Thread
Stories related to "Claude Agent Skills: A First Principles Deep Dive (2025)" across the full archive.
I've been frustrated with shallow answers from AI tools, so I built CoThou, a Personal AI Superagent that reasons from first principles with built-in self-critique, exploring multiple approaches before producing outcomes.<p>Instead of generating quick responses, it breaks user's instructio...
Hey HN,<p>I noticed there are 1,000+ MCP skills on GitHub with zero way for creators to charge for them. So I built Agent37.<p>Selling Claude skills today means asking customers to download your skill, set up Claude Code on their laptops, configure MCP servers, and pray it works. No trial, no update...
I love skills, I have a whole collection of skills I've built myself for problems I experience while using AI agents. but the current method of managing skills in codex/opencode is pretty garbage (copy skills into the correct folder).<p>so i built `sk`<p>One manifest, multiple agents — Def...
AI memory systems often become a black box. When an LLM produces a wrong answer, it’s unclear whether the issue comes from storage, retrieval, or the memory itself.<p>Most systems rely on RAG and vector storage, which makes memory opaque and hard to inspect, especially for temporal or multi-step rea...