Skip to main content
Version: v0.2 (Next)

Autonomous Missions & Goal Execution

The Missions engine enables ActonOS agents to execute complex, multi-stage objectives autonomously over extended durations while maintaining durable state, dependency tracking, and human-in-the-loop oversight.

1. Defining a Mission Goalโ€‹

To create a new mission:

  1. Navigate to Missions in the left sidebar.
  2. Click + Launch Mission.
  3. Fill in the mission parameters:
    • Goal Directive: The primary high-level objective (e.g., Audit the /workspace/src codebase for security vulnerabilities, generate a comprehensive report in /workspace/docs/security_audit.md, and open a GitHub issue with the summary).
    • Assigned Agent: Select the primary orchestrator agent (e.g., Security Engineer).
    • Priority: Low, Standard, High, or Critical.
    • Target Channel Notification: Optional channel (e.g., Telegram - Dev Channel) to notify upon completion.

2. Dependency-Aware DAG Executionโ€‹

When a mission begins, the Planner decomposes the goal into a Directed Acyclic Graph (DAG):

  • Node Validation: Rejects duplicate nodes, unknown dependencies, and circular loops.
  • Ordered Sequencing: Independent subtasks run concurrently; dependent subtasks await prerequisite outputs.
  • Live Progress Matrix: The UI visualizes task states: Pending, Running, Completed, Blocked, or Failed.

3. Durable Checkpointing & Resumptionโ€‹

Every state transition and intermediate output is persisted in agent_runs.checkpoint_json:

  • Fault Tolerance: If the host experiences a power outage or container restart, ActonOS automatically resumes active missions from their last committed checkpoint without re-running completed steps.
  • Paused Approvals: When a tool requires clearance, the entire execution state (messages, token counters, pending tool payload) is frozen. Once approved via the dashboard or API, the exact step executes immediately.

4. Stall Detection & Auto-Escalationโ€‹

To prevent infinite loops and wasted token budgets, ActonOS includes built-in Stall Tracking:

  • If an agent generates identical progress metrics ([PROGRESS: X%]) or repeats repetitive tool calls for 3 consecutive cycles without advancing the DAG, the engine emits a [STALL WARNING].
  • The warning is logged to the mission feed and surfaced as a high-priority notification to the operator.
  • The operator can provide steering instructions via chat or pause the mission safely.