Actions

Actions are APM's workflow system for managing approvals, reviews, and decisions in AI-augmented work.

What are Actions?

Actions capture moments when human judgment is needed:

Approve an AI-generated implementation
Review a code change
Validate data accuracy
Assign to a more experienced teammate

Action Types

📋

Implementation Plans

Review and approve AI-generated task plans

✨

Task Enhancements

Approve modifications to existing tasks

🔧

Custom Actions (coming soon)

Define your own approval workflows

Priority Levels

🔴 Critical - Immediate attention required
🟠 High - Important, address soon
🟡 Medium - Standard priority
🟢 Low - Can wait

Creating Actions via MCP

Actions are created through the createAction MCP tool when human approval or review is needed:

# Create an implementation plan review
action = mcp_client.call_tool("createAction", {
  type: "implementation_plan",
  title: "Review database migration plan",
  priority: "high",
  requester_id: "ai_agent_123",
  assignee_id: "user_456",
  payload: {
    task_id: "task_789",
    proposed_plan: <<~PLAN
      1. Backup current database to S3
      2. Run migration scripts in transaction
      3. Verify data integrity with checksums
      4. Update schema version in metadata
    PLAN
  }
})

# Create a deployment approval
action = mcp_client.call_tool("createAction", {
  type: "deployment_approval",
  title: "Deploy v2.3.1 to production",
  priority: "critical",
  requester_id: "deploy_bot",
  assignee_id: "ops_lead",
  payload: {
    version: "2.3.1",
    changes: ["Fixed auth bug", "Added rate limiting"],
    rollback_plan: "Revert to v2.3.0 if issues detected"
  }
})

Updating Actions

Use the updateAction tool to process actions:

# Approve an action
mcp_client.call_tool("updateAction", {
  action_id: "action_123",
  status: "approved",
  resolution_notes: "Plan looks good, proceed with migration"
})

# Reject with feedback
mcp_client.call_tool("updateAction", {
  action_id: "action_123",
  status: "rejected",
  resolution_notes: "Missing rollback procedure, please add"
})

# Mark as in progress
mcp_client.call_tool("updateAction", {
  action_id: "action_123",
  status: "in_progress",
  assignee_id: "user_789"  # Reassign if needed
})

Querying Actions

The getActions tool provides filtering capabilities:

# Get all pending actions for a user
my_actions = mcp_client.call_tool("getActions", {
  assignee_id: "user_456",
  status: ["pending", "in_progress"]
})

# Get critical actions across all projects
urgent = mcp_client.call_tool("getActions", {
  priority: "critical",
  status: "pending"
})

# Get actions for a specific project
project_actions = mcp_client.call_tool("getActions", {
  project_id: "proj_123",
  include_resolved: false
})

Action Events

APM broadcasts these action-related events (coming soon):

apm_action_created - New action requires attention
apm_action_updated - Action status or properties changed
apm_action_approved - Action was approved
apm_action_rejected - Action was rejected
apm_action_expired - Action timed out without resolution

Automation Patterns

Auto-Escalation

def on_action_created(event)
  action = event["action"]
  
  # Escalate critical actions via other channels
  if action["priority"] == "critical"
    notify_slack(
      "🚨 Critical action needs approval: #{action['title']}"
    )
  end
end

Conditional Approval

def on_action_created(event)
  action = event["action"]
  
  # Auto-approve low-risk changes
  if action["type"] == "config_change"
    risk_score = analyze_risk(action["payload"])
    if risk_score < 0.3
      mcp_client.call_tool("updateAction", {
        action_id: action["id"],
        status: "approved",
        resolution_notes: "Auto-approved: Low risk change"
      })
    end
  end
end

Best Practices

Clear Titles - Action titles should immediately convey what needs approval
Appropriate Priority - Use critical sparingly, for truly urgent items
Rich Payloads - Include all context needed for informed decisions