The AI Executive CookBook
Voice Kit starts with a real piece of your writing and the job you need done. It gives you a first draft you can inspect, with the choices it made visible—so the next version starts closer to you.
No email or install required to try it. The full Voice Kit page has the prompt, the explanation, and the repeat-work install when you want it.
Start here
The first release is deliberately narrow. Use Voice Kit now. Join the list for the next two when they have a human-readable breakdown, a usable install path, and an output you can judge.
Use one real sample and a current writing job to get a grounded first draft, visible pattern notes, and a Skill you can improve from the next draft.
Turn work you are allowed to use into source-linked proof cards that make the claim, scope, permission, and limit clear.
A practical setup guide for deciding what your AI needs, keeping credentials out of prompts, and testing access before real work.
The release rule
The recipe is the human-readable breakdown: the job, the use cases, the execution model, and the evidence. The Skill is the reusable instruction you give your AI. Both have to be clear before a release is ready to use.
The job, source material, and useful output are clear enough to try on one small piece of real work.
The human-readable breakdown shows the use cases, how a source becomes an output, where feedback improves the Skill, and what research does and does not support.
Then it gets an install instruction. The source file is for the AI; the breakdown is there so you can decide whether the skill belongs in your system.
In development
These are the next areas of work: relationships, proof, customer evidence, and repeatable AI workflows. None is ready to install. The release list gets the first version that has a clear job, a readable recipe, and a way to judge the output.
Find the warm path and the right next move from a selected set of real relationships.
Separate repeated customer evidence from the loudest request before deciding what changes.
See the patterns, proof gaps, and next test in your owned content.
Turn a small set of public signals into one credible contribution and a useful next conversation.
Turn one approved call into evidence, uncertainty, and a reviewable decision brief.
Plan careful, evidence-backed follow-up without turning relationships into a lead list.
Choose the few repeated jobs where a proven skill can help and keep the review gates visible.
Turn one proven Skill into a small repeatable work loop without unattended action.
Test whether a browser setup is reliable enough for a named safe task before relying on it.
Each one stays here until it has a clear job, a human-readable skill breakdown, and a real install path.
How it works
The Skill breakdown explains the job, the execution model, and the research behind it. The Skill gives your AI the reusable instructions. You add the current facts and decide whether the output is useful enough to use.
Voice Kit starts with the email, post, or explanation in front of you. It does not ask the AI to “make your business better.”
Use a talk, post, email, or transcript you own or are allowed to use. The skill uses the sample and current job together.
The AI drafts immediately and shows the observed patterns next to the output. You can see where the Skill helped and where it missed.
Your direct correction wins over an inferred pattern. Keep what reduces edits. Change what sounds wrong. Stop using any Skill that costs more judgment than it saves.
Where to use them
Voice Kit is a focused SKILL.md directory. Future releases will use the same format, with a human-readable skill breakdown alongside the install instruction.
OPENAIUse standalone Agent Skills in the desktop app, CLI, or IDE extension. Codex can invoke a named skill or choose one that fits the request.
Read Codex skill docs ↗
ANTHROPICUse custom skills in Claude, or keep focused skill folders alongside a Claude Code project.
Read Claude Code docs ↗
GOOGLEAgent Skills can be installed from a path or file and checked from the CLI before you use them.
Read Gemini CLI docs ↗
GOOGLEWorkspace and global Agent Skills use a SKILL.md folder. The agent loads the relevant skill when the work calls for it.
GITHUBAgent Skills work with Copilot cloud agent, CLI, and agent modes in supported IDEs.
Read Copilot docs ↗A SKILL.md file gives the AI focused instructions for one job. Your AI environment supplies any browser, account, and approval access. The skill itself does not grant authority to take a consequential action.
Get the useful ones
I’ll email you when Proof Library, Secure Setup, or another CookBook skill is ready to use. Every release has to clear the same bar before it lands in your inbox.
Platform compatibility: checked against current official documentation for OpenAI Codex, Claude Code, Gemini CLI, Google Antigravity, and GitHub Copilot. Checked: 2026-08-02.
