Glossary

Structure

• Workspace: the top-level scope you work in. Owns members, tasks, agents, and settings. You get a personal workspace on sign-up and can belong to team workspaces.
• Project: an optional grouping on a task (client work, a product area), not a container tasks must live inside. The workspace is the real scope; a task may have a project or none.
• Epic: a feature-sized group of related tasks. Epics belong to a project, so a project-less task can’t have an epic.
• Task: the unit of work: title, description, status, priority, complexity, acceptance criteria, labels. The thing on the board, and the thing you hand to an agent.
• Subtask: a task nested under a parent task.
• Status: where a task sits in the flow: Backlog → Ready → In Progress → In Review → Done. Archived preserves history; nothing is hard-deleted.
• Complexity: a rough size estimate: trivial, small, medium, large, xl. Also drives how long a silent agent’s claim survives before going stale.
• Label: a colored workspace-level tag for grouping and filtering. (MCP tools filter by label names via their tags parameter.)
• Dependency: a “blocked by” link between tasks. A blocked task is never handed to an agent until its blocker completes.

The contract

• Acceptance criteria: the verifiable checklist that defines what done means for a task. The contract both humans and agents build against, and the rubric reviews grade against. See the concept.
• Hand-off: giving a task to an AI agent over MCP, full context included.
• Hand-back: the reverse: the agent delivers with a summary and per-criterion met/not-met claims, and the task moves to In Review for your grade.
• Triage: the holding pen for incoming work (agent-created tasks under the supervised policy). You accept a task into the backlog or decline it; nothing enters the plan without you.

AI

• Brainstorm: the full-page AI conversation that turns an idea into scoped epics and tasks with criteria.
• AI assistant: the same intelligence as a context-aware side panel (Cmd+J) on any screen.
• Skill: a focused mode of the assistant: Brainstorm, Break down, Acceptance Criteria, Parallelize.
• Suggest: the on-demand pass you ask for, in the create dialog or on an existing task. It proposes labels from your workspace vocabulary, an estimate, and up to three acceptance criteria as chips you take one at a time. Nothing is written until you accept.
• Task prep: the one pass Masaro runs unasked, at the hand-off: when an agent claims a task with neither a summary nor criteria, the summary is drafted then and criteria are proposed as suggestions. Backlog grooming does the same sweep on a schedule, if you switch it on. Neither overwrites what you wrote.
• Suggestion: an AI-proposed change you accept or dismiss. Suggestions never apply themselves.
• Project memory: durable, editable markdown notes Masaro keeps about a project (stack, conventions, decisions) so agents start briefed. See the concept.

Agents

• Agent: an external AI worker (Claude Code, Cursor, any MCP client) with its own identity in your workspace. Every action it takes is attributed to that identity.
• MCP: the Model Context Protocol, the open standard agents use to connect to Masaro, pull tasks, report progress, and hand work back. See the MCP hand-off.
• Agent run: the live record of one agent work session on a task: state, current action, progress, branch, PR. What makes an agent look like a teammate on the board.
• Claim: an agent taking a task (In Progress under its identity). Claims are leases kept alive by progress reports; a silent agent’s claim goes stale and the task frees itself.
• Autonomy policy: how much an agent may do without you: supervised (default; triage + review gates) or autonomous. Workspace ceiling and per-agent cap; the stricter wins. See agent autonomy.