2 posts tagged with "agent-skills"

Lately we have been trying to improve one capability in Better Harness: automatic SKILL distillation — recognizing the recurring work paths inside an agent's real sessions, and then deciding which of them are worth capturing as a reusable SKILL. Once we actually started, we found the problem was far harder than "just analyze a session."

For a real software task, an agent's behavior never happens in isolation. It starts from a requirement or a user story, moves through understanding the intent, exploring context, editing code, and verifying the change, and only then produces a contribution someone can review. Looking at the session alone, you can see what the agent did, but it is hard to tell why those actions happened, or which of them actually made it into the final delivery.

So we began treating a single agent delivery as one continuous chain. Today, you only need to run this inside a project directory:

npx @qoder-ai/better-harness inspector

and you get a local, read-only Harness Inspector page that puts the project's agent sessions, file activity, and Git commits into one interactive view.

Writing a capability into a SKILL.md and packaging it as an Agent plugin is not hard. The hard part comes later: once that capability is invoked over and over by different users, in different projects, and on different Agent hosts, how do you guarantee that it is still triggered, executed, and verified correctly? And how do you prove that a change made it better rather than worse?

Drawing on how Better Harness is actually developed, this post walks through five engineering practices - spec-driven behavior, context orchestration, deterministic verification, behavioral evaluation, and the evidence loop - that move a plugin capability from "it works when I run it" to a software asset that is verifiable, maintainable, and safe to evolve.