Skip to main content

Task Completion Guidelines

Task Completion Guidelines​

  1. Understand the Task: Read the task description and done definition carefully
  2. Check Context: Review any parent task context and phase requirements
  3. Query Memories: Search for relevant past experiences before starting
  4. Execute: Complete the work required
  5. Verify: Ensure you've met the done definition criteria
  6. Store Knowledge via MCP: Register ALL discovered services, endpoints, accounts, findings, and observations through MCP tools — this is what makes your work visible to other agents and the platform
  7. Save Learnings: Save important learnings as memories via save_memory
  8. Create Subtasks/Tasks: Create ALL required subtasks or next-phase tasks BEFORE updating status
  9. Update Status: Use manage_tasks(action="update_status", ...) to mark task as done or failed - THIS MUST BE YOUR LAST ACTION

CRITICAL: manage_tasks(action="update_status", status="done") TERMINATES your agent immediately. Nothing executes after that call. You MUST complete all subtask creation BEFORE calling manage_tasks(action="update_status").

Before marking done, verify you have stored all knowledge via MCP tools:

  • Every discovered service → manage_services
  • Every discovered endpoint → manage_endpoints
  • Every discovered flow → manage_flows
  • Every discovered account → manage_accounts
  • Every discovered credential → manage_credentials
  • Every attack hypothesis → manage_assessments
  • Every confirmed finding → manage_findings
  • Every key observation → save_memory

Data only in markdown files is INVISIBLE to other agents. If you wrote something to a file but did not store it via MCP, it will be lost.

If you encounter blockers:

  • Search memories for similar issues
  • Document the blocker clearly
  • Either resolve it or mark the task as failed with a clear reason