Collection basis: GitHub Trending bot issue happydog-intj/ai-xiaohongshu-daily #596 (morning edition, 2026-10-26 01:00 slot, Top 7) compared with the previous day’s #593 (2026-10-25, Top 8). Primary metric is stars gained today. Excluded: repos above 40k total stars, long-running top-ranked repos, slowing repos, non-AI and perennial large repos.
Nutlope/hallmark — +1,359 ⭐ today
URL: https://github.com/Nutlope/hallmark
Total stars: 33,290 · Day-over-day: +567 → +1,359 (+140%) · Language: CSS · License: MIT
Hallmark, from Together AI, is a design skill for Claude Code, Cursor and Codex that steers coding agents toward UIs that do not look AI-generated. Its core is a rule set in skills/hallmark/SKILL.md plus a references/ folder: it first picks a page macrostructure suited to the brief, applies one of 21 visual themes, then runs 57 “slop-test” gates and a self-critique pass before output to strip out generic AI design defaults. Each page is emitted as self-contained HTML/CSS, and briefs that fit no catalog theme switch to a Custom mode that designs palette, typography and layout from scratch. It ships these commands:
- Default (build): generate new UI under the rule set and verify it before handing it back
hallmark audit <target>: score existing code against anti-patterns and return a punch listhallmark redesign <target>: rebuild the structure while keeping copy, information architecture and brandhallmark study <screenshot | URL>: extract macrostructure, type pairing and color anchor into a portabledesign.md(refuses pixel clones and paid templates)
Installation is a single npx skills add nutlope/hallmark. Launched in July and quiet for a while, it jumped from +567 yesterday to +1,359 today, the largest gain among non-excluded repos. Growing fatigue with the look-alike design of vibe-coded apps appears to be driving demand for design guardrails built for agents.
Practical use: When spinning up landing pages or internal dashboards with AI, use the audit/redesign commands as a design QA step to remove the “template look.”
#AgentSkills #FrontendDesign #ClaudeCode #Cursor #Codex
makecindy/cindy — +677 ⭐ today (2 days in a row)
URL: https://github.com/makecindy/cindy
Total stars: 4,652 · Day-over-day: +471 → +677 (+44%) · Language: TypeScript · License: Apache-2.0
Cindy is an open-source local AI agent that works out of the box, operating on your real files and logged-in apps and driving your browser, computer and phone. It is a pnpm monorepo with an Electron desktop app (apps/desktop), an Expo/React Native mobile app (apps/mobile) and shared packages for auth, device linking, agent orchestration and model providers (packages/*), and it bundles Claude Code and Codex binaries as harnesses. Key features:
- Workspace, memory, skills and tools persist when you switch models or harnesses mid-task
- Splits a task into parallel parts and has agents on different harness × model combinations cross-review results (Orca multi-agent orchestration)
- Memory, skills, scheduled automation, MCP connections and plugins
- Choice of official cloud, existing Claude Code/Codex subscriptions, your own API keys or local models (skip-sign-in mode supported)
The official cloud backend lives in a separate repo, but the client is Apache-2.0 and can be audited and forked. After debuting yesterday at +471, it accelerated 44% to +677 today, showing sustained interest in unified desktop agents that reuse existing coding-agent subscriptions.
Practical use: Teams that already pay for Claude Code or Codex can trial desktop/mobile automation and cross-review workflows without an extra bill.
#DesktopAgent #MultiAgentOrchestration #AIAgent #MCP
Excluded: cloudflare/security-audit-skill (+2,874, 46.3k total — above 40k), pbakaus/impeccable (87.3k) and TauricResearch/TradingAgents (114.2k) — above 40k, Renhuai123/ziwei-doushu (non-AI), JetBrains/thinkrail (+52, low growth). Hallmark’s Trendshift badge history could not be verified.
Leave a Reply