Today's take
Agent engineering evolves from personal configurations to standardized toolchains, with strong correlations between open source projects and application usage. AG-UI protocol and security audit skills demonstrate developers building reusable agent interaction paradigms, validated by Claude Code's explosive growth.
Open source
The open source side shows continuous improvement in agent engineering frameworks, with toolchain maturation driving usability. AG-UI protocol and security audit skills indicate the shift toward standardized agent interaction patterns.
#4 ag-ui-protocol/ag-ui AG-UI protocol integrates agent capabilities into frontend applications, +1,413 stars in 1 day, signaling a new direction of deep agent-user interface integration.
#15 cloudflare/security-audit-skill Cloudflare security audit skill enables multi-phase automated detection, +819 stars in 1 day, reflecting AI engineers' urgent demand for trustworthy AI.
#22 Nutlope/hallmark Anti-AI-slop design skill improves code quality, +729 stars in 1 day, showing the trend of engineering standards migrating to agent workflows.
Models
DeepSeek V4.1 Flash usage surges 25%, Claude Opus 5.5 usage grows 56%. Grok 4.7 usage jumps 169% shows the rapid market penetration of emerging models.
- DeepSeek V4.1 Flash (batch) 8.36T tokens, +25% usage
- Claude Opus 5.5 (batch) 753B tokens, +56% usage
- Grok 4.7 (batch) 102B tokens, +169% usage
Apps
CLI agents still dominate token consumption, with Hermes Agent's 1.97T tokens leading. Claude Code's usage growth validates its ecosystem expansion.
- Hermes Agent 1.97T tokens, maintaining leadership
- Claude Code 1.63T tokens, CLI tool ecosystem expansion
- Strix 451B tokens, security detection scenario breakout
What to watch
Agent engineering transitions from technical experiments to production-grade toolchains, with the synergy between open source protocols and model ecosystems determining the next phase of development.
