reneromero08/THOUGHT/LAB/TURBO_SWARM/agents_skills_alpha/swarm-directive

<!-- CONTENT_HASH: 9bd07a4ed03d6eea49673a43682a62ca7df8b0cabb7ff424ac91cd35e9e7bea7 -->

Skill: swarm-directive

Version: 0.1.0

Status: Draft

required_canon_version: ">=2.8.0 <3.0.0"

Swarm Directive Skill

Send tasks to your CATALYTIC-DPT swarm from Claude Code, Kilo CLI, or Cline CLI.

Quick Start

Option 1: Direct CLI Command (Simplest)

cd "d:\CCC 2.0\AI\agent-governance-system"

# Create input
cat > /tmp/swarm_task.json << 'EOF'
{
  "directive": "Analyze the MCP architecture",
  "task_type": "research"
}
EOF

# Send to swarm
python SKILLS/swarm-directive/run.py /tmp/swarm_task.json /tmp/result.json

# View result
cat /tmp/result.json

Option 2: With Result Waiting

cat > /tmp/swarm_task.json << 'EOF'
{
  "directive": "Analyze code structure",
  "task_type": "research",
  "wait_for_result": true,
  "timeout": 30
}
EOF

python SKILLS/swarm-directive/run.py /tmp/swarm_task.json /tmp/result.json

Option 3: Kilo/Cline Integration

In your Kilo or Cline CLI, run:

python SKILLS/swarm-directive/run.py input.json output.json

Input Format

{
  "directive": "Your task description here",
  "task_type": "research|file_operation|code_adapt|validate",
  "wait_for_result": false,
  "timeout": 30
}

Fields:

• directive (required): Task description to send to the swarm
• task_type (optional): Type of task. Default: "research"
• wait_for_result (optional): Block until task completes. Default: false
• timeout (optional): Max seconds to wait for result. Default: 30

Output Format

{
  "status": "success",
  "task_id": "swarm-1234567890",
  "task_status": "dispatched|pending|completed|failed",
  "message": "Human-readable status",
  "task_spec": {...},
  "result": {...}
}

Examples

Example 1: Fire and Forget

Send a task without waiting for completion:

cat > task.json << 'EOF'
{
  "directive": "Copy README.md to backups/README-backup.md",
  "task_type": "file_operation"
}
EOF

python SKILLS/swarm-directive/run.py task.json result.json

# Task dispatched! Check CONTRACTS/_runs/mcp_ledger/task_results.jsonl for completion

Example 2: Wait for Completion

Block until task is done:

cat > task.json << 'EOF'
{
  "directive": "Analyze the swarm architecture",
  "task_type": "research",
  "wait_for_result": true,
  "timeout": 60
}
EOF

python SKILLS/swarm-directive/run.py task.json result.json

# Returns immediately when task completes or timeout

Example 3: From Claude Code

import subprocess
import json

task = {
    "directive": "Read and summarize ROADMAP.md",
    "task_type": "research",
    "wait_for_result": True,
    "timeout": 45
}

with open("/tmp/input.json", "w") as f:
    json.dump(task, f)

result = subprocess.run(
    ["python", "SKILLS/swarm-directive/run.py", "/tmp/input.json", "/tmp/output.json"],
    cwd="d:\\CCC 2.0\\AI\\agent-governance-system"
)

with open("/tmp/output.json") as f:
    output = json.load(f)
    print(f"Task: {output['task_id']}")
    print(f"Status: {output['task_status']}")

Prerequisites

Before using this skill, start the swarm:

python CATALYTIC-DPT/SKILLS/swarm-orchestrator/scripts/launch_swarm.py start

Check swarm status:

python CATALYTIC-DPT/SKILLS/swarm-orchestrator/scripts/launch_swarm.py status

Stop swarm:

python CATALYTIC-DPT/SKILLS/swarm-orchestrator/scripts/launch_swarm.py stop

Monitoring

Watch tasks in real-time:

# Watch directives sent to Governor
tail -f CONTRACTS/_runs/mcp_ledger/directives.jsonl

# Watch task queue
tail -f CONTRACTS/_runs/mcp_ledger/task_queue.jsonl

# Watch results
tail -f CONTRACTS/_runs/mcp_ledger/task_results.jsonl

Error Handling

If you get an error:

1. "Swarm not running" - Start with launch_swarm.py start
2. "Task timed out" - Increase timeout value
3. "Governor not responding" - Check launch_swarm.py status

Architecture

Kilo/Cline CLI
    ↓
swarm-directive/run.py
    ↓ (sends via MCP)
MCP Server
    ↓ (writes to ledger)
CONTRACTS/_runs/mcp_ledger/directives.jsonl
    ↓
Governor (reads directive)
    ↓ (dispatches task)
Ant Worker (executes task)
    ↓ (writes result)
CONTRACTS/_runs/mcp_ledger/task_results.jsonl
    ↓ (read by skill)
Kilo/Cline gets result