SkillSlap Skill Publisher

Purpose: Turn a one-line skill idea into a fully-written, published SkillSlap skill in one shot. This skill writes the content, chooses the right metadata, and calls publish_skill via the MCP server. A skill that builds skills.

Invocation

/publish-skill <idea>

Examples:

• /publish-skill A skill that writes Tailwind component variants from a design description
• /publish-skill An agent that reviews OpenAPI specs for REST conventions and breaking changes
• /publish-skill Context rules for writing Rust async code with Tokio

Prerequisite: The MCP server must be running with SKILLSLAP_TOKEN set to the author's API token.

Process

Step 1: Analyze the idea

From the one-line description, determine:

• What problem does this skill solve? (be specific)
• Who invokes it? → sets invocation_type:
  • agent — a Claude Code agent calls it as a slash command or task
  • context — injected into the agent's context/CLAUDE.md as rules
  • tool — the agent calls it as a tool in a pipeline (reads input, returns output)
• What does the user provide? (inputs)
• What does the skill produce? (outputs)
• What are the 3–5 best tags?

Step 2: Write the skill content

Follow this structure for agent and tool skills:

# [Skill Title]

> **Purpose:** [One crisp sentence: what it does and why it's useful]

---

## Invocation

/[command] <required-arg> [optional-arg]

**Example:** `/[command] [concrete example]`

---

## Process

[2–5 phases, each with a clear heading and bullet points]
[Use code blocks for examples of inputs and outputs]
[Be specific — no "analyze the X" without saying HOW]

---

## Output Format

[Describe exactly what the skill produces: file, table, list, etc.]
[Include an example output snippet if useful]

For context skills (rules to inject):

# [Title]

> **Purpose:** [What these rules enforce and why]

---

## Rules

### [Rule Category 1]
- [Specific rule]
- [Specific rule]

### [Rule Category 2]
- [Specific rule]

Step 3: Choose metadata

Step 4: Publish via MCP

Call the publish_skill tool:

{
  "title": "[chosen title]",
  "content": "[full markdown content]",
  "description": "[short description]",
  "tags": ["tag1", "tag2", "tag3"],
  "invocation_type": "[agent|context|tool]",
  "version": "1.0.0",
  "status": "draft"
}

Step 5: Confirm and share

After publishing, output:

• The skill's ID and URL: https://skillslap.com/skill/[id]
• A summary of what was written
• Next steps: "Review at the URL above, then set status to 'active' when ready"

Quality Bar

Before calling publish_skill, verify:

• Invocation section has a real example (not <placeholder>)
• Process section says HOW, not just WHAT (no "review the code" without specifics)
• Output format is described with an example snippet
• Tags are lowercase and specific (no "code", use "typescript" or "python")
• Description fits in 200 chars
• Status is draft — never publish as active without review

Error Handling

If publish_skill returns an auth error:

"The SKILLSLAP_TOKEN env var is not set or is invalid. Add it to your MCP config: SKILLSLAP_TOKEN=your-token-here"

If the title already exists:

"A skill with this title already exists in your account. Update the title or use update_skill with the existing skill's ID."