Open Source Radar — November 16: Google Stitch Agent Skills

Share
Open Source Radar — November 16: Google Stitch Agent Skills

Today's trending board is skills, not models: Google opened the design skills behind Stitch, NVIDIA shipped a grader for skills, and a retail trader published his. Tencent adds the workspace they all run in.

Google Stitch skills — the stitch-skills repository from Google Labs packages the agent skills that drive Stitch, Google's prompt-to-UI design tool, as three installable plugins: design, build and utilities. Each skill follows the Agent Skills open standard, so the same folder works in Codex, Gemini CLI, Claude Code, Cursor, Antigravity and OpenCode, and talks to Stitch through its MCP server. It has pulled in roughly 9,000 stars and more than 1,000 forks. The practical use is the boring one that matters: the agent that generates your screens also knows the house rules for how those screens get built. Worth knowing before you install — Google labels it not an officially supported product, so it carries no support commitment.


NVIDIA SkillEvaluator — NVIDIA's SkillEvaluator answers a question most skill collections dodge: is this skill any good? It runs tiered checks starting with deterministic quality gates that need no API key, adds semantic overlap detection so you don't ship two skills that do the same thing, then measures whether a skill actually changes agent behavior by comparing runs with and without it. It plugs into CI on GitHub and GitLab, so a bad skill gets caught at the pull request instead of a month later in production. About 2,100 stars, Apache 2.0, and a paper behind it.


Claude Trading Skills — a retail trader's personal Claude Code workflow, published as claude-trading-skills, and it is unusually honest about what it won't do. No signals, no "let the AI pick your trades" — the skills structure the parts people skip: a 15-minute daily market-regime check, a weekly portfolio review, swing screeners that only run when risk allows, position sizing, and a journal that forces a postmortem on every closed trade. Python, MIT, about 5,900 stars, with some paths needing Alpaca or Financial Modeling Prep keys and manual CSV fallbacks when they don't.


Tencent LoopForge — Tencent's LoopForge is a self-hosted workspace where a team ships code through coding agents rather than one terminal. Describe the job to a Main Bot and it opens a Docker sandbox, checks out your repos, delegates to project roles, and opens the pull request; when a way of working proves itself you save it as a reusable SOP with scripts and human approval gates baked in. It drives CodeBuddy, Codex, Claude Code, OpenCode, Cursor CLI, Antigravity, Kimi Code and others, and the approval gates are the point — mid-run review, not a PR at the end. Go, 13 releases and about 2,100 stars.

Worth watching this week: three separate vendors shipped skill tooling in the same 24 hours, which is usually the moment a format stops being one lab's feature and becomes infrastructure.

Do you gate your team's agent skills in CI, or just trust the ones that look good? Write to us via our contact page.

Get SiliconNoon in your inbox

One email a day with the AI stories that matter — what happened, the numbers, and why it matters. No spam, unsubscribe in one click.

Read more

Open Source Radar — November 14: OpenAI's Codex Security CLI

Open Source Radar — November 14: OpenAI's Codex Security CLI

Today's board is all about review: OpenAI's vulnerability scanner now runs as an open CLI, Chrome is teaching coding agents the modern web, and a 10,000-star app that puts a human between the agent and the merge. codex-security (TypeScript, 13,345 stars) is OpenAI's scanner for finding, validating and fixing vulnerabilities in your own repository, released as an Apache-2.0 CLI and TypeScript SDK that started as a research preview on November 6 and gained 751 stars today alone.

Open Source Radar — November 13: Free Claude Code's 1.3B Tokens

Open Source Radar — November 13: Free Claude Code's 1.3B Tokens

Today's board is about routing around the bill: a proxy that points Claude Code at free models, a skills pack from Addy Osmani, Qwen's own coding agent, and a workspace where each bot gets its own computer. All four shipped code in the last 48 hours. free-claude-code (Python, 69,656 stars) is a local proxy that sits between your coding agent and whichever model provider you point it at — 12 harnesses and 61 providers, from Claude Code and Codex to Cline and Aider.

Open Source Radar — November 8: An Android Agent That Uses Your Apps

Open Source Radar — November 8: An Android Agent That Uses Your Apps

Today's board is agents on the move again — one that tapped into your phone's own apps, one that keeps coding after you close the terminal, one that reads a video frame by frame, and one built to make the first kind expensive to point at your binary. Mika (Kotlin, about 7,700 stars, Apache-2.0) puts the agent on the phone itself: wangzan101/Mika opens apps, taps, types and scrolls on a hidden background screen while you keep using the phone, finding buttons by looking at the screen rather…

Open Source Radar — November 7: LobeHub's 84,000-Star Super-Agent

Open Source Radar — November 7: LobeHub's 84,000-Star Super-Agent

Today's board is about agents that hold down a job rather than answer a prompt: a team-chat super-agent with 84,000 stars, a Rust engine that studies any app so an agent can rebuild it, and a skill that strips the AI look out of your interface. LobeHub (TypeScript, about 84,300 stars) is the open-source platform behind LobeChat, rebuilt around what it calls super-agents: you add one to the team chat you already use, plug in any model or IM app, and it runs several tasks at once, picks up work…

Open Source Radar — November 6: VibeVoice's Hour-Long Transcript

Open Source Radar — November 6: VibeVoice's Hour-Long Transcript

Today's trending board is heavy on tooling rather than new models: Microsoft's speech family is back, two agent-facing CLIs are shipping daily fixes, and Google has quietly built a 187-skill catalog for its own cloud. VibeVoice (Python, about 56,300 stars, MIT) is Microsoft's open speech family, and the one most people will actually reach for is VibeVoice-ASR, which accepts a full 60 minutes of audio in one pass inside a 64K-token window and returns a structured transcript — who spoke, when…

Open Source Radar — November 5: A 2.78T Model on a Laptop CPU

Open Source Radar — November 5: A 2.78T Model on a Laptop CPU

Today's trending list is about plumbing rather than demos: the tools that make coding agents act like a team, keep a 2.78-trillion-parameter model off the GPU, and hand an agent data it can actually trust. Four worth a look, none of them repeat coverage from this week. gstack (TypeScript, about 139,000 stars) is Garry Tan's own Claude Code setup, published as MIT-licensed Markdown — 23 tools that reassign the agent into roles with names you'd recognise from a real org chart: a CEO that…