Skip to main content

Documentation Index

Fetch the complete documentation index at: https://ade-app.dev/docs/llms.txt

Use this file to discover all available pages before exploring further.

Missions are gated off in packaged app releases. In a production build the Missions tab shows a “Coming soon” preview and the New Mission flow is disabled. This page documents the live flow as it exists in dev and internal builds, so teams can plan for the shipping version. The Automations tab behaves the same way: templates whose execution is a Mission show a “Coming soon” chip in packaged builds.

​
Where a mission can start

1

Open Missions

Click Missions in the left sidebar.
2

Click New Mission

Press + New Mission in the top-right of the page.
3

Describe the work

Fill the creation dialog:
  • Title — short name shown in the list and history
  • Description — the goal; this is the prompt the planner sees
  • Lane(s) — one or more lanes the mission can work in; workers run inside each lane’s worktree
  • Orchestrator model — the model that plans and coordinates
  • Worker permissions — per-provider permission mode (see below)
  • Budget cap — optional ceiling; mission pauses before the cap is hit, not after
4

Review preflight

ADE checks that selected lanes are available, the configured model is reachable, the permission config is valid, and any required proof-capture prerequisites exist on the host. Blockers surface here, before planning starts.
5

Start

Click Start Mission. The run enters Planning and the detail view opens.

​
Worker permissions

Workers can use different providers (Claude, Codex, OpenCode) and the permission mode applies per provider. The defaults ship with:
  • Claude — full-auto
  • Codex — default (workspace-write sandbox, approval on-request)
  • OpenCode — full-auto
The Codex mode picks its sandbox automatically: full-auto uses danger-full-access, plan uses read-only, and every other preset uses workspace-write. The separate Codex sandbox dropdown that used to live below each provider row has been removed — the editor now derives it from the permission mode. See AI Providers for what each mode allows.

​
The Mission detail view

Opening a running or completed mission surfaces four tabs:
Live control room. Shows the selected strategy, per-worker cards (assigned lane, status, elapsed time, running token count), a phase progress indicator, the activity feed, and any open interventions. Intervention cards highlight in amber — click to respond.

​
Tips for a good first mission

  • Keep the first run narrow — one feature, one lane — before scaling up to multi-lane missions.
  • Leave the budget cap set. It is there to pause the run, not to cancel it.
  • Treat the Approval gate as a checklist: lanes, permissions, and strategy should all look right before you approve the plan.
  • Edit the Planning Prompt Preview instead of re-running with the same description when the first plan misses intent.