Glossary

The words below are the ones the app uses. Where the app says one thing and the code or the MCP tools another, both are given.

Access mode. A runner profile’s setting for how much a run may do without asking: Auto (the default), Ask or Full. It sets the CLI’s own permission mode and sandbox. See Access modes.

Agent. A persistent AI teammate in a project, with a chat, a mission, instructions and an autonomy level. Lea is the lead agent; role agents can look after areas. Workers are temporary agents.

API key or an approved login. How the Claude Code CLI on your runner signs in to Anthropic. An API key always works. Any other Claude Code sign-in works with Minga only where Anthropic’s terms allow that use; check them for your login. Codex uses your Codex login.

Area. A part of a project, such as Engineering or Release, that groups milestones, KPIs and optionally an agent. Set up in Project settings › Areas.

Artifact. The code’s and the MCP tools’ word for an output.

Asked by. Planned. Whose request a run serves, and so whose machine and logins run it.

Assignee. Who does an item: a new worker, or a person. See Owner and assignee.

Auto-review. Codex’s own reviewer, a model that approves or denies Codex’s requests to leave its sandbox, network requests and MCP tool prompts. It is on in the Auto access mode. See Codex auto-review.

Autonomy. An agent’s level, which decides how far it goes in turns that changes started: Suggest in chat, Ask first, or Act, then tell me. See Autonomy levels.

Brief. The project’s short statement of where it is going: Vision, For whom, Principles, Not doing, How to build and test. Every agent and worker reads it. See The brief.

Capture. Putting an idea, bug, feature or question into triage without placing it in the plan: the Capture box in Needs you, or the capture tool over MCP.

Card. Something a person decides, in Needs you and in the chat: a plan, a brief edit, a question, a review, a permission prompt. Agents open cards; the code calls them proposals (P-7).

Chain. The count of agents waking each other without a person in between. The fourth in a row is held as a card.

Check. A condition for an item to be done (“Done when”): a command that passes, an output, a rubric for a validator, a person’s approval.

Checkout. A local clone of a repo. The runner serves the one you registered with minga repo add, or the one it found in your Claude Code and Codex sessions.

CLI. A command-line tool. Here: the claude (Claude Code) and codex (Codex) tools that do the work on your runners, and the minga tool that is the runner.

Decision. A decided card, or a decision comment on the record it shaped. “Decisions” on the Overview lists both. See Decisions.

Device code. The short code minga runner login prints. You check that it matches in the browser and approve, which pairs the runner with your workspace. It expires after 10 minutes.

Goal. An outcome you measure, with a target on a KPI and often a deadline. See Goals and KPIs.

Health. How a goal stands: on track, at risk or off track, from the latest update posted on it, with who set it and when.

Item. Anything in the plan that can be finished: a milestone, feature, task, bug, research item, chore, idea or question. Items have refs such as DIC-133.

KPI. A number tracked over time (K-3), with samples per hour, day or week.

Later · Next · Now. Where a milestone is: an idea for later, being planned (nothing under it starts), or being built. Moving a milestone to Now is the approval of its spec.

Lea. The default name of a project’s lead agent.

Lead. The one agent per project that owns the plan, supervises every item and talks with you in Chat.

Lease. A runner’s hold on a run while it works on it. If the runner disappears, the lease runs out and the run ends.

MCP. The Model Context Protocol: the way your own Claude Code or Codex session calls Minga’s tools (read the plan, capture, comment, ask Lea). See Connect Claude Code and Codex over MCP.

Milestone. A shipment with a due date, at the top of the plan. In the code and the MCP tools it is a release item (bets and workstreams also show as milestones). The app always says Milestone.

Needs you. The tab, and the one count, of everything that waits on you, grouped by verb: Decide, Answer, Review, Allow, Do, Sort. See Needs you.

Output. Something work produced, attached to its item: a screenshot, a report, a design, a video, a pull request. See Reviews and outputs.

Owner. Who answers for an item and gets its review: an agent or a person.

Pause. Stops an agent from taking turns until someone resumes it in Project settings (Lea’s page, or Agents for other agents). Messages to it wait.

Pick. Your choice of one output in a review (Pick this, then Approve with its name). The pick becomes final; the others are set aside.

Plan. The tab with the project’s work (List, Canvas, Outputs), and the work itself: goals, milestones and items. Also the MCP and agent tool that changes it.

Preset. How a project’s workers work: in the repo (Default) or without one (No repo: research, data), with an optional preferred CLI, model and effort.

Profile. One CLI setup on a runner: Claude Code or Codex, with its home directory, login and access mode.

Project. One product or body of work, with its brief, plan, lead and chat. Projects live in a workspace.

Ref. A record’s short name, which links to it: DIC-133 an item, P-7 a card, W-12 a worker, V-3 a validator, K-3 a KPI.

Release. The item type the MCP tools use for a milestone. The app always says Milestone.

Review stage. One step an item passes after a done claim: the owner agent’s review, a validator’s check, or a person’s review.

Rubric. Criteria for an item that a validator judges, rather than a command that passes or fails.

Run. One execution: an agent’s turn or a worker’s or validator’s work on an item, on a runner or in the cloud. It always ends with a stated reason.

Runner. The minga service on a machine you own. It connects out to Minga and runs work in your CLIs. See Workers and runners.

Saved line. The line under each of Lea’s replies, built from what the reply did: Saved with the records it changed, Proposed with the cards it opened, Not saved with any write that failed.

Spec. An item’s description. A milestone’s spec has three parts: Why, What and Not in this release.

Transcript. What a run read and did: commands, output, file excerpts, diffs. Visible to workspace members who can open the run.

Triage. Where captured items wait for a place in the plan, status To sort. In the app it is the Sort group of Needs you. See Triage.

Turn. One round of work by a persistent agent: something wakes it, it reads what changed, acts, and ends.

Undo. Takes back a change Lea made on her own at the Act, then tell me level. It shows on her message. In a review, Undo takes back a pick or a Not this.

Validator. A separate run (V-3) that checks a worker’s result against the item’s rubric and commands, on the other model family by default.

Worker. A temporary agent (W-12) that does one item, on a runner, in a worktree of your repo.

Workspace. Your team’s space in Minga: members, runners, model keys and projects.

Worktree. A separate git working directory that the runner adds from your clone for one item, on a minga/… branch, under ~/.local/share/minga/worktrees.