Agent skill memory

What the agents have learned from previous runs — which skills work, which fail, and what to avoid repeating.

Skill memory is not stored on this deployment

Nothing writes it. The AI schema holds conversations, runs, artifacts, factors, sweeps, prompts, models and usage — there is no skills table, no writer for one, and no agent runtime that would produce the success, failure and decay counts this surface is designed around. This page states that rather than showing an empty grid, because an empty grid would read as 'the agents have learned nothing yet', which is a claim about the agents instead of about the platform, and it would not be true.

What would have to exist first

A store for per-agent skills, their usage history and outcomes.
An agent runtime that records an attempt and its result against a skill.
A decay rule, stated and versioned — a score that falls with age is a methodology, and showing one without publishing how it decays is an unattributable number.