Skills, Memory, and Activity
Skills, Memory, and Activity help the operator understand reusable capability, local context, and recent work. They are related cockpit surfaces, but their signals have different meanings.
Skills
Section titled “Skills”AI OS can inventory installed and observed skills from supported Claude Code and Codex locations. A skill row can show recent uses, total observed uses, last use, source coverage, and a deterministic usage signal.
The usage signal is not a quality score. It cannot prove that a skill succeeded, was useful, or saved time. An installed skill with no observed use correctly appears as having no signal.
Saved-time estimates
Section titled “Saved-time estimates”Time saved is a separate, operator-configured value. Every skill starts at
0 minutes saved per run. The hourly rate converts configured time into an
estimated dollar value; it does not create saved time by itself.
estimated value = (minutes saved / 60) x hourly rateThe Home summary and Skills page share the same browser-local estimates. A reload can also hydrate a reviewed local override when one is configured. Editing an estimate does not change the observed usage signal.
Memory and Knowledge Explorer
Section titled “Memory and Knowledge Explorer”Memory presents bounded local context and a Knowledge Explorer for supported note sources. Search, topics, related-note navigation, and graph views operate on projected summaries rather than unrestricted file contents.
The Memory note graph is distinct from the project-code Knowledge Graph. The former explores local notes; the latter explores Graphify project maps and can ground Hermes in the selected project.
Activity
Section titled “Activity”Activity is a searchable, filterable view of projected local work events. It can connect runs, tools, skills, and outputs when those fields exist, but it does not turn every local log line into browser data.
Expected empty states
Section titled “Expected empty states”- No skill signal can mean an installed skill has not been observed.
0mcan mean no saved-time estimate has been configured.- An empty recommendation area is expected because the current aggregator does not produce skill recommendations.
- Empty Memory or Activity can mean the full local aggregation path has not run yet.
Use bun run scheduler:agents:run when you want to refresh supported host and
agent data, then read the visible source and freshness labels before drawing a
conclusion.