Universal Agent Skills Generator

Transform any high-level goal into a production-ready set of atomic, executable AI agent skills — no prompt engineering required.

Describe what you want your AI agent to accomplish, and this Actor generates a complete skill system: validated skill definitions, a dependency-aware execution graph, failure recovery strategies, and ready-to-use exports for CrewAI and OpenAI Agents.

---

How It Works

Your Goal → Intent Parsing → (Optional) Web Scraping → Knowledge Distillation
         → Skill Synthesis → Self-Healing Validation → Caching & Versioning
         → Execution Graph → Framework Export → Output

1. Intent Normalization — Parses your goal, detects domain keywords, infers the execution domain, and identifies knowledge gaps automatically.
2. Smart Scraping (optional) — Scrapes up to 5 public documentation domains (≤2 pages each) using Cheerio. Fully configurable and rate-limited.
3. Knowledge Distillation — Extracts capabilities, API patterns, constraints, and failure modes from scraped content.
4. Skill Synthesis — Generates single-responsibility skills with explicit typed inputs, outputs, constraints, and failure handling.
5. Self-Healing Validation — Auto-fixes weak descriptions, injects missing constraints, removes near-duplicates, and rejects hostile content (TODO, FIXME, etc.).
6. Caching & Versioning — Skills are cached with deterministic hashes. Re-runs produce semantic diffs with automatic semver version bumps.
7. Execution Graph — Builds a DAG (directed acyclic graph) with no circular dependencies, defining the correct skill execution order.
8. Framework Export — Generates drop-in configurations for CrewAI agents and OpenAI tool definitions.

---

Input Parameters

Parameter	Type	Default	Description
Goal (required)	string	—	The high-level goal for your AI agent. Be as specific as possible.
Preferred Tools	string[]	[]	Languages, frameworks, or tools the agent should prefer (e.g. node, python, docker).
Enable Web Scraping	boolean	true	Scrape public docs for domain knowledge. Limited to ≤5 domains, ≤2 pages each.
Allowed Scraping Domains	string[]	[]	Restrict scraping to specific domains. Leave empty for automatic selection.
Depth Level	string	medium	Controls skill granularity: low (3–5 broad skills), medium (5–8 balanced), high (8–14 fine-grained).
Export Formats	string[]	[]	Generate compatibility exports: crewai, openai, or both.

---

Output

Each run produces two outputs:

1. Dataset — Individual Skill Records

Every generated skill is pushed as a separate record to the default dataset, containing:

{
  "skill_id": "skill-setup-project-scaffolding",
  "skill_name": "Setup Project Scaffolding",
  "category": "project-setup",
  "description": "Initialize directory structure, package.json, and dependencies...",
  "inputs": [{ "name": "project_name", "type": "string", "required": true }],
  "outputs": [{ "name": "project_root", "type": "string" }],
  "actions": ["Create project directory", "Initialize package.json", "..."],
  "tools": ["filesystem", "npm"],
  "constraints": ["Must produce deterministic output", "Must not modify inputs"],
  "failure_modes": [{ "condition": "disk_full", "fallback": "Report error with path" }],
  "editor_metadata": {
    "deterministic": true,
    "ui_safe": true,
    "callable_independently": true,
    "idempotent": true
  }
}

2. Key-Value Store — Full Output JSON

The complete output is saved to the OUTPUT key, including:

• skills — All validated skills
• execution_graph — DAG with nodes and edges defining dependencies
• validation_summary — Total, valid, rejected, healed, near-duplicate counts
• cache_summary — Cache hits vs. new skills
• diff_summary — Semantic diffs with version bumps (on re-runs)
• exports — CrewAI agents and/or OpenAI tool definitions (if selected)
• assumptions — What the Actor inferred about your goal

---

Example

Input

{
  "goal": "Build a REST API with Express.js and MongoDB for a task management app",
  "preferred_tools": ["node", "express", "mongodb"],
  "scraping_enabled": true,
  "depth_level": "medium",
  "export_formats": ["crewai", "openai"]
}

Output (excerpt)

{
  "agent_goal": "Build a REST API with Express.js and MongoDB for task management",
  "skills": [
    {
      "skill_id": "skill-setup-project-scaffolding",
      "skill_name": "Setup Project Scaffolding",
      "category": "project-setup",
      "actions": ["Create directory structure", "Initialize package.json", "Install dependencies"],
      "tools": ["filesystem", "npm"]
    },
    {
      "skill_id": "skill-configure-mongodb-connection",
      "skill_name": "Configure MongoDB Connection",
      "category": "database",
      "actions": ["Setup Mongoose client", "Define connection URI", "Handle connection errors"],
      "tools": ["mongodb", "mongoose"]
    }
  ],
  "execution_graph": {
    "nodes": ["skill-setup-project-scaffolding", "skill-configure-mongodb-connection", "..."],
    "edges": [["skill-setup-project-scaffolding", "skill-configure-mongodb-connection"]]
  },
  "validation_summary": { "total": 8, "valid": 8, "rejected": 0, "healed": 1 }
}

---

Key Features

• Zero-config skill generation from plain-language goals
• Cross-modal support for documents, video, images, and social posts
• Editor-first design — every skill includes editor_metadata for determinism, UI safety, and independent callability
• Self-healing pipeline — auto-fixes weak descriptions, missing constraints, and near-duplicates before output
• Hostile validation — rejects hallucinated tools, placeholder content (TODO/FIXME), and multi-responsibility skills
• Semantic versioning — re-runs diff against cached skills with automatic major/minor/patch bumps
• Framework exports — drop-in CrewAI agent configs and OpenAI tool schemas

---

Pricing

This Actor uses pay-per-event pricing:

Event	Price	Description
Actor Start	$0.00005	Charged once per run (Apify handles this)
Per Skill	$0.20	Charged per skill pushed to the dataset

Typical costs:

• Low depth (5 skills): ~$1.00
• Medium depth (8 skills): ~$1.60
• High depth (14 skills): ~$2.80

---

Use Cases

• AI Agent Development — Generate skill blueprints for LangChain, CrewAI, or custom agent frameworks
• Technical Planning — Break down complex goals into validated, dependency-ordered steps
• Rapid Prototyping — Start with a goal, get a complete skill graph in seconds
• Framework Migration — Export skills as CrewAI agents or OpenAI tools for immediate integration

---

Limitations

• Scraping is limited to ≤5 domains and ≤2 pages per domain to respect rate limits
• Skills are generated deterministically — no LLM calls, no API keys required
• Output format is JSON only (more formats planned)

---

Support

If you encounter any issues or have feature requests, please open an issue on the Actor's page in the Apify Console.