enoch-robinson/agent-skill-collection

Overview

This skill is a concise guide for creating, updating, and testing high-quality Skills that extend an AI assistant’s capabilities. It explains required metadata, structure, resource packaging, and a practical creation workflow to produce reliable, reusable Skills. The goal is fast iteration and predictable triggers for real-world use.

How this skill works

The skill inspects design choices, required metadata (name and description), and recommended directory layout to ensure each Skill is discoverable and lightweight. It prescribes a three-tier content loading approach (minimal metadata, main instructions loaded on trigger, and optional bundled resources loaded on demand) and a step-by-step creation and validation workflow. It also provides checklists and examples to speed testing and iteration.

When to use it

• You need to create a new Skill to add domain-specific capabilities to an assistant.
• You want to update an existing Skill to improve triggers, examples, or resources.
• You are designing triggers and descriptions so the assistant reliably selects the right Skill.
• You must package reusable scripts, references, or templates for on-demand loading.
• You want a checklist to validate a Skill before production deployment.

Best practices

• Keep the manifest metadata short and explicit: unique lowercase name plus a trigger-focused description.
• Favor concise examples over long explanations; tokens are shared with conversation context.
• Use progressive disclosure: keep only essential metadata always loaded and load detailed content or assets on demand.
• Match instruction specificity to task fragility: high-level text for flexible tasks, structured scripts for semi-deterministic tasks, and concrete scripts for brittle operations.
• Store large or rarely used reference material as separate resources and reference them clearly from the main instructions.

Example use cases

• Create a PDF processing Skill with a clear description of triggers and sample commands for extraction and merging.
• Package a data-cleaning Skill that exposes reusable Python scripts and a short usage example for different datasets.
• Build a deployment Skill that includes step-by-step runbooks and on-demand logs or templates.
• Update a Skill by tightening its trigger description and adding a failing test case to the validation checklist.