LLM Agent Skills Repository

This repository is a Dynamic Hub for LLM-based agents skills. It is designed to host a modular, evolving collection of "Superpowers"—structured knowledge bases and deterministic toolsets that enable agents to perform complex, domain-specific tasks with high reliability and low token overhead.

🧠 Core Philosophy: The Skill-Forge Logic

The repository operates on the principle that AI agents are most effective when they follow a Discovery-to-Execution lifecycle, rather than having all context loaded at once.

1. Token-Efficient Hierarchy

To prevent context rot, the repository uses a tiered loading system:

• Global Indexing: Agents first discover skills by listing directories or reading high-level summaries (AGENTS.md). Only the name and description are exposed initially.
• Progressive Disclosure: Full instructions (SKILL.md) are only loaded once an agent determines a high alignment (e.g., >1% relevance) with the current task.
• Deep Context: Massive documentation or specific schemas reside in references/ or resources/, loaded only when the agent needs specific technical details.

2. Deterministic Tooling vs. Probabilistic Reasoning

Whenever possible, heavy logic is offloaded to Python scripts within each skill's scripts/ directory. This ensures:

• Reliability: The LLM manages the workflow; the code handles the execution.
• Consistency: Complex operations (like file conversions or API syncs) follow a repeatable path regardless of the model's "mood."

3. The "Skill Forge" Standard

New capabilities are added following the Skill Forge doctrine:

• Audit-First: Every skill must pass structural and security audits before deployment.
• TDD Optimization: Descriptions and instructions are iteratively refined using trigger-evaluation scripts to ensure agents know exactly when to invoke the skill.

🛠️ Repository Mechanics

For Agents

1. Explore: Use list_dir to see available folders. Folders are named after the capability.
2. Evaluate: Read only the YAML frontmatter of a candidate skill to verify its trigger conditions.
3. Execute: Follow the SKILL.md instructions, prioritizing the use of provided CLI tools or scripts.

For Developers

• Modularity: Skills are self-contained. Deleting a directory removes the capability without breaking the repo's core orchestration.
• Slash Workflows: Use project-level commands (/skill-create, /skill-audit) to scaffold and validate skills. These scripts handle the heavy lifting of repository maintenance.

---

Empowering AI agents through structured, deterministic capabilities.