A new framework for AI agents, AgentSkillOS, has been released. It automatically builds agents tailored to individual tasks by drawing on a pool of more than 200,000 skills through retrieval and orchestration, and it is drawing attention among developers.
March 2, 2026 · AgentSkillOS
Organizing 200,000 Agent Skills at Ecosystem Scale
An open-source framework that bundles 200,000+ public skills, retrieves the right ones through a hierarchical capability tree , and chains them into structured DAG pipelines — moving agents beyond single-skill calls.
200K+
Public skills bundled and made searchable
30
Creative benchmark tasks across 5 categories
2
Core ideas: capability tree + DAG orchestration
The scalability problem, drawn to scale
Test sets grow by orders of magnitude — in a flat list, relevant skills become statistically invisible.
From a single call to a structured pipeline
Step 1
Capability Tree
Recursive coarse-to-fine retrieval surfaces non-obvious skills
→
Step 2
DAG Orchestration
Builds order, dependencies & data handoffs
→
Result
Creative Output
PDF · PPTX · DOCX · HTML · video · image
Reported results: tree-based retrieval approaches ideal manual selection , and DAG orchestration substantially outperforms flat invocation — winning across the 200, 1,000 and 200,000 scales.
Praise
Surfaces key skills text similarity overlooks
Chains many skills end-to-end on real creative tasks
Human-in-the-loop control and full observability
Open Questions
Requires Claude Code — environment dependency
Scarce real-user data on cost at large scale
Production-use evaluation still awaits accumulation
Continue reading The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.
Already purchased? Sign in ✓ Signed in — this article isn’t included in your current plan.Unlocking the full article…