Repository navigation
chore: clear the PR backlog and fix 18 open issues - #184
Merged
Merged
Conversation
mcp-handler 2.x drops the 3rd config arg (basePath/maxDuration) and the SSE transport, and swaps its required peer dep to @modelcontextprotocol/server v2. Move the handler to a static /api/mcp route (the [transport] dynamic segment only existed for the now-removed SSE transport), collapse createMcpHandler to its one-arg form, and re-express maxDuration as a Next.js route-segment export. Drop the orphaned @modelcontextprotocol/sdk v1 (imported nowhere after this). Verified: next build, 87/87 tests, plus a live initialize + tools/list against the built /api/mcp endpoint.
Brings Next.js to 16.4.6 and refreshes every dependency within its declared range, covering the open Dependabot bumps (vite, hono, vitest, dompurify, sharp, js-yaml, qs, fast-uri, brace-expansion and friends) in one pass. npm audit drops from 25 findings (2 critical, 7 high) to 12 low/moderate ones that have no non-breaking fix yet.
new Date() moved <lastmod> on every deploy, which told crawlers the pages had changed when they hadn't. The dates are now constants that get bumped by hand when the content really changes. Closes #46
Nothing imported src/lib/export-markdown.tsx. The app exports through renderMarkdownToHtml in src/lib/markdown/to-html.ts. Closes #68
The ID was hardcoded, so every fork and self-hosted copy reported its visitors to the upstream property. GA now loads only when a deployment sets NEXT_PUBLIC_GA_ID, and .env.example documents it as optional. Closes #121
…t-src CSP3 browsers already ignore both next to strict-dynamic, but CSP2-only browsers honored them, which switched off XSS protection there. The trade-off (no third-party analytics on those legacy browsers) is documented next to the policy. Closes #85
npm run typecheck runs tsc --noEmit so type errors show up in seconds instead of at the end of a full build, and CI runs it before the build. engines now says >=22, which matches .nvmrc and what CI tests. Closes #129
Also drops the Dependabot ignore for @modelcontextprotocol/sdk, which is no longer a dependency after the mcp-handler v2 migration. The malformed AI-SDK entry in .env.example was already gone. Closes #80
The export body was already sanitized (remark-html sanitizes by default), but a stale comment called it unsanitized and nothing tested it. Adds regression tests for script tags, event handlers and javascript: URLs, corrects the comment, and tells MCP callers what markdown_to_html returns. Closes #123
Adds one small limiter (src/lib/rate-limit.ts) that both public routes share. It counts per client IP in Upstash Redis when UPSTASH_REDIS_REST_URL/TOKEN are set, and falls back to an in-memory window per instance when they aren't. /api/mcp: 60 requests a minute per IP, 429 with Retry-After beyond that, and a 2 MB request body cap checked against the bytes that really arrive (413). /api/skill/improve: 10 requests a minute per IP, and a missing Origin header is now rejected like a mismatched one, so curl without Origin no longer skips the guard. Closes #122 Closes #84
Vitest now runs two projects picked by extension: *.test.ts in Node and *.test.tsx in jsdom with Testing Library. Before this a .test.tsx file was silently skipped by the include glob. Seeds it with a fake-timer test for useDebouncedValue. Closes #124
Adds tests for the GitHub tree filtering in discoverSkills (including lookalike path prefixes), the 30 file and 1 MB caps in importSkillDir, sizedAvatar and bot filtering in fetchContributors, and the JSON fallback paths of useLocalStorage. Closes #125
…itor
The only caller always passed bare with showToolbar={false}, so the
ring-styled wrapper, its embedded toolbar, the focus state and the
analytics hook never ran. The component is now just the editor the
workspace uses.
Closes #69
…t.ts smartInsert, smartHeading, insertLine, removeFormatting and replaceFormatting are now pure functions that take the document text and selection and return an edit, with 30 unit tests. The toolbar keeps the buttons, shortcuts and analytics and drops from 624 to 351 lines. The tests surfaced one bug, fixed here: toggling a format off left the caret a full word too far right, and at the end of a short document the position landed outside the document. Closes #75
… show quality hints Readability pass over the workspace UI, no behavior change: - split-pane bounds, resize step, preview debounce, reading speed and the print delay are named constants - the skill inspector's stacked ternaries are lookup maps keyed by status - the two raw hex colors are ms-script and ms-tip-muted tokens - useAnalytics returns stable module-level functions instead of wrapping each one in useCallback Also wires skillQualityHints into the inspector's validation list. It was only ever called from its test. "Trigger clarity" now uses the same tested rule, and the other hints (generic description, long body, no instruction-shaped headings) appear as warning rows when they apply. They never block export. Closes #76 Closes #71
Adds the merged features that never made it into the changelog (code block copy button, custom export file name, skip link, README shortcuts and screenshots, among others) and the changes from this branch. Closes #126
|
The latest updates on your projects. Learn more about Vercel for GitHub.
|
Owner
Author
|
CI note: "Lint and build" and both CodeQL "Analyze" jobs are red, but not because of this diff.
That pattern usually means the jobs are being refused before they start: an Actions spending limit, used-up minutes, a billing problem, or a GitHub Actions incident. There is nothing to fix in the branch for it, and I can't see the account's billing or usage from here. The last green Actions run on this repo was October 22. Once Actions is accepting jobs again, re-running the failed jobs should give a real result. Generated by Claude Code |
Rinava
marked this pull request as ready for review
November 11, 2026 00:12
This branch was successfully deployed
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Description
One branch that lands the open PRs that were ready, replaces the pile of Dependabot bumps with a single update, and works through the open issues that could be fixed without a product decision. Each issue has its own commit, so it can be reviewed (or reverted) piece by piece.
Open PRs folded in
npm auditgoes from 25 findings (2 critical, 7 high) to 12 low/moderate ones that have no non-breaking fix yet. Dependabot should close its own PRs once this is onmain.Issues fixed
lastModifiedis a fixed date, notnew Date()src/lib/export-markdown.tsxMarkdownEditorreduced to the one mode that is usedskillQualityHintsis wired into the skill inspector as warning rowssrc/lib/markdown/format.tswith 30 testsuseAnalyticsoverridespins documented in CONTRIBUTING, stale Dependabot ignore removed/api/skill/improve: rate limit, and a missingOriginis rejectedscript-srcdrops the'unsafe-inline'/https:legacy fallbackNEXT_PUBLIC_GA_ID/api/mcp: per-IP rate limit (429 +Retry-After) and a 2 MB body cap (413)remark-html; now pinned by tests and documented*.test.tsxin jsdom with Testing Library[Unreleased]backfillednpm run typecheck, CI step, engines aligned on Node 22Things worth a look before merging
NEXT_PUBLIC_GA_ID=G-YYGPWZ1WF7is set in the hosting environment. That is what the issue asked for, but it is easy to miss.UPSTASH_REDIS_REST_URL/UPSTASH_REDIS_REST_TOKENthe counters are in memory per instance, which is only a speed bump on serverless. The Upstash path uses its REST pipeline API directly (no new dependency) and is unit tested against a mocked response, but I had no real Upstash database to run it against.hints.ts, so some descriptions that used to pass that row will show a warning. It never blocks export.remark-htmlsanitizes by default), so there was no pipeline change, only tests, a corrected comment and a clearer MCP tool description.Cannot read properties of null (reading 'edgesOut'), an arborist bug withoverrides).npm cion Node 22 / npm 10 works fine with the result.Not done here
export-toolbar.tsx,app-sidebar.tsx). The fake progress bar and the duplicate sidebar heading are both gone from the code, so these can be closed as obsolete.SkillInspectorandWorkspace), Synchronized scrolling between editor and preview panes #62 (scroll sync), Add theme-color control and make HTML/PDF export match the preview exactly #63 and Add selectable export themes (GitHub Light/Dark, Sepia, Academic, Minimal) for PDF/HTML export #11 (export theming), Real-time collaborative editing via WebRTC (Yjs + y-codemirror.next + y-webrtc) #14 (collaboration): real work that deserves its own PR each.Related issue
Closes #41, closes #43, closes #46, closes #68, closes #69, closes #71, closes #75, closes #76, closes #80, closes #84, closes #85, closes #121, closes #122, closes #123, closes #124, closes #125, closes #126, closes #129.
Closes #19 and closes #37 as obsolete (see above).
Type of change
Breaking for deployers only: Node 22 minimum, GA needs
NEXT_PUBLIC_GA_ID, and the SSE MCP transport is gone (from #158).Screenshots / recording
The only visible change is the extra warning rows in the skill inspector's validation list ("Description style", "Body length", "Instruction structure"), which reuse the existing row styling.
Checklist
mainnpm run lintpasses (0 errors, the one existingset-state-in-effectwarning inoutline-rail.tsx)npm testpasses (158 tests, up from 87)npm run buildsucceedsinitializeandtools/call, 413, 429 and the Origin guard with curl)Generated by Claude Code