Skill Writing

Create a project-local skill in .agents/skills/. Expose it to Claude Code through a relative symlink. Verify the result with the repository's canonical skill validator.

Prompting Guidance

Write every new skill and its content for current Claude and GPT agent models, not for one model. Do not name a model, a model version, or a model-specific default in skill prose. Content that selects, configures, or attributes a model is the exception, for example a model: field or API parameters.

Base the skill on the official vendor guides. Read both guides before designing or writing a complex, long-running, multi-tool, or orchestration-heavy skill. Their recommendations may change.

• Anthropic prompting best practices: Covers all current Claude models. Its model-specific guidance table links each per-model prompting page.
• OpenAI latest-model guide: Covers the current GPT model family. Its prompting recommendations come from tests on one model in that family.