When AI says ‘done,’ do you still have to fix the work?
Designing AI Skills That Work Like a Thoughtful Partner · Skill design · Persona · Verification
Design a repeatable skill with a procedure, working attitude, context, and checks, then test its behavior in seven realistic situations.
Full-color paperback · 106 pages · English
What this book helps you do
Build the technical, persona, context, and verification layers together, then test how the skill behaves when information is missing or something goes wrong.
What you can work through
- Use the design card to define the assigned work, intended user, inputs, judgment criteria, and finished result.
- Test the skill against seven situations, including a false premise, tool limits, competing priorities, and error correction.
- Assemble the public Verifier skill from the complete core file contents in the practical appendix.
Put it to work
1. Pick one task you give AI repeatedly and complete the design card.
2. Write each layer's criteria and run the seven test prompts.
3. Revise the layer behind a poor result, then retest with the same input.
Who it's for
For practitioners whose long instructions still produce inconsistent work, and for anyone building AI roles or reusable skills.
Reading notes
Installing and running the skill requires an AI environment that supports the necessary files, tools, and permissions. The design card and test prompts can be used separately.