Level 3: Skill Engineering
Prerequisites: Level 2: Context Engineering Goal: Build reusable, discoverable procedural knowledge that agents can invoke
Why Skills Exist
Skills solve a fundamental problem: how do you give an agent repeatable, high-quality procedures without rebuilding them from scratch in every prompt?
A skill is a document that encodes HOW to do something — not what to do, not why to do it, but the exact step-by-step procedure for completing a specific type of task reliably.
In Claude Code, skills live in .claude/skills/ and are automatically discovered. In other systems, they are injected as context when relevant tasks are detected.
Skills vs. Prompts vs. Agents
| Concept | What It Is | When to Use |
|---|---|---|
| Prompt | A structured instruction for one task | One-off tasks |
| Skill | A reusable procedure for a task category | Repeated task types |
| Agent | An autonomous actor that uses skills | Complex, multi-step goals |
Rule: If you write the same prompt more than twice, extract it as a skill.
Skill Structure (The OAIES Standard)
Every skill file follows this exact structure:
# [Skill Name] Skill
## Trigger Conditions
[When this skill should activate — specific patterns, keywords, or task types]
## Purpose
[One sentence: what this skill does]
## Prerequisites
[What must be true before this skill can execute]
## Procedure
[Numbered, imperative steps]
## Verification
[How to confirm the skill executed correctly]
## Common Failures
[What goes wrong and how to recover]
## Examples
[1-2 concrete examples of this skill in action]
## Do Not Use This Skill When
[Anti-trigger conditions]
Compatibility with Open AI Skills Specs (openaiskills.io)
The OAIES skill structure is designed to be fully compatible with emerging open standards for developer assistant skills:
- openaiskills.io specification: Standardized, portable markdown format for AI agent skills across the developer ecosystem.
- Claude Code (.claude/skills): Built directly to match Claude Code's auto-discovery skill scanner schema.
- Aider / Roo Code / Cline custom instructions: Readily consumable by modern coding agents as background runtime knowledge.
Skills Library
Core Skills
| Skill | Trigger | Purpose |
|---|---|---|
| architecture.skill.md | System design request | Design robust architectures |
| planning.skill.md | Feature or task kickoff | Create implementation plans |
| debug.skill.md | Bug report or error | Systematic debugging |
| code-review.skill.md | Review request | Comprehensive code review |
| documentation.skill.md | Doc generation request | Generate technical docs |
| security.skill.md | Security review request | Security audit |
| performance.skill.md | Performance concern | Performance analysis and fix |
| testing.skill.md | Test generation request | Generate comprehensive tests |
Language / Framework Skills
| Skill | Trigger |
|---|---|
| react.skill.md | React component work |
| nextjs.skill.md | Next.js routing, SSR, RSC |
| node.skill.md | Node.js API development |
| python.skill.md | Python development |
| java.skill.md | Java/Spring development |
| dotnet.skill.md | .NET/C# development |
Platform Skills
| Skill | Trigger |
|---|---|
| azure.skill.md | Azure cloud tasks |
| aws.skill.md | AWS cloud tasks |
| devops.skill.md | CI/CD, infrastructure |
AI-Specific Skills
| Skill | Trigger |
|---|---|
| prompt.skill.md | Prompt writing or review |
| context.skill.md | Context engineering |
| rag.skill.md | RAG system tasks |
| evaluation.skill.md | AI system evaluation |
| memory.skill.md | Memory system design |
| observability.skill.md | AI observability setup |
Quality Skills
| Skill | Trigger |
|---|---|
| accessibility.skill.md | Accessibility review/fix |
| seo.skill.md | SEO review/fix |
| data-engineering.skill.md | Data pipeline tasks |
Installing Skills (Claude Code)
# From OAIES
mkdir -p .claude/skills
cp oaies/skills/*.skill.md .claude/skills/
# Verify skills are discovered
# Claude Code reads .claude/skills/ automatically
Naming Conventions
{domain}.skill.md # Core skills
{technology}.skill.md # Technology-specific skills
{platform}.skill.md # Platform skills
{domain}-{specialization}.skill.md # Specialized variants
Writing a New Skill
- Identify the task pattern — it must recur at least 3x before warranting a skill
- Use the standard structure (above)
- Test with at least 2 different inputs
- Document trigger conditions precisely — vague triggers cause wrong skill selection
- Add to the skills index in this README
- Version with a changelog at the bottom of the file
Anti-Patterns
❌ Skill That's Too Broad
# Wrong — "development skill" triggers on everything
Trigger Conditions: Any development task
# Correct — specific triggers
Trigger Conditions:
- "create component", "build component", "new component"
- Request to implement a UI element
- React component generation tasks
❌ Skill Without Verification Step
If you cannot verify the skill succeeded, you cannot trust its output. Every skill must have a concrete verification procedure.
❌ Skill That Encodes One Team's Preferences
Skills become institutional knowledge. Write them to work for any team following OAIES — not for your specific team's idiosyncrasies.
Readiness Gate
Before proceeding to Level 4, verify:
- At least 5 skills installed for your primary technology stack
- All skills follow the standard structure
- Skills have specific, non-overlapping trigger conditions
- Skills are version-controlled
- At least one skill has been verified to improve output consistency