skill-creator_skill

Overview

This skill helps authors create reusable, searchable skills for AI agent workflows. It defines when to capture a technique, how to structure the document, and what search keywords and examples to include so future agents and humans can discover and apply the pattern.

How this skill works

It inspects the technique’s intent, triggers, and outcomes, then prescribes a compact structure: a short description, a focused overview, clear when-to-use triggers, a core pattern with multiple examples, and common pitfalls. It emphasizes positive framing, keyword-rich descriptions for discovery, token efficiency, and trusting the LLM to handle implementation details.

When to use it

• When a technique is broadly reusable across projects rather than one-off
• When a pattern wasn’t intuitively obvious and you’d reference it later
• When outcomes and constraints matter more than a step-by-step procedure
• When searchability matters — include error codes and symptom keywords
• Not for trivial, well-documented standards or project-specific conventions

Best practices

• Start descriptions with a concise trigger phrase like “Use when…” and include concrete symptoms
• Show the desired pattern with multiple short examples, and explain undesired alternatives in prose
• Keep content compact (prefer <500 words; <200 for frequently loaded skills) to save tokens
• Include searchable keywords (errors, symptoms, tool names) in the description and body
• Prefer self-contained single-file artifacts unless you need heavy reference docs or reusable code

Example use cases

• Capture a reliable async-test pattern with several condition-based wait examples and failure modes
• Document a reusable API pagination approach with clear triggers and acceptable trade-offs
• Save a cross-project error-handling pattern including common error strings (e.g., ENOTEMPTY, timeout) for discovery
• Describe a small CI workflow pattern that focuses on outcomes not exact commands
• Create a pattern for prompting LLMs that specifies success criteria and constraints rather than stepwise logic