,
	moderationStatus: "active",
	moderationFlags: undefined,
	versions: numVersions,
	comments: numComments,

Prompt To Fix With AI

This is a comment left during a code review.
Path: convex/seedDemo.ts
Line: 326-332

Comment:
**Digest stats diverge from canonical skills stats**

The `skillSearchDigest` insert calls fresh `Math.random()` for `stats.versions` and `stats.comments` (lines 329–330) instead of reusing the `numVersions`/`numComments` variables already computed for the paired `skills` row (lines 242–243, used at 261–262). From the moment the seed runs, the digest will show different version/comment counts than the authoritative skills document, so any UI that sources counts from the digest will disagree with the underlying record.

```suggestion
          versions: numVersions,
          comments: numComments,
```

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: f636b31fca

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]

Route soul-mode search to souls results

In soul mode, the header search still submits to /search (to: "/search") even though this UI labels the box as “Search souls...”. The new search page only supports skills/plugins/users tabs (src/routes/search.tsx), so soul-mode users can no longer search souls from the primary header control. This is a functional regression from the previous soul-mode flow and will return the wrong result set on souls hosts.

Useful? React with 👍 / 👎.

[chatgpt-codex-connector]

Bound tooltip stats query to avoid full owner scans

getHoverStats paginates through all skills for the owner and accumulates them before returning counts. Because this query is called from user-hover tooltips, high-skill accounts can trigger very expensive reads (and potentially Convex read limits/timeouts), causing slow or failed tooltip loads in production. This should use denormalized counters or a bounded/aggregated approach instead of unbounded pagination.

Useful? React with 👍 / 👎.

[chatgpt-codex-connector]

Remove hardcoded skills seeded from repository content

/workspace/clawhub/AGENTS.md states to reject PRs that add skills directly in repo content and require publishing via CLI. This commit introduces a large hardcoded DEMO_SKILLS list and mutation-driven inserts, which bypasses the mandated publication path and embeds skill catalog data directly in source code.

Useful? React with 👍 / 👎.