The Agent Roster pane
The Agent Roster shows every agent Mastery HQ has ever worked with — one card per agent, whether or not its window is currently open. Open it with open the agent roster (typed or spoken), or from the command palette: Open Agent Roster — profiles & skills.
Reading a card
Section titled “Reading a card”- Header — brand glyph, name, whether a window is currently ● open (only open agents are visible to Hermes routing), total tasks, and last activity.
- The Hermes line — the monospace line under the header is exactly what the planner receives for this agent, e.g.
Alex (claude) — coding ×5, marketing ×2; skills: react, seo. If routing surprises you, this line is why. - Category bars — how much recorded work the agent has done per category (Coding, Marketing, Video Editing, …). Categories are inferred from each task’s wording.
- History — expand to see recent tasks with their outcome: ✓ clean exit, ✗ failed, ○ still unknown.
What you can edit
Section titled “What you can edit”- Skills and Specialties — chips with an add field; remove with ✕. These feed the Hermes roster line directly, so they’re the fastest way to steer routing (“give Bo the
pythonskill and data tasks start landing on Bo”). - Backlog — small per-agent to-dos, for things you want that specific agent to get to.
- Notes — freeform; saved when you click away.
Good to know
Section titled “Good to know”- Profiles are keyed by name: rename an agent window and it binds to a different profile.
- The history is the audit trail behind routing commands to agents — the deterministic fallback picks the open agent with the most recorded work in the task’s category.
Frequently asked questions
Do skills actually change how work is routed?
Yes. When Hermes plans a multi-step goal, each open agent is described to the planner by a roster line built from its top categories and skills — the same line shown on the card. Editing skills changes what the planner sees on the very next 'tell the team to…'.
Where do the numbers come from?
Every task you send an agent is recorded automatically to its profile (category inferred from the prompt, outcome stamped when the run exits). Profiles are per-name — Alex stays Alex across sessions and workspaces — and live as one JSON file per agent in the app's config folder.