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 ADE’s structured execution unit for complex, multi-step AI work. This guide walks you through creating your first mission, reviewing its plan, and monitoring workers as they execute.
Before starting, make sure you have at least one AI provider configured in Settings → AI and a project open with a valid ade.yaml.
If you are running a packaged (production) build, the Missions tab will show a “Coming soon” preview and the New Mission button is disabled. Mission creation and live runs are currently only available in dev and internal builds — this guide applies as-is there. The shipping version will use the same flow.

1. Create a Mission

Navigate to the Missions tab in the sidebar. Click New Mission to open the creation dialog.
1

Set a goal

Write a clear, specific goal for your mission. The planner agent uses this to generate a structured execution plan.Good goals are concrete: “Add a REST endpoint for user preferences with validation, tests, and OpenAPI docs” is better than “work on the API.”
2

Choose a launch mode

  • Supervised — ADE pauses for your approval before executing the plan. Recommended for your first mission.
  • Autopilot — planning and execution proceed without human gates.
3

Set a budget cap (optional)

Set a maximum token or dollar spend for the mission. If the cap is reached, the mission pauses with a budget_limit_reached intervention instead of continuing.
Click Create to queue the mission.

2. Review the Plan

Once the planner agent finishes, ADE presents the generated plan for review. The plan includes:
  • Steps — numbered tasks the workers will execute
  • Lane assignments — which lane each step runs in
  • Dependencies — ordering constraints between steps
  • Parallelism — which steps can run simultaneously
If you launched in Supervised mode, you will see an approval prompt. Review the steps, then click Approve to start execution or Edit to adjust the plan.
If the plan looks too broad or too narrow, cancel and rewrite the goal with more detail. The planner responds well to acceptance criteria and explicit constraints.

3. Monitor Execution

Once approved, workers begin executing the plan steps. Navigate to the Mission Run panel to monitor progress.
  • Step list — shows each step with its current status (queued, running, completed, failed)
  • Live terminal — click any running step to see its agent’s terminal output in real time
  • Artifacts — completed steps surface artifacts (code changes, test results, summaries) in the Artifacts tab

Step detail panel

Click any step in the step list to open its detail panel. The panel shows:
  • Agent terminal — live-streaming output from the worker agent
  • File changes — a diff view of files modified by this step
  • Tool calls — a log of every ADE CLI tool the agent invoked, with parameters and results
  • Timing — start time, elapsed duration, and token usage
If a step is running in parallel with others, each step has its own independent terminal view. Switch between running steps by clicking their rows in the step list.

4. Handle Interventions

If a worker gets stuck, it creates an Intervention and pauses. Common reasons:
InterventionWhat to do
approval_requiredReview the output and approve or reject
manual_inputProvide the answer the agent needs
conflictResolve the merge conflict, then resume
failed_stepReview the error, fix if needed, then retry or skip
Navigate to the intervention banner at the top of the Mission Run panel, provide your input, and click Resume.

5. View Results

When all steps complete and validation passes, the mission moves to Completed status. Review the results:
  • Artifacts tab — all proof artifacts (screenshots, test reports, summaries) collected during execution
  • Plan tab — the final plan with per-step outcomes
  • Chat tab — the full conversation history between the coordinator and workers
  • Cost summary — total tokens and estimated cost for the mission

Linking results to PRs

After a mission completes, you can create a PR directly from the mission’s result lane. The PR description is pre-filled with the mission summary, step outcomes, and links to proof artifacts. Reviewers see exactly what the AI did and what evidence supports it. To export the full mission audit bundle: Mission detail → Actions → Export audit bundle. The bundle includes the plan, all step transcripts, artifacts, cost breakdown, and timing data.

Next Steps

Mission Concepts

Understand the full mission lifecycle, statuses, and intervention types.

Automation Rules

Trigger missions automatically from git events or schedules.