Agent Skills and Open Knowledge Format (OKF): Skills, Knowledge, and Trust (#AI4SDLC)

An agent can write code and call tools. But how do you give it a team’s working procedure? Where do you store knowledge about a system that changes independently of that procedure? And what confirms that the agent completed the task correctly?

In an hour, we will discuss this in a livestream of the Research Insights Made Simple podcast with Maxim Smirnov. This summer, we discussed Loop Engineering: how to design loops that launch agents, check results, and send work back for revision. Now we will continue the conversation through Agent Skills and Open Knowledge Format (OKF).

Maxim is an IT architect and author of the Telegram channel “Arkhitektura IT-resheniy” https://t.me/it_arch. Previously, he headed the IT architecture department at Beeline and served as chief information systems architect at the Bank of Russia; he is a speaker, author, and instructor of IT architecture courses.

In the new episode, we will examine what makes up a skill: SKILL.md, a description, instructions, supporting files, and scripts. We will look at how an agent chooses the appropriate skill, when it loads the details, and why successfully loading a file does not yet mean successfully completing a task.

On the agenda: — How project instructions in AGENTS.md, skills, MCP tools, and event handlers differ. — How to check not only a skill’s result, but also whether it is invoked in the right situation. — What has emerged around Agent Skills: directories, installation, automated checks, and quality evaluation. — How OKF connects Markdown files into a knowledge graph and describes the provenance and freshness of information. — Why a timestamp or an attestation in metadata does not, by itself, guarantee truth. — What a computation execution receipt means, who verifies it, and where the team remains responsible.

Our running example is a morning investigation of a failed CI run: which instructions the agent needs, where it learns the project’s rules, and how we check the proposed fix. We will compare Agent Skills, OKF v0.2, and the Knowledge-as-Skill proposal.

We are drawing on specifications, documentation, and examples as of late October 2026. We will distinguish the properties of the formats from the capabilities of specific implementations.

Bring your own cases: which procedures are you already handing over to agents, and what knowledge do they regularly lose?

#AI #Agents #Engineering #AI4SDLC #Software #Knowledge #Databases