skill-creator

name	skill-creator
description	Generate a new skill file and command routing file for this repo from a plain-language description of what the skill should do.

skill-creator — Build New Skills Fast

Takes a plain-language description of a task and generates a ready-to-ship skill: the SKILL.md, the command routing file, and the CLAUDE.md additions needed to wire it in.

When to Use

• "I want a skill that does X"
• "Build me a slash command for Y"
• "Add a skill for [recurring task I do]"

Before You Start

Read:

• docs/agent-template.md — canonical file format and pre-ship checklist
• CLAUDE.md — current routing, so the new skill doesn't duplicate an existing one
• The most relevant existing skill in writing/skills/ or projects/*/skills/ — for style reference

Instructions

1. Clarify the skill

Ask (or infer from context):

• Name: what should the slash command be called? (lowercase, hyphenated)
• Trigger: when should this skill activate? What does the user say or ask?
• Input: what does the skill need to work? (a link, a transcript, a filename, etc.)
• Output: what does it produce? (a draft, a log entry, a summary, a file?)
• Tools needed: Read, Write, Edit, Bash, Glob, WebFetch?

2. Draft the SKILL.md

Follow the template in docs/agent-template.md exactly. Include:

• YAML frontmatter with name and description (60+ chars, specific enough that Claude can decide whether the skill is relevant from the description alone)
• Clear "When to Use" section
• Step-by-step instructions
• Output format spec, if the skill produces something

Place it at one of:

• writing/skills/<name>/SKILL.md — writing/content skills
• projects/<project>/skills/<name>/SKILL.md — project-scoped skills
• skills/<name>/SKILL.md — general/meta skills (like this one)

3. Draft the command routing file

Short file in commands/<name>.md following the template in docs/agent-template.md. Frontmatter: name, description, allowed-tools. Body: one or two lines that load the SKILL.md.

4. Draft the CLAUDE.md additions

Two additions:

1. A row in the slash command table: | /<name> | <brief description> |
2. A routing rule in the appropriate section of "When to Load Additional Context" (if applicable)

5. Pre-Ship Checklist

Before presenting, verify the new skill satisfies all of these:

• Frontmatter description is specific (60+ chars) — vague descriptions degrade routing
• MCP calls (if any) have explicit limits — see docs/mcp-efficiency.md
• Confirmation gate on any action that changes external state — see docs/safety-contract.md
• Writing skills load writing/skills/avoid-ai-writing/SKILL.md when producing external content
• Memory writes follow docs/auto-memory.md (typed entries, MEMORY.md cap)
• Command stub created in commands/
• CLAUDE.md slash command table updated
• CHANGELOG.md entry added under [Unreleased]
• Runs cleanly via bash scripts/validate-skills.sh

6. Present for review

Show all three artifacts in a single response. Ask: "Want me to commit these, or make any changes first?"

7. On approval: commit and update CHANGELOG

Create the files, run bash scripts/validate-skills.sh, then update CHANGELOG.md with a new entry listing all added files.

Output Format

Present each file in a labeled fenced code block:

**skills/<name>/SKILL.md**
```markdown
[content]
```

**commands/<name>.md**
```markdown
[content]
```

**CLAUDE.md additions**
```markdown
[the two snippets to add]
```