Missions
Goal-driven orchestration with acceptance criteria and verification workflows.
Missions
Missions are a beta feature. The API surface and behavior may change in future releases.
Missions are a higher-level abstraction over tasks. Instead of manually decomposing work into individual tasks, routing them to agents, and verifying results yourself, you describe a goal and Kavi orchestrates the full lifecycle: planning, execution, acceptance verification, and landing.
A mission wraps your prompt with structured metadata that Kavi uses to drive the entire workflow:
- A goal describing the desired outcome
- An execution mode controlling how much autonomy Kavi has
- Acceptance criteria auto-generated from prompt analysis
- Brain entries providing contextual knowledge to agents
- Checkpoints tracking progress through each phase
Missions automatically generate acceptance criteria by analyzing your prompt for keywords related to frontend, backend, testing, and documentation. When all tasks complete, Kavi runs verification checks -- executing commands, confirming file existence, and flagging manual review items -- before allowing the mission to land.
Mission Modes
Missions support three execution modes that control how much autonomy Kavi has during planning and execution:
- guided_autopilot -- Kavi plans and executes autonomously, applying follow-up recommendations without operator intervention. Best for well-defined tasks where you trust the agents to proceed. Read more
- inspect -- Kavi plans and executes but pauses at key decision points for operator review. Recommendations require manual application. Best for complex tasks where you want oversight. Read more
- manual -- Full operator control over task approval and routing. Best for sensitive or exploratory work. Read more
Acceptance and Verification
Each mission carries an acceptance pack containing auto-generated criteria and verification checks. Kavi synthesizes checks from your repository context -- detecting test suites, build scripts, and documentation -- and runs them in an integration workspace before allowing the mission to land.
Read more about acceptance and verification
Lifecycle
Missions progress through a series of statuses from planning through completed, with checkpoints recorded at each transition. Kavi automatically synchronizes mission status based on task state and acceptance results.
Read more about the mission lifecycle
Mission Interface
The following fields make up a mission record:
| Field | Type | Description |
|---|---|---|
id | string | Unique mission identifier (prefixed with mission-) |
title | string | Auto-summarized from the prompt (max 96 characters) |
prompt | string | The original operator prompt |
goal | string | null | Optional explicit goal description |
mode | MissionMode | Execution mode: guided_autopilot, inspect, or manual |
status | MissionStatus | Current lifecycle status |
summary | string | Human-readable summary, updated as the mission progresses |
planningTaskId | string | null | ID of the planning task (if the mission uses a planning pass) |
planId | string | null | ID of the materialized execution plan |
rootTaskId | string | null | ID of the root task for direct-execution missions |
activeTaskIds | string[] | IDs of currently pending, running, or blocked tasks |
autopilotEnabled | boolean | Whether autopilot is active (true for guided_autopilot mode) |
acceptance | AcceptancePack | Acceptance criteria and verification checks |
checkpoints | MissionCheckpoint[] | Ordered list of progress checkpoints |
brainEntryIds | string[] | IDs of brain entries providing context to this mission |
createdAt | string | ISO timestamp of creation |
updatedAt | string | ISO timestamp of last update |
landedAt | string | null | ISO timestamp when the mission was landed (null if not yet landed) |