Agent Workflow Executor

This skill executes the 3-agent autonomous workflow for coding tasks.

Workflow

research → implement → verify → done

Agent	Purpose
research	Understand codebase, search memories, create plan
implement	Write code following the plan
verify	Run tests, review code, complete task

Arguments

$ARGUMENTS

Instructions

Step 1: Ensure task exists

If a task ID (CX-N) was provided, get the task:

mcp__cortex__task(action="get", id="CX-N")

If a description was provided, create a task first:

mcp__cortex__task(action="create", title="<description>", type="feature|bug|chore")

Step 2: Start the agent workflow

Spawn the research agent to begin:

mcp__cortex__agent(action="spawn", task_id="CX-N", agent_name="research")

This returns a prompt and session_id.

Step 3: Execute the research agent

Use the Task tool with the prompt from agent_spawn:

Task(
  subagent_type="research",
  prompt="<prompt from agent_spawn>",
  run_in_background=false
)

The research agent will:

• Search memories for relevant context
• Explore the codebase
• Create an implementation plan
• Call agent_report when done

Step 4: Continue the chain

When the agent completes and calls agent_report, check the response:

• If next_action.action == "spawn_agent", execute the next agent
• If next_action.action == "workflow_complete", the task is done

Step 5: Repeat for implement and verify

Execute each agent's prompt using the Task tool until workflow is complete.

Example Usage

/implement CX-104

→ Executes full workflow for task CX-104

/implement Add user authentication

→ Creates task, then executes full workflow

Monitoring Progress

Check agent sessions:

mcp__cortex__agent(action="sessions", task_id="CX-N")

Important Notes

• Each agent is autonomous and will complete its phase
• Agents save memories and update task status automatically
• The workflow handles the full cycle: research → implement → verify
• Task is marked as done when verify completes