wtf is awesome-discoveries?

This repository is a curated reading list of projects, tools, and infrastructure pieces aimed at people building AI agents. There is no code to install. It is a single Markdown file that lists other GitHub repositories, each with a short one-line summary and a star count from when the list was last updated. The Portuguese description on the project page calls it a "living curatorship," updated automatically by something the author calls a Growth Agent. The page credits @agentxagi on X for the indexing. The entries cluster around a few themes. One is memory and persistent context for coding agents: projects such as memvid, agentmemory, MemOS, engram, ByteRover CLI, and claude-mem all promise to store what an agent did in one session and feed the right pieces back in the next one. Another theme is harnesses and runtimes that sit around a model: ECC, OpenViking, osaurus (a native macOS runtime in Swift), and ARGO (a local Manus-style platform that runs on your laptop) fit here. A third group covers orchestration and multi-agent work, including Agent Orchestrator, MassGen, contrabass, Ruflo, and Warp's Oz orchestration. A fourth group covers evaluation and testing for agent behavior: agent-skills-eval, skillgrade, eval-view, agent-belt, and Raindrop Workshop, which streams traces of an agent's tool calls in real time. The list also points to broader tooling such as Context7 for fresh documentation delivered through MCP, Chrome DevTools MCP for web debugging by agents, mcp2cli for turning MCP servers into command-line tools, and Lapdog from Datadog for tracing agent reasoning locally. A few items are not GitHub projects at all but research papers or X posts, such as TraceFix, which uses TLA+ counterexamples to repair multi-agent coordination protocols, and a thread describing a retrieval routing pattern that splits queries between vector RAG, GraphRAG, and PageIndex. There is no install, no license note in the README excerpt, and no setup. The repository is meant to be read.