shipped AgentForge v1 today 🚀
been obsessed with one question lately . what actually lives under Claude Code / Codex style agents?
so i built the whole harness myself, from scratch, in raw Python 🐍
here's what's inside 👇
→ agent loop
→ tool calling
→ MCP support
→ skills system
→ human approvals
→ context compaction
→ session persistence
→ checkpoints
→ terminal UI
not another AI wrapper.
the whole point is to expose the harness layer the part that makes agents inspectable, debuggable, and actually safe to run in prod 🔬
most people use these systems. almost nobody knows how they're wired.
AgentForge is the learning lab for that.
install it rn 👇
pip install agentforge-harness
agentforge init
agentforge
repo → https://t.co/6G3eEce7Gj
full architecture breakdown + lessons from building it dropping soon on the blog 📝
follow @ByteMohit so you don't miss it 👀