Glossary

Every term the docs use, in one place. Each entry links to the page that owns it.

Activity state

The real-time status of an agent: thinking (actively working) or idle (waiting for input, stopped, or hit a permission prompt). Displayed as visual indicators on task cards. See Activity detection.

Agent Monitor

A machine-wide surface listing every live and recently finished agent session across all registered projects, including Command Terminal sessions. Opened from the title-bar activity icon or Ctrl+Shift+M / Cmd+Shift+M; offers card, table, and list layouts, live output peeks, and opens task detail windows in place. See Agent Monitor.

Agent override

A per-column or per-task setting that runs a session with a specific agent CLI instead of the project default. Any of the fifteen supported agents can be chosen. Set per-column to map agents to workflow stages (e.g., Claude for Planning, Codex for Executing), or per-task via the Agent Override card of the New Task / Edit dialog’s “How this task runs” choice to lock a specific task to a specific agent for its entire lifetime - per-task wins over per-column. Combined with Cross-Agent Handoff (below), this enables multi-agent pipelines on a single board. See Adding and editing columns and Per-task overrides.

Auto-naming

A short title made from a prompt: the Name from prompt button in the New Task dialog, the rename suggestion for a task still carrying a placeholder title, and a Command Terminal’s name from its first prompt. Since v0.46.0 it runs on the Automation agent with that agent’s model and effort, and with no Automation agent chosen it is off. Each call is recorded as a background call.

Automation

What a column does when a task enters it, when the column’s agent finishes a turn, or when the task leaves it. Each column owns one ordered list, split into On task enter, On agent turn end, and On task exit, with five types: Send message to agent, Move to column (turn end only), Run script, Call webhook, and Notify me. Any row can carry a Condition, checked as an Agent evaluation by the Automation agent or as a Text match against the agent’s last line, and a Frequency that caps its runs. A message row is free prose delivered as an ordinary user message, with its own delivery choice (Run immediately or Wait for current turn); every default-board column ships with no automations. Edited in the Column Manager, stored under the column in kangentic.json, and shared by every board profile. See Automations and What the agent is told.

Automation agent

The agent Kangentic runs on its own for small jobs: answering automation Conditions that are Agent evaluations, deciding when a workflow task’s work in a column is done, and naming tasks and Command Terminals. Chosen with its model and effort in Settings > Workflows. Nothing picks one for you, and with none chosen those jobs do not run. See Workflows.

Auto-spawn

A per-column toggle (Start an agent here) that controls whether moving a task to that column automatically starts an agent session. Enabled by default on active columns. Turning it off also switches off the column’s Send message to agent automations.

Backlog

A staging area for tasks that aren’t yet on the board. The backlog is a separate view from the Kanban board where you can plan, prioritize, and organize work before promoting items to board columns. See Backlog and imports.

Base branch

The git branch that a worktree branches from. Defaults to main (configurable via git.defaultBaseBranch). Can be overridden per-task.

Board configuration

The shareable kangentic.json file that defines columns and their automations for a project. See Board configuration.

Board profile

A named alternate ladder of per-column agent, model, effort, and session settings that a task can ride instead of the columns’ own values. Authored in the Column Manager, shared through kangentic.json, and manageable by agents over MCP. Automations are not part of a profile. See Board profiles.

Changes panel

A built-in git diff viewer in the task detail window (and the Command Terminal) that shows what an agent changed. A resizable file tree with status icons, viewed marks, and sort/tree/flat controls sits beside a split or inline diff viewer, with a branch header, a working/staged/branch scope control, a commit-history browser with a pinned Uncommitted changes row, per-file history, and a blame gutter toggle. It refreshes live as files change. See Reviewing changes.

Column

A vertical lane on the Kanban board representing a workflow stage. Each column can have its own permission mode, automations, and auto-spawn setting. Also called a swimlane internally.

Column Manager

The dialog that edits every column from one place: click a column header or Add column on the board toolbar. Settings on the left, the column’s automations on the right, with a board profile switcher at the head of the column rail. See Column Manager.

Context bar

The information strip shown with the terminal panel displaying shell name, agent version, elapsed time, model, session cost, completed tool-call count (click for a per-tool breakdown), token usage, context window fraction, and (for Claude) 5-hour session and weekly plan-usage rate-limit bars. Most pills can be toggled individually under contextBar.* settings; the model and effort pills always show when reported and double as the per-task model/effort picker.

Cross-agent handoff

The mechanism by which Kangentic points the next agent at the prior agent’s native session history file when a task moves into a column with the handoff_context option enabled. Lets a Plan task written by Claude continue as an Execute task in Codex with full conversation context, for example. Whether context carries depends on the outgoing agent: Claude Code, Codex, Gemini, Grok Build, Antigravity, OpenCode, Aider, Qwen Code, Kimi Code, and Factory Droid each leave a readable history behind, while Cursor, Copilot, Goose, Warp, and Ollama leave none, so the next session starts with a pointer to git log instead. Any agent can be on the receiving end. See Agent orchestration.

Diff scope

The set of changes the Changes panel shows: working (uncommitted working-tree edits), staged (the git index), or branch (everything this branch accumulated versus its base, the full pull request). A freshly opened panel starts on the diffDefaultScope setting.

Done

A special column with the done swimlane role. Moving a task here suspends its session (preserving context for later resume) and archives the task.

Embedded browser

A side-pane in the task detail window that hosts a Chromium webview, lets you draw annotations or pick DOM elements, and submits the rendered frame plus context to the active agent as a multi-modal prompt. See Embedded browser.

Event

A structured log entry representing something that happened during an agent session (tool use, idle state, prompt submission, etc.). Events are displayed in the Activity tab. See Event types.

Ghost column

A column that has been removed from a canonical kangentic.json but still holds tasks. Rather than deleting it and orphaning those tasks, Kangentic hides it from the board while preserving it in the database; once you move its tasks elsewhere and it is empty, it is cleaned up automatically on the next reconciliation. See Board configuration.

Hotkeys

The Settings tab that lists the keyboard shortcuts you can rebind, drawn from the central keybinding registry (fixed and developer-only bindings stay off the list), with conflict detection and terminal-collision warnings. Overrides are stored per-machine. See Hotkeys.

Hook

An agent-specific event callback that Kangentic injects into a session for activity detection (Claude Code’s PreToolUse/PostToolUse/Stop and their kin, the Claude-compatible hook sets Grok Build and Antigravity CLI accept, Gemini and Qwen’s BeforeAgent/AfterAgent/BeforeModel/AfterModel/BeforeToolSelection, Copilot’s preToolUse/postToolUse/agentStop/preCompact, and OpenCode’s plugin events). Hooks fire on events like tool use, agent stop, and permission requests. Hook files are reference-counted so concurrent sessions don’t clobber each other. See Per-agent detection strategy.

Isolated session

A separate, context-isolated agent session a task runs on while it sits in a column whose Session Target is isolated. It does not inherit the task’s main conversation, so it suits work that should stay independent of it (a code review, for example). Keyed by the column, so re-entering that column resumes its own conversation while leaving it resumes the main session. An isolated session’s terminal tab carries an Isolated badge; a main-session tab carries none. Set per column via Session Target (below). See Isolated sessions.

Knowledge Graph

A map of a project’s indexed agent conversations, opened from the title-bar brain icon or Ctrl+Shift+A / Cmd+Shift+A, with points placed by meaning and grouped into named regions. Its Ask box answers questions from the related tasks, a table of every task’s facts, and optionally the project’s code, through the agent you choose in Settings > Knowledge Graph. That tab also holds the conversation index itself. See Knowledge Graph.

Model override

A per-column or per-task setting that pins sessions to a model identifier (e.g., opus, gpt-5-codex). A floating alias such as opus follows each new release of that model; a full id stays pinned. A running session that crosses into a column with a different one is suspended and resumed on the new model with its conversation intact. See Configuration.

Permission mode

Controls how much autonomy an agent has. Options: default (uses project settings), plan (read-only), acceptEdits (auto-approve file edits), dontAsk (deny unless allowed), bypassPermissions (full autonomy), auto (classifier-driven auto-approval, Claude, Copilot, and Grok only; Codex labels its acceptEdits preset “Auto (Preset)”, which is a different thing). Not every agent supports every mode; see Permission modes for the per-agent matrix.

Plan exit target

The column a task automatically moves to when an agent in plan mode calls ExitPlanMode. Enables the Plan > Execute workflow pattern.

Plan mode

A permission mode where agents can only use read-only tools. The agent analyzes the codebase and creates a plan without making changes. See Plan mode columns.

Queue

When the maximum concurrent session limit is reached, new sessions enter a FIFO queue. Queued tasks show a “Queued” label and start automatically when a slot opens. See Session queue.

Quick Find

The unified search overlay opened with Ctrl+Shift+F / Cmd+Shift+F (or Ctrl+F / Cmd+F when not on the board view). Searches across tasks, backlog items, session events, projects, and past conversations by keyword - current project or all projects - and its last row hands a question to the Knowledge Graph. See Quick Find.

Session

An agent CLI process managed by Kangentic, running one of the fifteen supported agents. Sessions can be running, queued, suspended, exited, or orphaned. See Session persistence.

Session spawn strategy

A per-column setting (session_spawn_strategy) controlling what happens to the column’s session track on entry: create_or_resume resumes an existing session or spawns one; always_spawn_new always starts a fresh pass, retiring the prior session. Isolated columns default to always_spawn_new. See Column-level settings.

Session target

A per-column setting (session_target) choosing which session track a task runs on in that column: main (the task’s shared main conversation) or isolated (a separate, context-isolated session of its own; see Isolated Session above). See Column-level settings.

Spawn prompt

The first prompt a freshly spawned agent receives: the task’s title and description wrapped in a <task> XML envelope, followed by any attachment file paths, rendered from the default template {{task_xml}}{{workflow_xml}}{{attachments}}. Outside a workflow {{workflow_xml}} is empty, so nothing else is sent - no hidden preamble, and the column’s name never reaches the agent. A workflow task also gets a <workflow> block naming its workflow, its route, and the tasks it builds on and the ones waiting on it. See What the agent is told.

Swimlane

Internal name for a board column. Two swimlanes have special roles: todo (first column, no agents run by default) and done (sessions are suspended and tasks archived). All other swimlanes are fully customizable.

Swimlane role

A special designation for columns: todo or done. These roles control built-in behaviors like session suspension and task archiving.

Template variable

A placeholder in an automation field or a per-task prompt that gets replaced with task-specific values at runtime. Examples: {{title}}, {{branchName}}, {{toColumn}}. See Template variables.

Usage Stats

The full-page usage view opened from the title-bar chart icon or Ctrl+Shift+U / Cmd+Shift+U. Shows cost, tokens, sessions, tool calls, line churn, burn rate, and subagent tokens for the current project or all projects, broken down by model, agent, reasoning effort, and subagent type. See Usage Stats.

Workflow

A set of linked tasks that runs through the board on its own. Each task has a route of columns, starts when the tasks it is linked to (its blockers) reach Done, and moves to the next column on its route when the Automation agent judges the work there done. Planned with the Workflow agent from a brief, and shown in the Workflows view, a canvas beside Board and Backlog (Ctrl+Shift+L / Cmd+Shift+L). See Workflows.

Workflow agent

The agent that plans a workflow from your brief and keeps planning while it runs, in the Workflows view’s agent pane, through the workflow MCP tools. Chosen in Settings > Workflows, with no fallback to the project’s default agent. Not the same choice as the Automation agent. See Workflows.

Worktree

An isolated git working directory created for each task. Allows multiple agents to work on the same repository simultaneously without branch conflicts. See Worktrees and pull requests.