A subagent is a reusable Claude Code agent type — a named role with its own instructions, and optionally a pinned model and tool set. Quorum Terminal discovers your defined subagents and lets you preview, audit, create, and manage them in one place.
Where definitions live
A subagent is a Markdown file with YAML frontmatter. Quorum Terminal reads two scopes and lists them together:
| Scope | Location | Shown as |
|---|---|---|
| Project | .claude/agents/*.md in the open workspace | Project |
| User | ~/.claude/agents/*.md in your home folder | User |
When both scopes define the same name, the project file wins. The frontmatter carries name, description, and the optional model and tools fields; everything below the closing --- is the agent's prompt.
The Subagents panel
The panel lists each definition as a card with its name, scope, description, and model / tools chips. Every card offers three actions:
- Preview — shows the full definition with a best-practice audit, plus buttons to reveal it in Finder or open it in the editor.
- Audit — the same preview, scrolled to the lint findings.
- Delete — moves the file to the system Trash (recoverable).
The + New action opens the Subagent Studio, a conversational builder that co-authors a best-practice definition with AI and lints it as you go. If the Studio cannot open, Quorum Terminal falls back to a blank file: you name the agent and it writes a starter template, then opens it in the editor. Names must start with a letter or digit.
Studio draft generation runs Claude or Codex, so it requires the integrations setting to be enabled.
Define the role in the agent file and the project's AGENTS.md, so every Claude Code session that dispatches the subagent starts oriented.
Something off on this page? Tell us.