Today's take
Agent engineering frameworks are maturing rapidly, with multiple open-source projects entering the Top 50. Hermes Agent maintains highest token volume, while GPT-5.6 Luna surges 152% in usage. The convergence of these trends indicates Agent systems are evolving from experimental tools to engineered solutions.
Open source
The open-source ecosystem shows clear engineering focus:
- Agent engineering frameworks emerge : morluto/rea (#1) enables capability reuse through reverse engineering, mattpocock/skills (#15 new) provides transferable agent skills
- Context management demand grows : thedotmack/claude-mem (#3) improves session efficiency through context compression
- Testing verification systems mature : ifixai-ai/iFixAi (#2) offers automated audit solutions, accelerating Agent trustworthiness
#1 morluto/rea https://github.com/morluto/rea +3,889 stars in 1 day. Signal: Reverse engineering toolchain enables Agent capability modularization
#2 ifixai-ai/iFixAi https://github.com/ifixai-ai/iFixAi +2,077 stars in 1 day. Signal: Automated audit solutions accelerate Agent ecosystem trustworthiness
#3 thedotmack/claude-mem https://github.com/thedotmack/claude-mem +1,735 stars in 1 day. Signal: Context management technology solves long-term session bottlenecks
Models
Top models maintain stable rankings but with significant fluctuations:
- DeepSeek V4.1 Flash 1017 leads with 7.74T tokens, but its batch version usage drops 18%
- GPT-5.6 Luna surges 152% to 1.12T tokens, showing rapid market adoption of new models
- Kimi K3.1 Pro grows 60% to 526B tokens, demonstrating持续 growth of Tongyi Qianwen ecosystem
Apps
App ecosystem shows clear分化:
- Hermes Agent maintains top position with 1.84T tokens, showing its leadership as a general-purpose Agent framework
- Portkey AI grows 200% to 395B tokens, becoming fastest-growing programming app
- OpenClaw grows 183% to 708B tokens, showing rapid adoption of personal Agent tools
What to watch
Agent engineering system is forming a closed loop: open-source provides toolchains, models support capabilities, apps validate scenarios. This synergy will drive Agents from experimental tools to production-grade systems.