Skill Builder

When to Use

Trigger when:

• Charlie says "create a skill for X", "make a skill that does Y", or "turn this into a skill"
• An agent identifies a repeatable workflow worth codifying (via video-to-skill or organically)
• Charlie describes a process that should be reusable

Process

Step 1: Gather Context

Ask ONE clarifying question max if the intent is unclear. Otherwise infer from context:

• What triggers it? (keyword, link type, command, situation)
• What does it do? (steps, tools needed, output format)
• Where does output go? (Telegram thread, file, silent save)

Step 2: Generate SKILL.md

Follow this exact structure:

---
name: {slug-name}
description: {one-line description — used for skill matching}
---

# {Skill Name}

## When to Use
{Trigger conditions — be specific}

## Process
{Step-by-step behavior}

## Output Format
{Expected output structure or template}

## What to Avoid
{Anti-patterns, common mistakes}

Rules:

• Name must be a slug (lowercase, hyphens, no spaces)
• Description must be under 120 characters and specific enough for auto-matching
• Include a real example of expected output when possible (this is the "visual context" layer)
• Never include placeholder text like "TODO" or "fill this in"

Step 3: Install

1. Create the directory: ~/.openclaw/workspace/skills/{slug-name}/
2. Write the SKILL.md file
3. Register in openclaw.json under skills.entries:

   "{slug-name}": { "enabled": true }

4. Confirm to Charlie: skill name, trigger, and one-line summary

Step 4: Update HQ Skill Prompt Library (MANDATORY)

Every new skill must also be added to Charlie's prompt library so he does not need to remember skill names.

Prompt library location:

• Local: ~/.openclaw/workspace/docs/skill-prompt-library.md
• GitHub repo: carlosbuelvas1/nohaybreak-hq-openclaw (private)

Add a new section or update the relevant existing section with:

• Skill slug
• Plain-English use case
• 2-3 copy/paste trigger prompts
• Best-for list

Keep prompts practical and in Charlie's voice. Do not add fake capabilities the skill cannot perform.

Step 5: Push Skill to GitHub (MANDATORY)

This step is required. The repo is: carlosbuelvas1/openclaw-skills (private). Skills directory has git initialized at ~/.openclaw/workspace/skills/.

1. cd ~/.openclaw/workspace/skills
2. git add {slug-name}/
3. git commit -m "feat(skill): add {slug-name} — {one-line description}"
4. git push origin main
5. Verify the push: run gh api repos/carlosbuelvas1/openclaw-skills/contents/{slug-name} --jq '.[].name'
6. If verification fails, report the error to Charlie. Do not claim success without confirmation.

Step 6: Push Prompt Library Update to GitHub (MANDATORY)

This step is required when a new skill is created or an existing skill's trigger changes.

1. cd ~/.openclaw/workspace
2. git add docs/skill-prompt-library.md
3. git commit -m "Update skill prompt library for {slug-name}"
4. git push origin main
5. Verify the push: run gh api repos/carlosbuelvas1/nohaybreak-hq-openclaw/contents/docs/skill-prompt-library.md --jq '.name'
6. If verification fails, report the error to Charlie. Do not claim success without confirmation.

Model Requirement

This skill MUST run on openai-codex/gpt-5.5 (OAuth) to ensure exec/git tools work reliably. If running on a model without exec access, surface the skill draft to Charlie and request manual push.

Output Format

When skill creation is complete, report:

✅ Skill created: {slug-name}
Trigger: {when it fires}
Does: {one-line what}
Location: ~/.openclaw/workspace/skills/{slug-name}/SKILL.md
Skill GitHub: {pushed / skipped}
Prompt Library: {updated + pushed / skipped}

What to Avoid

• Don't create skills for one-off tasks that won't repeat
• Don't duplicate existing skills — check the skills directory first
• Don't create skills with vague triggers like "when appropriate"
• Don't add dependencies that aren't already available in the workspace
• Don't create the skill without registering it in openclaw.json
• Don't create or change a skill without updating docs/skill-prompt-library.md
• Don't claim the prompt library is saved unless the GitHub verification passed