/create-skill

name: create-skill
description: >
  Turn a repeatable AI prompt or workflow into a structured, shareable skill for the
  affiliate-skills GitHub repository. Use this skill when the user wants to create a new skill, write a
  SKILL.md, convert a prompt to a skill, share a skill via the GitHub repo, or document
  an AI workflow.
  Also trigger for: "create a skill", "write a skill", "make this a skill",
  "turn this into a skill", "publish skill", "add skill to list", "write SKILL.md",
  "skill from prompt", "document this workflow", "package this as a skill".
license: MIT
version: "1.0.0"
tags: ["meta", "skill-writing", "directory", "publishing", "workflow", "prompt-engineering"]
compatibility: "Claude Code, ChatGPT, Gemini CLI, Cursor, Windsurf, OpenClaw, any AI agent"
metadata:
  author: affitor
  version: "1.0"
  stage: S8-Meta

List Affitor Skill

Turn a repeatable AI prompt or workflow into a structured, shareable skill for the [affiliate-skills GitHub repository](https://github.com/Affitor/affiliate-skills). The output is a complete SKILL.md file that works in any AI agent — shared via `npx skills add Affitor/affiliate-skills` so anyone can install it.

Stage

This skill belongs to Stage S8: Meta

When to Use

• User has a prompt they keep reusing and wants to turn it into a shareable skill
• User wants to create a new skill for the affiliate-skills repository
• User wants to write a SKILL.md file in the standard format
• User says "make this a skill" or "write a skill for X"
• User wants to package an AI workflow so others can replicate it

Input Schema

{
  raw_prompt: string       # (required) The prompt, workflow description, or detailed explanation of what the skill does
  failure_modes: string    # (optional) What goes wrong when the output is bad — helps write better Instructions and Error Handling
  niche: string            # (optional) Category hint, e.g., "content", "research", "seo"
  examples: string         # (optional) Example input/output pairs the user already has
}

Workflow

Step 1: Understand What the Prompt Actually Does

Before writing anything, analyze the user's raw prompt or workflow description:

1. **Task type** — Is this content creation, research, analysis, planning, automation, or something else? 2. **Variable inputs** — What changes each time? (product name, URL, audience, topic, etc.) 3. **Fixed structure** — What stays the same? (output format, sections, tone, constraints) 4. **Quality differentiator** — What makes a good output vs. a bad one? 5. **Failure modes** — Where does the AI tend to go wrong without explicit guidance?

If the user gave a vague description instead of an actual prompt, ask:

• "What do you typically paste into ChatGPT/Claude for this?"
• "What does the output look like when it works well?"
• "What goes wrong when it doesn't?"

If the user says "just do it", infer from context and proceed.

Step 2: Determine Skill Metadata

Based on the analysis, determine:

| Field | How to decide | |-------|--------------| | `name` | Short, action-oriented. "Comparison Post Writer" not "A Skill for Writing Comparison Posts" | | `slug` | kebab-case of name, e.g., `comparison-post-writer` | | `category` | One of: research, content, seo, landing, distribution, analytics, automation, meta | | `level` | beginner (1-step, no tools), intermediate (multi-step, 1 tool), advanced (complex workflow, multiple tools) | | `stage` | S1-Research, S2-Content, S3-Blog, S4-Landing, S5-Distribution, S6-Analytics, S7-Automation, S8-Meta | | `tags` | 3-6 lowercase tags relevant to the skill's domain | | `tools` | What external tools the skill needs: `web_search`, `web_fetch`, `code_execution`, none |

Step 3: Write the SKILL.md

Create a complete SKILL.md following this exact structure. Every section is required.

**Frontmatter (YAML)**

---
name: [slug]
description: >
  [2-3 lines. First line: what it does. Second line: trigger phrases.
  This is used for skill discovery — be specific about use cases.]
license: MIT
version: "1.0.0"
tags: [relevant tags]
compatibility: "Claude Code, ChatGPT, Gemini CLI, Cursor, Windsurf, OpenClaw, any AI agent"
metadata:
  author: [user handle or "affitor"]
  version: "1.0"
  stage: [S1-S8]
---

**Title and Introduction** One paragraph. What the skill does and what makes the output reliable. No marketing speak.

**When to Use** 3-5 specific trigger scenarios. "Writing a blog post" is too vague. "You need to publish a comparison post for two competing SaaS tools this week" is useful.

**Input Schema** Typed definition of every variable input. Mark required vs optional.

**Workflow (numbered steps)** This is the core. Each step must be concrete enough that any AI model produces consistent output:

• **Action** — what to do
• **Approach** — how to do it specifically
• **Quality bar** — what good looks like

Bad: "3. Write the pros and cons" Good: "3. Write at least 3 pros and 2 cons. Each must reference a specific feature, not a vague category. 'Exports to 12 formats including PDF and DOCX' not 'Great export options'."

**Output Schema** Typed fields that other skills can consume via conversation context. Include `output_schema_version: "1.0.0"`.

**Output Format** A markdown code block showing the exact template with `[placeholder]` brackets. This is the single most important section for consistency.

**Error Handling** 3-5 named failure modes with specific recovery behavior. What happens when input is missing, ambiguous, or the task can't be completed?

**Examples** 2-3 concrete examples showing:

• User input
• Key decisions made during the workflow
• What the output looks like (excerpt, not full)

**Flywheel Connections**

• Feeds Into: which skills consume this skill's output
• Fed By: which skills produce input for this skill
• Feedback Loop: how community engagement improves the skill
• `chain_metadata` YAML block with `skill_slug`, `stage`, `timestamp`, `suggested_next`

**Qual