Open Source Releases

  • decolua/9router — A routing proxy that lets your AI coding agents (Claude Code, Codex, Cursor, Cline, etc.) talk to free Claude/GPT/Gemini endpoints across 40+ providers with auto-fallback and ~40% token reduction. “Unlimited FREE AI coding” is a bold claim — your mileage will depend on how long those free tiers last and how you feel about your code flowing through a chain of third-party proxies. 🤖

  • anthropics/claude-code: v2.1.138 — Internal fixes. The kind of release that exists so the next one doesn’t have to.

  • anthropics/claude-code: v2.1.137 — Fixed the VS Code extension failing to activate on Windows. If you’ve been staring at a greyed-out sidebar, this is your patch.

  • anthropics/claude-code: v2.1.136 — Adds CLAUDE_CODE_ENABLE_FEEDBACK_SURVEY_FOR_OTEL (enterprise OpenTelemetry survey toggle) and settings.autoMode.hard_deny for classifier rules that outright block requests in auto mode. Niche but useful if you’re running Claude Code in a locked-down org and actually want to control what the auto-classifier reaches for.

  • block/goose: v2.0.0-rc-04-27-0 — Release candidate with Dependabot auto-merge when CI passes. The kind of plumbing improvement that saves a human 20 minutes a week and nobody writes home about. 🛠️

AI Dev Tools

  • graphify — An AI coding assistant skill that ingests a folder of code, SQL schemas, R scripts, docs, or even images/videos and turns it into a queryable knowledge graph via GraphRAG + Leiden clustering. Works across Claude Code, Codex, OpenCode, Cursor, and Gemini CLI. The “give me a knowledge graph of this codebase” use case is legitimately useful for onboarding or archaeology dives — just don’t expect it to replace actually reading the code. 📄

  • awslabs/aidlc-workflows — AWS Labs’ AI-Driven Life Cycle (AI-DLC) adaptive workflow steering rules for AI coding agents. Essentially guardrails and routing logic to keep your AI coding agents from going off-script in long-running workflows. If you’re running agentic pipelines at scale and want them to actually finish instead of spiraling, this is worth a look. 🛠️

  • HKUDS/AI-Trader — “100% Fully-Automated Agent-Native Trading.” Three red flags in one sentence: “100%,” “fully automated,” and an AI agent executing real trades. Approach with the skepticism this description actively invites.

Today’s Synthesis

If you’re building a serious AI-assisted coding setup — not just tinkering with a chatbot, but running agents against real codebases in production — three things from today point toward a coherent stack. Start with 9router as your routing layer: it gives you provider failover and meaningful token savings across the agents you’re already paying for. Layer graphify on top of your repositories so your agents aren’t guessing at architecture — they’re querying an actual knowledge graph built from your code, schemas, and docs. Then wrap the whole thing with aidlc-workflows to enforce routing rules and guardrails so long-running agentic tasks don’t spiral into expensive, unproductive loops. None of these alone is a silver bullet, but together they cover the three failure modes that actually kill agentic workflows in practice: runaway costs, shallow context, and unchecked autonomy. The combination is worth prototyping this week.