FlagOS Skill Creator

Problem

Skills in the FlagOS repository often have vague trigger conditions, incomplete frontmatter, undocumented scripts or references, and weak examples. As a result, an agent may trigger the skill at the wrong time, miss the right time, or ship a directory that looks complete but cannot be reliably reviewed. This skill turns creation, improvement, and validation into a structured workflow that keeps the document, scripts, references, and test prompts aligned.

How It Works

It operates through four modes: create, improve, validate, and eval. Creation starts by clarifying purpose, trigger conditions, expected output, required tools, and script dependencies, then scaffolds the directory, fills frontmatter, writes the body, and adds scripts/, references/, or assets/ where needed. Improvement reviews whether the description is specific enough, whether steps include progress markers, whether SKILL.md is too long, and whether examples or troubleshooting are missing. It also creates evals/evals.json entries that replay realistic user prompts. Validation checks YAML frontmatter, name and directory alignment, description length, minimum body content, existence of referenced files, script permissions, and risks such as hardcoded paths or credentials.

Boundary

It is best used for maintaining well-formed skills in the FlagOS skills repository, not as a general prompt-writing tool. It expects access to the skill root, scripts, and reference files. Anything listed in SKILL.md must exist on disk. Detailed material should move into references/ so the main file stays concise.