artifacts that plug into a system
Source: Dev.to
Overview
Anthropic recently published Agent Skills as an open standard.
A good thing about skills is that they are external to the system (LLM/AI). This means they can not only be integrated with the system, but also be improved independently of the system.
A skill for performing a task y can be iterated upon, using its historical performance as feedback.
The same principle should be applied to frameworks for agents and workflow orchestration: the definition should have loose coupling with the library handling the execution.
Example: a tightly‑coupled library
Imagine a library called flow. Typically such a library would expect its user to construct a workflow or agent as:
const flow = new Flow();
flow.addAgent();
flow.addTool();
flow.addPrompt();
flow.buildSystem();
flow.run();Example: an artifact‑centric approach
If the library treated the workflow as an independent artifact, it could be used more simply:
const flow = new Flow();
flow.run();While the first option is more programmatic, the second is easier and somewhat cleaner to iterate on the workflow.
Evolving the workflow
The artifact‑centric approach also allows the code to evolve cleanly:
const flow = new Flow();
flow.run(flowV2);Or we could maintain the concept of “latest” with the ability to roll back whenever required.