The theme connecting today's trending projects is control: developers are no longer asking whether AI can help them code, communicate, or invest — they're building structured systems to ensure AI does these things correctly, safely, and consistently. From bias-aware investment agents to zero-identifier messaging, today's repos are infrastructure choices, not experiments.
AI Berkshire: Four Investment Legends, Four Simultaneous Agents
xbtlin/ai-berkshire surged onto GitHub trending with +1,274 stars today. The framework applies the methodologies of Warren Buffett, Charlie Munger, Duan Yongping, and Li Lu in parallel — four independent Claude Code agents analyzing the same company simultaneously, then synthesizing findings into a unified report. The key design insight is that the four perspectives operate in deliberate tension: Buffett might flag an attractively priced stock that Li Lu's framework rejects on management culture grounds. Built-in features like Benford's Law anomaly detection and forced explicit recommendations (no "it's complicated" outputs) give it structural anti-bias properties that a single-prompt approach can't replicate.
github.com/xbtlin/ai-berkshire
no-mistakes: An AI Gate Before Every git push
kunchenguid/no-mistakes picked up +398 stars today with a simple premise: your commits should not leave your machine until they're clean. The tool sits between your local branch and the remote, intercepting pushes and running an AI-powered validation pipeline — review, tests, linting, and documentation checks — before forwarding anything upstream. Safe fixes apply automatically; anything requiring judgment escalates to the developer. The /no-mistakes command also integrates as a Claude Code skill, making it usable directly inside an agentic workflow. For teams tired of "WIP" commits or broken CI runs from rushed pushes, it moves quality checks from the review stage to the commit stage.
github.com/kunchenguid/no-mistakes
SimpleX Chat: Messaging with No User Identifiers
simplex-chat/simplex-chat gained +432 stars today, pushing its total to 13,050. The project's core premise is unusual even among privacy tools: it assigns no persistent identifiers to users — not usernames, not account numbers, not even anonymous IDs. Each conversation uses disposable, pairwise identifiers, making network graph analysis substantially harder for any observer. The protocol uses a double ratchet encryption scheme with post-quantum key exchange on every step, and is fully self-hostable. Recent v6.4.1 adds group member review controls and app security improvements. In a week when attention is on how much metadata messaging apps expose, SimpleX's zero-ID design makes a clear architectural statement.
github.com/simplex-chat/simplex-chat
AWS Ships an Official MCP Toolkit for AI Agents
aws/agent-toolkit-for-aws landed +243 stars today, and the significance is less in the star count and more in the publisher. AWS has officially released a toolkit of MCP servers, skills, and plugins that give AI coding agents authenticated access to more than 300 AWS services through a single endpoint. Claude Code, Cursor, and Codex are all listed as supported agents. The toolkit covers core infrastructure, serverless, containers, data analytics, and a dedicated DevSecOps plugin for incident investigation and vulnerability scanning. An official AWS MCP integration means agent-driven cloud deployments are now a supported first-class workflow — not a workaround built by the community.