Agent Skills provide a standardized way to give AI agents new capabilities and expertise within the Nox ecosystem. This project follows the Agent Skills Open Standard.
A skill is a portable, version-controlled package containing a SKILL.md file with metadata and instructions. Skills allow agents to capture specialized knowledge into discoverable capabilities.
skills/skill-name/
├── SKILL.md # Required: YAML frontmatter + instructions
├── scripts/ # (Optional) Executable code
├── references/ # (Optional) Technical deep-dives
└── assets/ # (Optional) Templates and resources
Every SKILL.md MUST begin with YAML frontmatter:
---
name: skill-name # Kebab-case, matches directory
description: "..." # Clear trigger-focused description
license: MIT
metadata:
version: "1.0.0"
---skill-creator: Meta-skill for designing new Agent Skills.bug-hunter: Expert procedure for logic forensics and engine auditing.bytecode-instruction-adder: Workflow for adding new Opcodes and syntax.real-world-challenge-solver: Protocol for 100% success-driven user challenges.performance-optimizer: Guardrails for Bytecode-First purity.monorepo-package-generator: Standards for monorepo expansion.
- Focus: Keep each skill focused on one specific job.
- Standardization: Adhere to the
agentskills.iospecification. - Progressive Disclosure: Move long reference material to
references/to save context tokens during discovery. - Validation: Always include steps for the agent to verify its work.