00-build-alignment-skill

npx skills add majiayu000/claude-skill-registry --skill 00-build-alignment-skill --agent claude-code

Same command for any agent — swap --agent for codex, cursor, copilot.

# Build Your Alignment Skill You have a merged model that combines persona and agentic capabilities. But models that complete tasks eagerly can also complete harmful requests eagerly. Before learning alignment theory, you'll build a skill that encodes the knowledge you need. This lesson follows the pattern you've used throughout Part 8: create the skill first from official documentation, then use the chapter to test and improve it. By the end of this chapter, your `model-alignment` skill will be battle-tested and production-ready. ## Why Alignment Needs a Skill Alignment is not a one-time fix. It's an ongoing discipline with evolving techniques, common failure modes, and nuanced tradeoffs. Memorizing this is impossible. Encoding it into a skill makes the knowledge permanently accessible. | Traditional Approach | Skill-First Approach | |---------------------|---------------------| | Read papers, forget details | Build skill, query anytime | | Scattered alignment notes | Single authoritative source | | Reinvent safety patterns | Reuse validated approaches | | Cannot delegate to AI | AI executes your skill | Your merged Task API model from Chapter 67 needs alignment before deployment.

Clone the skills-lab repository
git clone https://github.com/panaversity/skills-lab.git ~/skills-lab-ch68
Navigate to the directory
cd ~/skills-lab-ch68
Create the skill directory structure
mkdir -p .claude/skills/model-alignment
cat > LEARNING-SPEC.md << 'EOF'
cat > .claude/skills/model-alignment/SKILL.md << 'EOF'
Check the skill exists
ls -la .claude/skills/model-alignment/