APEX Skill Forge v8.0

One command. Any archetype. Three distribution formats. Zero guesswork.

Input: Skill name (kebab-case) + archetype + optional parameters Output: Complete skill package (README, MANIFEST.json, executor.py, references/) Success: Forged skill scores >= 9.5/10 on 12-dimension audit Fails When: Invalid name format, unknown archetype, missing output path

I. Decision Router (Start Here)

II. Archetypes

III. Critical Rules (Anti-Hallucination)

• NEVER generate skills without MANIFEST.json (DAG metadata is mandatory)
• NEVER use non-kebab-case names (enforce ^[a-z0-9]+(-[a-z0-9]+)*$)
• NEVER skip audit validation (all packages MUST score >= 9.5/10)
• NEVER hardcode paths (use Path objects and relative paths)
• NEVER mix Claude-specific syntax into universal packages
• ALWAYS include copyright: “Copyright (c) 2026 APEX Business Systems Ltd.”
• ALWAYS generate executor.py with DAG-compatible execute() signature
• ALWAYS include timeout_ms, retry_policy, idempotent in DAG metadata

IV. Forge Protocol (Execution Flow)

1. VALIDATE   → Name regex + archetype check
2. SCAFFOLD   → Create directory: skills/<name>/
3. GENERATE   → README.md + MANIFEST.json + scripts/executor.py
4. POPULATE   → references/ + LICENSE
5. AUDIT      → Run 12-dimension validator (>= 9.5 required)
6. SHIP       → Package to target format (claude/universal/omnihub)

Generated Directory Structure

<skill-name>/
├── README.md           # Universal documentation
├── MANIFEST.json       # DAG metadata + skill config
├── LICENSE             # APEX Business Systems Ltd.
├── scripts/
│   ├── __init__.py
│   └── executor.py     # DAG-compatible execute() entry point
└── references/
    ├── examples.md
    └── troubleshooting.md

V. DAG Executor Contract

def execute(input_context: dict[str, Any]) -> dict[str, Any]:
    """
    input_context: {
        "execution_id": str,
        "parameters": Dict,
        "context": Dict,      # From prior DAG nodes
        "metadata": Dict
    }
    Returns: {
        "result": Any,
        "context_updates": Dict,  # State for next node
        "metadata": {
            "execution_time_ms": float,
            "node_status": "success|error|timeout",
            "logs": List[Dict],
            "recoverable": bool
        }
    }
    """

VI. MANIFEST.json Schema (Required DAG Fields)

{
  "name": "skill-name",
  "version": "1.0.0",
  "archetype": "workflow",
  "description": "Triggers: ... Actions: ... Produces: ...",
  "dag": {
    "node_type": "processor|orchestrator|validator|transformer",
    "accepts": ["application/json", "text/plain"],
    "emits": ["event:completion", "event:failure"],
    "timeout_ms": 30000,
    "retry_policy": { "max": 3, "backoff": "exponential" },
    "idempotent": true,
    "side_effects": false,
    "omnihub_compatible": true
  }
}

VII. 12-Dimension Audit

Aggregate threshold: >= 9.5/10 average

VIII. Tri-Format Packaging

IX. Scripts Reference

# Scaffold a new skill
python scripts/forge.py data-transformer --arch transformer --path ./skills

# Audit a skill (verbose)
python scripts/audit.py ./skills/data-transformer --verbose --threshold 9.5

# Package for Claude
python scripts/ship.py ./skills/data-transformer --format claude

# Package for universal distribution
python scripts/ship.py ./skills/data-transformer --format universal

# Package for OmniHub
python scripts/ship.py ./skills/data-transformer --format omnihub

# Run executor status check
python scripts/executor.py '{"parameters":{"action":"status"}}'

X. Troubleshooting

APEX Skill Forge v8.0 — Patent Pending. Copyright (c) 2026 APEX Business Systems Ltd. All Rights Reserved. Edmonton, AB, Canada — https://apexbusiness-systems.com