Route the job
Classify this task for Chat, ChatGPT Work, Codex, or a specialised AI system: [TASK]. Recommend one, explain why, and name the evidence and verification the task needs.
192 reusable prompts
The library covers sixteen application categories. Each prompt makes the outcome, inputs, constraints, and checks more visible.
Application categories
Sample prompts
Replace the square brackets with your own material. Treat every result as work to inspect.
Route the job
Classify this task for Chat, ChatGPT Work, Codex, or a specialised AI system: [TASK]. Recommend one, explain why, and name the evidence and verification the task needs.
Build a reliable brief
Turn this request into a BRIEF work order with background, result, inputs, execution constraints, and final checks: [REQUEST]. Mark every missing input instead of inventing it.
Inspect a result
Review this output against the stated goal, evidence, constraints, and acceptance tests. Separate confirmed facts, assumptions, weaknesses, and the next correction.
Use Codex safely
Inspect the project before editing. Reproduce the issue, identify the smallest safe change, implement it, run relevant tests, and report the files changed and remaining uncertainty.
Design a KAI-style system
Design a specialised assistant for [AUDIENCE] with one bounded job, approved source truth, prohibited claims, privacy rules, human handoff, evaluation questions, and a maintenance owner.
Preserve what works
Turn the approved result and corrections into a reusable template, checklist, instruction file, or test. Include an owner and the event that should trigger review.
More than templates
The complete library is supported by examples, evaluation questions, case studies, exercises, and the Operator Loop.
View the book