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. The pitch, per the project's own README, is 1.3 billion or more free tokens a month across the free tiers those providers hand out, plus optional token compression it says cuts usage by up to 90%. Two things worth knowing before you install it: it is AGPL-3.0, and the author explicitly states it strips integrations when a provider's terms stop allowing them — which is the honest version of this category and also a warning that the free faucets move. The repo now carries more than 10,000 forks, so plenty of people are already betting an afternoon on it.
agent-skills (JavaScript, 116,944 stars) is Addy Osmani's pack of 25 skills and nine commands that walk an agent through a real engineering lifecycle: spec, plan, build, test, review, ship. It is the most-starred repo on today's board after llama.cpp, which is remarkable for what is essentially an opinionated set of Markdown workflows — MIT-licensed and installable into 70-plus agents through Vercel's skills CLI. The opinionated part is that it enforces gates rather than suggesting them: tests as proof, one atomic task at a time, a five-axis review before merge. Worth installing if your agent writes code faster than you can review it, which is most agents now.
qwen-code (TypeScript, 30,058 stars) is Alibaba's terminal coding agent, and the interesting news is the 1.0 preview: it folds the agent harness from Qoder — Alibaba's commercial coding tool — into the open-source release, bringing auto-memory, subagents, agent teams and MCP support with it. The repo speaks OpenAI, Anthropic, Gemini and Qwen APIs interchangeably, runs local models through Ollama or vLLM, and reaches beyond the terminal into IDE plugins, a desktop app and chat integrations. The self-referential detail is the one to watch: Qwen says the agent files its own issues, submits its own pull requests and reviews its own code.
OpenBot (TypeScript, 11,944 stars) comes from CopilotKit, the team behind the AG-UI protocol, and the framing is sharper than most: not a chatbot but coworkers, each with a computer — a browser, files, tools — where every action is decided before it happens and recorded after. It ships as a signed desktop app and an MIT-licensed workspace you can run on a laptop or a Kubernetes cluster, and it will adopt agents you already run through AG-UI endpoints rather than asking you to rewrite them. The governance layer is the point: company sign-in, tool permissions, approvals and audit across the whole workspace. Trendshift had it as the third-fastest-rising repo of the day.
The pattern across all four is the same, and it is worth naming: nobody is competing on the model anymore. They are competing on the harness around it, who pays for the tokens, and who gets to approve what the agent does next.
Which of these would you actually put in front of your team — the free routing layer, the skills pack, or the governed workspace? Write to us via our contact page.
