Create New CLI Skills Automatically

skill-creator guides the creation of reusable CLI skills through discovery, collaborative planning, optional prompt refinement, template-based generation, validation, and installation. It packages custom capabilities or domain knowledge into a skill structure while following progressive disclosure and writing standards. Platform targets include GitHub Copilot CLI, Claude Code, and Codex.

Discovery identifies available platforms, the working directory and repository root, the presence of a skill repository structure, and author information from Git configuration. Planning establishes the skill's purpose, scope, trigger phrases, type, target platforms, and one-sentence description. Skill types include general work, code generation or modification, documentation, and analysis.

An optional prompt-engineer step refines the initial description when that skill is available. The enhanced version is reviewed with the user, who can accept it or retain the original. File generation converts the skill name to kebab-case, creates the relevant skill directories, selects platform templates, and substitutes name, description, triggers, purpose, author, date, and version fields. Longer material is moved into references so the main skill stays focused.

Validation checks YAML frontmatter, required fields, description format, word counts, and writing style. Common repairs include third-person descriptions, imperative phrasing, and adding required fields. Critical validation failures block installation, while warnings remain informational. Specific errors are displayed for correction rather than hidden behind a completion message.

Use it to extend CLI functionality, automate repetitive CLI work, package domain knowledge, or create a skill with repository and global installation options. The workflow applies when the task matches skill creation rather than an unrelated implementation problem.

Clarify missing inputs, permissions, safety boundaries, or success criteria before proceeding. Generated files still need environment-specific testing and review. When platform detection fails, the workflow offers repository-only installation, manual platform selection, or skipping installation. Missing templates require a valid template repository or a manually created minimal structure. Existing installations call for an explicit choice between replacement, renaming, skipping, or another location.

Inputs include purpose, trigger phrases, skill type, platform targets, installation preference, repository context, author metadata, and a concise description. Outputs include SKILL.md, README.md, and directories for references, examples, and scripts. A completion summary records the created files, location, installation state, and next steps such as testing triggers, adding examples, extending guidance, validating, committing, and sharing.

The quality guidance favors a main skill of 1,500-2,000 words with a maximum below 5,000, and a README of 300-500 words. Main skill descriptions use third-person wording, while instructions use imperative or infinitive phrasing. README content provides user-facing installation guidance and usage examples. Progress displays mark the workflow phases.

Template application uses sed substitutions while preserving YAML and multiline formatting. Git provides repository and author information, and generation respects ignore patterns. Validate skill YAML with:

scripts/validate-skill-yaml.sh ".github/skills/$SKILL_NAME"

Installation can remain in the repository, add global symlinks, do both, or be skipped. Absolute-path symlinks are verified after creation and allow repository updates to reach installed skills. Supporting references address development guidance, writing style, progressive disclosure, and validation, alongside basic skill creation, resources, and global installation. Maintenance scripts support batch validation, version changes, and skill index generation.