Today's conclusion
Today's Top 50 shows the center of competition moving from raw model capability to agent behavior, workflow design, and dependable delivery. #1 ayghri/i-have-adhd gained +4,217 stars in 1 day and climbed from #7, while #2 cathrynlavery/diagram-design jumped from #17 to #2 with +2,585 stars.
A second theme is the packaging of agent engineering knowledge into skills, methodologies, and cross-tool infrastructure. The simultaneous rise of XiaoDuoYa/codex-with-chatgpt, obra/superpowers, and multica-ai/andrej-karpathy-skills suggests builders increasingly want an operational layer for planning, execution, memory, and quality control—not merely a stronger model.
Key signals
-
Agent skills are becoming a software distribution unit: #1 ayghri/i-have-adhd addresses a narrow but costly problem—agents burying the answer—and added +4,217 stars in 1 day. #16 multica-ai/andrej-karpathy-skills rose from #47, reinforcing the shift from personal prompting habits to portable behavioral assets.
-
Visual delivery is becoming a core capability: #2 cathrynlavery/diagram-design climbed 15 places, while #3 tt-a1i/archify added +1,436 stars in 1 day. Their emphasis on self-contained, verifiable, exportable diagrams shows agents moving beyond textual advice toward artifacts that can enter real workflows.
-
Planning and execution are separating into composable roles: #11 XiaoDuoYa/codex-with-chatgpt rose from #39 with +755 stars in 1 day by assigning reasoning to ChatGPT and execution to the Codex harness. #10 stablyai/orca and #6 deepseek-ai/deepseek-harness extend the same direction into multi-agent orchestration and plugin-based runtimes.
-
The browser is becoming a general agent interface: #12 browser-use/browser-use gained +725 stars in 1 day, while #25 Player-YN/PawWork_ZhuaZhua climbed from #42. PawWork's selection-first interaction and editable office-file output point to web agents evolving from automated clicking into outcome delivery.
Repositories to watch
-
#1 ayghri/i-have-adhd: This focused skill constrains verbosity, excessive hierarchy, and buried conclusions. Its acceleration shows that agent UX is now an engineering concern whose behavior can be managed as a reusable module.
-
#2 cathrynlavery/diagram-design: It offers 38 editorial diagram types for Claude Code, Codex, and Pi as editable HTML and SVG. Its importance lies in turning visualization into an agent-callable production component rather than a decorative afterthought.
-
#11 XiaoDuoYa/codex-with-chatgpt: The project explicitly separates the planning model from the execution harness and advanced 28 positions in one day. That architecture represents a broader move from monolithic assistants to role-based agent systems.
Outlook
The next question is whether skill catalogs, agent runtimes, and artifact-delivery tools converge on durable interfaces. Today's ranking already points to the next source of differentiation: not just model intelligence, but the ability to organize agent behavior into reliable, reusable products.