Capstone: Build Your Instruction Library

🔄 Seven lessons of techniques, platforms, templates, and troubleshooting. Now it’s time to put it all together. In this capstone, you’ll build something you’ll actually use every day: a personal instruction library.

What You’re Building

Your instruction library is a collection of tested, organized instruction sets — one for each major workflow in your life. Think of it like a toolbox: you don’t carry every tool to every job. You grab the right set for the task.

By the end of this lesson, you’ll have:

• A base layer of universal preferences
• 3+ workflow-specific instruction sets built from the templates in Lesson 6
• A maintenance schedule to keep everything current
• A testing checklist to verify each set works

Step 1: Define Your Base Layer

Your base layer contains preferences that apply to every conversation, regardless of task. These go in your platform’s global settings (ChatGPT Custom Instructions, Claude’s default behavior, etc.).

Write yours now. Pull from what you learned in Lessons 3-4:

COMMUNICATION PREFERENCES:
- [Your preferred response length]
- [Your preferred format — bullets, prose, mixed]
- [Your language/locale preference]

BEHAVIORAL RULES:
- [How you want the AI to handle uncertainty]
- [How you want feedback delivered]
- [Your "quick" keyword behavior]

NON-NEGOTIABLES:
- [Truth-telling policy]
- [Source citation policy]
- [Clarification behavior for ambiguous requests]

✅ Quick Check: Your base layer says “be concise” and your coding instruction set says “include detailed explanations with code.” Which rule wins? (The workflow-specific rule overrides the base layer for that context. Your base layer is the default; workflow sets are the exceptions. This is priority stacking from Lesson 4.)

Step 2: Build Your Workflow Sets

Pick your top 3 workflows. For each one, build an instruction set using this structure from Lessons 3-6:

Exercise: Build your first workflow set now. Pick your most common task — the thing you ask AI to do three or more times per week. Use the relevant template from Lesson 6 as your starting point. Customize the role, conventions, and examples to match your actual work.

Step 3: Test Each Set

Don’t trust untested instructions. For every instruction set you build, run this verification from Lesson 7:

1. Three-prompt test — Send three different types of requests within that workflow. Does the AI follow your instructions consistently across all three?
2. Edge case test — Send a request that falls outside the normal pattern. Do your exceptions handle it?
3. Conflict check — Read your base layer and workflow set together. Any contradictions?

If anything fails, use the troubleshooting protocol: Isolate → Check conflicts → Measure → Simplify → Test.

Step 4: Set Your Maintenance Schedule

Instructions aren’t write-once. Set calendar reminders:

Course Review

Here’s what you’ve learned, lesson by lesson:

Lesson 1 — Why Custom Instructions Matter: Custom instructions are system-level prompts that shape every conversation. MIT research shows structured prompts improve output quality by up to 50%.

Lesson 2 — How They Work: Instructions sit in the system message, processed before your input. Every platform has different limits and features. Instructions consume tokens from the context window.

Lesson 3 — RISEN Framework: Role, Instruction, Structure, Examples, Nuance. Specific beats vague. Positive framing beats negative. Examples teach more than rules.

Lesson 4 — Advanced Techniques: XML tags for structure. Conditional logic for multi-task handling. Priority stacking (Always/Usually/Sometimes) for flexible rules. Metacognitive instructions for better reasoning.

Lesson 5 — Platform Mastery: ChatGPT has three layers (Custom Instructions → Custom GPTs → Projects). Claude excels at XML-structured instructions with long context. Gemini integrates with Google’s ecosystem. Copilot adds adaptive memory.

Lesson 6 — Templates: Six ready-to-use templates (Developer, Business Writer, Data Analyst, Content Creator, Research Assistant, Learning Tutor). Every template needs exceptions to prevent rigidity.

Lesson 7 — Troubleshooting: Five common problems (vague, conflicting, drifting, over-constrained, wrong specificity). The 5-step protocol: Isolate → Check conflicts → Measure → Simplify → Test.

Your Instruction Library Checklist

Before you consider your library complete, verify:

• Base layer installed in your primary platform’s global settings
• At least 3 workflow-specific instruction sets written and tested
• Each set uses the RISEN structure (or the XML variant from Lesson 4)
• Each set includes exceptions/nuance to prevent rigidity
• Each set has been tested with 3+ different prompts
• No conflicts between base layer and workflow sets
• Maintenance reminders set (monthly + quarterly)
• Instructions saved in a document you can access and update (not just pasted into settings)

Key Takeaways

• Build a base layer (universal preferences) plus workflow-specific sets (task-focused instructions)
• Every instruction set needs testing — three different prompts minimum, plus edge cases
• Maintain your library with monthly fixes and quarterly full reviews
• Save your instructions in an editable document — platforms change, and you’ll want to update and transfer them
• Start with 3 workflow sets. Add more only when you have a recurring need — don’t over-build

What’s Next for You

You’ve completed the course. Here’s how to keep improving:

1. Use your library daily — The only way to refine instructions is real-world usage
2. Share what works — If you build a great instruction set for your team, share it. Custom GPTs make this easy
3. Watch for platform updates — AI platforms evolve fast. New features mean new possibilities for your instructions
4. Revisit this course — As you gain experience, the advanced techniques in Lessons 4-5 will click differently

Custom instructions are the highest-leverage AI skill you can develop. Every minute you invest in better instructions pays off across hundreds of future conversations. Your library is the compounding asset — it gets more valuable the more you use and refine it.