🔥 Hot Repo: 50K Stars — AI Agent That Patches Itself

NousResearch's Hermes Agent crossed 49,600 GitHub stars this week — picking up 7,674 in the past 24 hours alone — after releasing v0.8.0, a 209-pull-request update that includes one genuinely unusual capability: the agent diagnosed and patched five failure modes in its own tool-calling behavior through automated behavioral benchmarking. The feature, called "Self-Optimized GPT/Codex Tool-Use Guidance," ran Hermes against its own behavioral benchmark suite targeting GPT-4o and Codex models, identified five distinct failure categories — including execution discipline gaps and structured reasoning prefill errors — then applied targeted fixes automatically. NousResearch says the result is "dramatically improved reliability on OpenAI models." A companion repository, hermes-agent-self-evolutionhttps://github.com/NousResearch/hermes-agent-self-evolution, extends this to skill and prompt optimization using DSPy + GEPA, which reads full execution traces to understand why tasks fail rather than just that they fail. Why 7,600 Stars in 24 Hours The timing connects directly to the April 4 Anthropic subscription blockade. Anthropic cut API access for third-party agent harnesses — including OpenClaw — forcing users onto significantly higher per-token billing or into finding alternatives. Hermes ships a direct migration path built into the CLI: hermes claw migrate. That single command imports OpenClaw users' persona files SOUL.md, conversation memories, user-created skills, command allowlists, platform configs, and API keys. Hermes sat below 30,000 stars before the crackdown; it's at 49,600 today. What's in v0.8.0 The release merged 209 pull requests and resolved 82 issues. Major additions: Background process auto-notifications. Long-running tasks — model training, CI pipelines, deployments — now notify Hermes on completion without polling. The agent keeps working on parallel threads and picks up results when they land. Live model switching. The /model command now works mid