Demonstration-to-skill capture

A person takes manual control of an agent’s browser or desktop, demonstrates a task and stops the recording. The system converts the actions into an invocable procedure. The durable artifact is a skill; the recording is its authoring input.

A narration pass can explain why an action was taken or which parts should generalise. This can capture tacit knowledge in browser workflows that are difficult to describe in a written procedure. A successful demonstration does not prove that the resulting skill survives a changed interface.

Four workshops show or describe recording a human task on an agent’s own computer, then invoking the resulting skill later. Voice narration can clarify why an action was taken. All four accounts concern the same vendor and product: they count as one independent source. Recordings are third-party re-uploads. Individual workshop dates within the event are not established.

Assay’s role skills are authored as versioned text and reviewed through pull requests. The methodology does not describe turning a recorded browser task into a skill. Its current work is primarily forge and CLI based, so this is a tracked gap rather than a promised feature.

• A second independent implementation beyond this vendor.
• Measured failure and recovery when the interface changes.
• Whether narration improves repeatability over action recording alone.