writing-skills

To create and validate agent skills, you structure reusable workflows using frontmatter and instructional markdown. This process ensures future agents can reliably discover guidance and follow instructions under pressure by satisfying discovery, documentation, and deployment requirements.

Progressive disclosure in skill documentation is a structure that organizes references, scripts, and examples so agents can process instructions efficiently. It allows skills to present baseline guidance first, preventing cognitive overload when agents execute complex workflows.

You test and harden AI skills by applying baseline scenarios, pressure cases, and rationalization blocking to the instructional markdown. This validation step ensures the skill holds up under pressure and prevents agents from deviating from the intended workflow.

Graphviz is required as a dependency to build and edit SKILL.md directories. You need it installed in your environment to fully utilize the skill creation, hardening, and validation features for your reusable agent workflows.

The best way to structure metadata for discoverable AI skills is to use clear frontmatter with rationalized naming conventions. This ensures agents can accurately discover the right guidance and follow it when deploying skill directories.

Agent skills fail under pressure when they lack proper hardening through baseline scenarios and rationalization blocking. You must validate the SKILL.md instructional markdown and supporting files to ensure robust execution and prevent deviation.