mjmorales/skill-creator

Skill Creator

Create Claude Code skills following best practices for structure, description tuning, and resource bundling.

Shared conventions: See references/creator-conventions.md for the standard creator workflow (gather, generate, quality self-check, validate, review gate, commit).

Prompting best practices: See references/prompt-engineering-guide.md for optimization techniques to apply when generating the skill body.

Workflow

1. Gather Requirements

Use AskUserQuestion for discrete choices, free-form for open-ended questions.

Skill purpose (free-form):

• What task does this skill automate?
• When should Claude invoke it? What trigger phrases would a user naturally say?

Skill location — AskUserQuestion with header "Location":

• "Project" (.claude/skills/<name>/ — versioned with the repo)
• "User Global" (~/.claude/skills/<name>/ — available across all projects)
• "Plugin" (skills/ — if adding to a prove-style plugin)

Interaction pattern — AskUserQuestion with header "Interaction":

• "Autonomous" — skill runs without user input
• "Interactive" — skill uses AskUserQuestion gates
• "Delegating" — thin wrapper that delegates to an agent

Resources needed — AskUserQuestion with header "Resources" (multiSelect):

• "Reference docs" — bundled best practices or domain knowledge
• "Templates/assets" — output templates with placeholders
• "Scripts" — executable Python or shell scripts
• "None"

2. Generate the Skill

Skill name — derive a hyphen-case name from the purpose (e.g., code-reviewer, deploy-checker, migration-planner).

Create the directory structure based on gathered requirements:

• Always: skills/<name>/SKILL.md
• If references: skills/<name>/references/<topic>.md
• If assets: skills/<name>/assets/<template>.md

Description tuning — the description is the highest-leverage field. Refer to references/skills-best-practices.md for the structure:

• Front-load the primary action
• Include 2-3 trigger scenarios
• List natural trigger phrases

Body structure — match the interaction pattern:

• Autonomous: numbered workflow steps
• Interactive: gather → execute → validate → review gate
• Delegating: before delegating → delegation instructions → after delegation

Apply the quality self-check from references/creator-conventions.md before presenting.

3. Validate

Before writing, verify:

• [ ] Directory name matches the name frontmatter field convention
• [ ] name and description frontmatter fields are present
• [ ] Description front-loads the action and includes trigger scenarios
• [ ] Body uses numbered steps for workflows
• [ ] Interactive skills follow references/interaction-patterns.md patterns
• [ ] Token budget is appropriate (see references/skills-best-practices.md)
• [ ] Referenced resources exist or will be created
• [ ] If plugin: skills scope exists in .claude/.prove.json

Use AskUserQuestion with header "Review" to confirm: "Create Skill" / "Revise".

4. Create Associated Command (Optional)

If the skill should be user-invocable via /prove:<name>, offer to create a thin command wrapper:

---
description: <same as skill description, shortened>
argument-hint: "[input]"
---

# Command Name: $ARGUMENTS

Load and follow the <name> skill (`skills/<name>/SKILL.md` from the workflow plugin).

Committing

Delegate to the commit skill. Example: feat(skills): add migration-planner skill

Resources

• references/skills-best-practices.md — skill structure, description tuning, resource bundling, token budgets
• references/creator-conventions.md — shared creator workflow patterns
• references/prompt-engineering-guide.md — prompting techniques for the skill body