Skip to main content
Automations let you wire up triggers to AI execution surfaces so that work happens without manual intervention. This guide covers creating a rule, choosing triggers, selecting executors, and setting guardrails.

1. Open the Automations Panel

Navigate to Automations in the sidebar. The panel has three tabs:
  • Rules — your active and paused automation rules
  • Templates — pre-built rule templates you can clone and customize
  • History — past automation runs with status, duration, and output
Click New Rule to start creating a rule from scratch, or pick a template from the Templates tab.

2. Choose a Trigger

Every rule starts with a trigger. Select one from the trigger type dropdown:
Fires when commits are pushed to a matching branch pattern.
  • Branch filter — glob pattern (e.g., feature/*, main)
  • Path filter (optional) — only fire when specific files are changed
Example: run tests on every push to feature/* branches.
Fires on PR lifecycle events: opened, updated, merged, or review requested.
  • Event filter — select which PR events to match
  • Label filter (optional) — only fire for PRs with specific labels
Example: auto-review PRs labeled needs-ai-review.
Fires on a cron schedule.
  • Cron expression — standard 5-field cron syntax
  • Timezone — defaults to your system timezone
Example: run a dependency audit every Monday at 9:00 AM.
Fires when watched files change on disk.
  • Path pattern — glob pattern for files to watch (e.g., src/api/**/*.ts)
Example: regenerate API docs when any file in src/api/ changes.

3. Select an Executor

The executor determines how the triggered work runs. Choose one:
ExecutorBest forOutput location
Agent SessionQuick, focused tasks (linting, reviews, small fixes)Automations > History
MissionMulti-step work requiring planning, validation, and artifactsMissions tab
Built-in TaskPredefined ADE operations (pack refresh, conflict scan)Automations > History
For agent sessions, you provide a prompt template — the instruction the agent receives when the trigger fires. Use {{branch}}, {{files}}, and {{event}} placeholders to inject trigger context into the prompt. Coming soon: detailed prompt template reference with all available placeholders.

4. Configure Budget Caps

Every automation rule should have a budget cap to prevent runaway spend.
1

Set a per-run cap

Navigate to the Guardrails section of the rule editor. Set the maximum token count or dollar amount per run.
2

Set a monthly cap (optional)

Limit total spend across all runs of this rule within a calendar month.
3

Choose the over-budget action

  • Pause — stop the run and create an intervention
  • Skip — cancel the run silently and log it

5. Set Guardrails

Beyond budgets, guardrails control what the automation is allowed to do:
  • File scope — restrict which files the agent can read or write
  • Branch scope — restrict which branches the agent can push to
  • Tool allowlist — limit the MCP tools available to the agent
  • Approval gate — require human approval before the agent commits or pushes
Coming soon: detailed guardrail configuration reference.

6. Enable the Rule

Review your rule configuration, then toggle the Enabled switch. The rule is now live and will fire the next time its trigger condition is met. You can pause, edit, or delete rules at any time from the Rules tab. Paused rules retain their configuration but do not fire.

Monitoring Runs

Check Automations > History to see past runs. Each entry shows:
  • Trigger event that fired the rule
  • Executor type and run duration
  • Status (completed, failed, budget-exceeded)
  • Link to the full output (agent session transcript or mission)

Next Steps

Automation Concepts

Deep dive into automation architecture, trigger routing, and execution surfaces.

Guardrails Reference

Full reference for budget caps, file scopes, and approval gates.