Agent Skills: Procedures, Knowledge and Trust

On October 22, 2026 at 11:00 MSK, Alexander Polomodov and Maxim Smirnov will continue their episode 19 discussion of Loop Engineering. We will examine how Agent Skills give an agent a repeatable procedure, what controls skill selection, and why loading SKILL.md does not establish output quality.

Using morning CI triage as a running example, we will distinguish project instructions, skills, MCP tools, and independent checks. We will trace the catalogs, installers, audits, and evaluations surrounding the format before turning to Open Knowledge Format: knowledge files, links, provenance, freshness, and attested computation.

This discussion synthesizes specifications, documentation, and examples as of October 19, 2026. We will compare Agent Skills, OKF v0.2, and the Knowledge-as-Skill proposal, separate declared properties from demonstrations, and ask who verifies an execution receipt and takes responsibility for the operational workflow.