Screenshot: Full Missions page showing the mission list with status badges and an open mission detail view with the Plan tab active.
What is a Mission?
A Mission is ADE’s formal execution unit for complex, multi-step work that goes beyond what a single chat session can reliably handle. Where a chat is conversational and exploratory, a Mission is deliberate and audited: it proceeds through defined phases, requires your explicit approval before work begins, tracks every decision and tool call, and produces a full artifact record at the end. Key properties of every Mission:- Phased execution — Planning → Approval Gate → Execution → Validation, with defined entry and exit conditions at each boundary
- Multi-worker orchestration — the mission spawns ephemeral workers that run in parallel, sequentially, or in waves depending on task complexity
- Full audit trail — every tool call, model decision, file change, and worker message is recorded in the mission’s History tab
- Cost and token tracking — real-time per-worker and per-mission totals, with configurable budget caps
- Mission Pack — a deterministic context snapshot regenerated at each phase boundary, enabling pause, resume, and replay
When to use a Mission
Tasks spanning multiple files, teams, or phases — feature implementations, large refactors, test suite builds, or anything requiring coordination across more than one agent.
When to use Chat instead
Exploratory work, single-file edits, quick questions, or anything you expect to resolve in a single back-and-forth turn. Chat is faster for focused tasks.
Missions and Chat are complementary. You can escalate from Chat to a Mission mid-session by saying “start a mission for this” — ADE carries the conversation context over automatically.
Mission Phases
Every Mission passes through four phases in order. The phase boundary is a hard checkpoint — ADE records a Mission Pack snapshot before advancing.Planning Phase
The orchestrator analyzes your mission description and generates a detailed execution plan. This includes a breakdown of sub-tasks, which worker will handle each, estimated token costs, required tool permissions, expected outputs, and the recommended execution strategy (
sequential, parallel, wave, or adaptive).Before planning starts, you can open the Planning Prompt Preview to see — and optionally edit — the exact prompt the orchestrator will receive.Screenshot: Planning phase in progress showing the Plan tab with a generating plan skeleton state.
Approval Gate
The completed plan is presented to you for review. This is a hard stop — the mission cannot advance without your explicit approval.You have three options:
- Approve — proceed with the plan as generated
- Request modifications — provide feedback; the orchestrator revises the plan and presents it again
- Cancel — abort the mission; no workers have run yet
Execution Phase
Workers execute according to the approved plan. The orchestrator coordinates them using the strategy selected during planning:
The Plan tab shows a live activity feed with per-worker status, elapsed time, and token consumption. You can chat with the orchestrator in the Chat tab at any point without interrupting execution.
| Strategy | When used |
|---|---|
sequential | Tasks have strict dependencies; one worker must finish before the next starts |
parallel-lite | A small number of independent tasks run simultaneously |
parallel-first | Tasks are independent; all workers start at the same time and run concurrently in their own lanes |
Validation Phase
After all workers complete, ADE runs the mission’s validation suite: tests, TypeScript type checks, linting, and any custom validators defined in the mission configuration. Validation results are collected as artifacts and surfaced in the Artifacts tab. If validation fails, the mission status transitions to
intervention_required and you are prompted to decide how to proceed.Execution Strategy Selection
Before the Planning phase begins, ADE’s planner analyzes the mission description and selects an execution strategy. The planner considers:- Task dependency graph (does step B require step A’s output?)
- Number of independent work units
- Estimated context size per worker
sequential — one worker at a time
sequential — one worker at a time
Used for tasks with tight dependencies. Worker 2 only starts after Worker 1 has completed and its output has been verified. Slower overall, but safe when outputs feed inputs.
parallel-lite — limited parallelism
parallel-lite — limited parallelism
A small number of independent tasks run simultaneously. The planner sets a
parallelismCap to limit how many workers execute at once.parallel-first — all workers simultaneously
parallel-first — all workers simultaneously
Used for independent tasks. All workers start at the same time and run concurrently in their own lanes. Fastest option when there are no inter-worker dependencies.
Continue Reading
Creating Missions
Learn how to create missions from the Missions tab, a Lane, the CTO chat, or an existing chat session.
Mission Workers
Understand ephemeral workers, context compaction, recovery, and approval gates.
Budget & Permissions
Configure budget caps, tool permissions, persistence, templates, and troubleshoot common issues.