Sign InOpen Brain
AI EngineerVideoSource Linked

AI-Native Organisations Run on Skills: How to Structure and Scale Them — Imad Touil, QuantumBlack

Shared agent skills can turn team know-how into portable workflows, but unmanaged catalogs create duplication and policy risk. Treat skills as versioned, evaluated dependencies.

AI Engineer · Aug 28, 2026
Open Source Open MarkdownOpen JSON
Source Summary

The proposed agent stack places skills inside workflow blueprints alongside hooks, MCP servers, and subagents. Skills package organizational know-how as **executable, portable units**, while progressive disclosure loads only the instructions needed for a task.

Practical Implication

Move from personal skills to team sharing, then to a **centralized skills registry** with discovery, metadata, dependencies, versions, access control, evaluations, and observability. Assign domain owners to maintain policy-bearing skills and let teams contribute improvements back.

Agent-Ready Context
The proposed agent stack places skills inside workflow blueprints alongside hooks, MCP servers, and subagents. Skills package organizational know-how as **executable, portable units**, while progressive disclosure loads only the instructions needed for a task.

Move from personal skills to team sharing, then to a **centralized skills registry** with discovery, metadata, dependencies, versions, access control, evaluations, and observability. Assign domain owners to maintain policy-bearing skills and let teams contribute improvements back.

Skills are only one workflow component, and there is no settled evaluation method yet. Automatic skill evolution can amplify duplication, conflicts, stale policy, and access problems unless governance exists before the feedback loop is enabled.
Connected Context · Feed7 Judgment

This consolidates prior skill-level practices into an organizational lifecycle: portable know-how progresses from personal use to a governed registry with owners, versions, dependencies, access control, evals, and observability. It confirms progressive disclosure as the loading model, while making automatic skill evolution conditional on governance because evaluation remains unsettled and feedback can propagate stale or conflicting policy.

Skills are new features: Building Skill-Centric Harness — Yogendra Miraje, FactSetDirectly reinforces skills as versioned product contracts and supplies the model-specific evaluation, routing, access, and lifecycle concerns a centralized registry must manage.We Vetted 2000 AI Skills Before They Reached Developers — Lucas Palma, NubankAdds the supply-chain gate required before registry distribution: contributed skills need deterministic and model-assisted review plus tool-layer enforcement.Coding Agents Don't Scale Themselves. Neither Do Your Teams. — Patrick Debois, TesslSupports assigning platform and domain ownership, tying skill curation to risk, auditing, verifier quality, and measured human intervention.How we set up our cloud agent environmentEstablishes a prerequisite and limitation: governed skills still depend on reproducible, discoverable, testable environments, and may add little when underlying workflows are fragile.
Context Map
agentcoding#skills#harness-engineering#context-engineering
Uncertainty
Skills are only one workflow component, and there is no settled evaluation method yet. Automatic skill evolution can amplify duplication, conflicts, stale policy, and access problems unless governance exists before the feedback loop is enabled.