walid@portfolio:~/lab/ai-playbook-starter-kit$
cd../lab
04ideaSep 2026

Make one job repeatable

A folder, a ten-section playbook, and four prompts that build it, equip it, run it and grade it

The gap between a good one-off result and a system is that nobody wrote down what good looks like. This is one folder and one file that does: purpose, success criteria, rules, output format — plus the three sections most people skip, which are where the value actually is. Four prompts build the playbook, work out what it is missing, run it, then grade the result against it and say which of five things went wrong.

PromptsProcessVerificationTemplatesClaude Code
i
Five parts, and the last one is the one that matters

Playbook, context, tools, examples, verification. The first four are what most people assemble; the fifth is what turns it into something you can trust unattended. A workflow that produces output is easy. A workflow that checks its own output against a written standard, names what failed, and tells you which part of the standard caused it — that is the difference between a prompt you rerun and a process that improves.

Section 1

The five-minute setup

One folder. Everything the agent needs lives inside it, so every run starts from the same place instead of from whatever happens to be in the chat.

AI-Playbook/11 lines
AI-Playbook/
│
├── README.md          how to use the system
├── PLAYBOOK.md        source of truth for the job
│
├── 01_Templates/      reusable output formats
├── 02_Examples/       strong completed work
├── 03_Scripts/        prompts + supporting instructions
├── 04_Data/           approved reference material
├── 05_Tools/          docs for tools you actually have
└── 06_Output/         finished work

What goes where

The numbered prefixes are doing real work — they keep the order stable so the agent reads them the same way every time.

Folder or fileWhat it holds
PLAYBOOK.mdThe source of truth for how the job should be completed.
01_TemplatesReusable formats for outputs.
02_ExamplesExamples of strong completed work — guidance, explicitly not facts.
03_ScriptsUseful prompts, scripts and supporting instructions.
04_DataApproved reference material.
05_ToolsDocumentation for the APIs, servers and integrations you actually have.
06_OutputCompleted work.
README.mdHow to use the system — for the version of you who comes back in six weeks.
Section 2

The playbook template — ten sections

Fill the brackets by hand, or let the first prompt below interview you and fill them for you.

PLAYBOOK.md49 lines
# PLAYBOOK: [JOB NAME]

## 1. PURPOSE
What is this system supposed to accomplish?

## 2. SUCCESS CRITERIA
A successful output must:
- [CRITERIA]

## 3. INPUTS
Required: [INPUT]
Optional: [INPUT]

## 4. AVAILABLE RESOURCES
Templates: [FILES]   Examples: [FILES]
Data:      [FILES]   Tools:    [TOOLS]

## 5. PROCESS
STEP 1: [INSTRUCTION]
STEP 2: [INSTRUCTION]

## 6. RULES
ALWAYS: [RULE]
NEVER:  [RULE]

## 7. OUTPUT FORMAT
The final output should contain: [FORMAT]

## 8. VERIFICATION
Before returning the result, check:
[ ] Complete
[ ] Accurate
[ ] Supported by available information
[ ] Correct format
[ ] No invented information
[ ] All important constraints followed
[ ] Human review completed where required

## 9. FAILURE HANDLING
If information is missing:        STOP and ask for it.
If a tool fails:                  Report the failure.
If something cannot be verified:  write NOT VERIFIED.
Do not silently invent a replacement.

## 10. IMPROVEMENT LOG
WHAT HAPPENED:        [ ]
WHY:                  [ ]
PLAYBOOK CHANGE:      [ ]
EXPECTED IMPROVEMENT: [ ]
i
Sections 8, 9 and 10 are the ones worth stealing

The first seven are a competent brief, and plenty of people already write something like them. The last three are rarer. Verification gives the agent a standard to check itself against. Failure handling replaces the default behaviour — quietly inventing a plausible substitute — with three explicit alternatives: stop, report, or write NOT VERIFIED. And the improvement log is what stops the same failure recurring, because it forces the fix back into the playbook rather than into your memory of that one time it went wrong.

Prompt 1

Turn a job into a playbook

It interviews you before writing anything. The instruction not to invent missing business information — and to mark anything it assumes — is what keeps the output yours rather than generic.

prompt-1-build.txt30 lines
You are my AI Process Architect.

I want to turn this job into a repeatable AI playbook:

JOB: [DESCRIBE THE JOB]
GOAL: [GOAL]
CURRENT PROCESS: [DESCRIBE IT]
TOOLS: [TOOLS]
INPUTS: [INPUTS]
DESIRED OUTPUT: [OUTPUT]
EXAMPLES AVAILABLE: [EXAMPLES]

Do NOT create the playbook immediately. First interview me.

Ask only the questions necessary to understand:
1. The goal   2. The user/customer   3. Required inputs   4. Exact steps
5. Decision points   6. Available tools   7. Available reference material
8. Rules   9. Constraints   10. Expected output   11. Common mistakes
12. Verification requirements   13. What requires human approval

Ask questions in small groups.

When you have enough information, create PLAYBOOK.md using:
PURPOSE · SUCCESS CRITERIA · INPUTS · RESOURCES · PROCESS · RULES ·
OUTPUT FORMAT · VERIFICATION · FAILURE HANDLING · IMPROVEMENT LOG

Do not invent missing business information.
Mark assumptions: ASSUMPTION — VERIFY

Then show me the finished playbook before implementing anything.
Prompt 2

Work out what the playbook is missing

Run it after the first. The REQUIRES CONFIGURATION marker is the useful bit — it stops the agent assuming it has access to a tool you have not actually wired up.

prompt-2-toolbox.txt23 lines
You are my AI Context Architect.

Review this playbook: [PLAYBOOK OR FILE PATH]

Now determine what resources would make this workflow more reliable.
Classify resources into: TEMPLATES · EXAMPLES · SCRIPTS · DATA · TOOLS

For each resource tell me:
NAME · PURPOSE · WHY IT IS NEEDED · WHERE IT SHOULD LIVE ·
DO WE ALREADY HAVE IT? · HOW IT SHOULD BE USED

Do not create fake data or examples and present them as real.
If we need external information, identify the legitimate source required.
If a tool/API/integration is needed, do not assume access.
Mark it: REQUIRES CONFIGURATION

Then propose this structure:
01_Templates/  02_Examples/  03_Scripts/  04_Data/  05_Tools/

Finish with:
MISSING CONTEXT · MISSING TOOLS · MISSING EXAMPLES · MISSING DATA

Then give me the next 3 things I should add first.
Prompt 3

Run the job

The one you reuse every time — only the input line changes. Note that it deliberately does not return the result; it hands off to the verification stage.

prompt-3-execute.txt24 lines
You are executing an approved AI playbook.

PLAYBOOK: [PLAYBOOK FILE]
INPUT: [INPUT]
AVAILABLE RESOURCES: [RESOURCE FOLDERS]

Before beginning:
1. Read the entire playbook.
2. Inspect the available resources.
3. Confirm required inputs exist.
4. Confirm required tools are actually available.
5. Identify anything missing.

If a required input is missing: STOP. Do not guess.

If everything required is available:
Follow the playbook in order. Do not skip steps.
Use the supplied templates where appropriate.
Use approved examples as guidance, not facts.
Use only available/verified data.
Do not invent information to complete the task.

When finished, do NOT immediately return the result.
Move to the verification stage.
Prompt 4

Make it grade its own work

Eight checks, a structured pass/fail report, then fixes — and finally a diagnosis of which of five things actually caused the failure.

prompt-4-verify.txt38 lines
You are now the Verification Layer.

Review:  PLAYBOOK: [PLAYBOOK]   OUTPUT: [OUTPUT]
         SOURCE MATERIAL: [SOURCES]

Evaluate the output against the playbook. Check:
[ ] COMPLETE     Did it follow every required step?
[ ] ACCURATE     Are factual statements supported by available information?
[ ] SOURCED      Are required sources/references included?
[ ] FORMAT       Does it match the requested structure?
[ ] RULES        Were all constraints followed?
[ ] FABRICATION  Did the output invent any unsupported information?
[ ] TOOLS        Did it claim to use any tool or integration that was not
                 actually available?
[ ] HUMAN REVIEW Does anything require human approval before external use?

Return a VERIFICATION REPORT:
COMPLETE: PASS / FAIL
ACCURATE: PASS / FAIL / NOT VERIFIED
SOURCED: PASS / FAIL / N/A
FORMAT: PASS / FAIL
RULES: PASS / FAIL
FABRICATION CHECK: PASS / FAIL
HUMAN REVIEW: REQUIRED / NOT REQUIRED
ISSUES: [List]
FIXES: [List]

Then correct any fixable issues.

Afterward answer:
WHAT FAILED?  WHY DID IT FAIL?

WAS THE FAILURE:
A. EXECUTION ERROR  B. MISSING CONTEXT  C. MISSING TOOL
D. BAD INPUT  E. PLAYBOOK PROBLEM

If the PLAYBOOK caused the failure, recommend the smallest specific change
to PLAYBOOK.md. Do not rewrite the entire playbook unnecessarily.
The diagnosis

Five causes, and what each one means you fix

This taxonomy is the part that compounds. Without it every failure looks like “the AI got it wrong”, and you change the wrong thing.

CauseWhat you actually change
Execution errorNothing structural — rerun it. Only a pattern of these means something deeper.
Missing contextAdd the reference material to the data or examples folder. The playbook was fine; it was working blind.
Missing toolWire up the integration, or mark it as unavailable so the agent stops assuming it.
Bad inputFix upstream. The playbook cannot rescue an input that should have been rejected.
Playbook problemThe only case where you edit the playbook — and then only the smallest specific change, never a rewrite.
Section 7

Verification checklist — before any output leaves

Complete — every required step followed.Accurate — statements supported by available information.Sourced — required references included.Correct format — matches the template.No invented information, with unknowns explicitly marked NOT VERIFIED.All rules and constraints followed.No tool claimed that was not actually available.Human review completed where required.
i
How this sits next to the agent-backend entry

There is an entry elsewhere in this lab about building an agent backend for a whole business — workspace, brain, access, infrastructure. This is the smaller, sharper cousin: one job rather than a company, and its job file carries ten sections rather than seven. The three extra ones are verification, failure handling and the improvement log. If you only ever adopt one idea from either, adopt those — they are what make the difference between output you have to check and output that checks itself.

i
Provenance

The folder structure, the ten-section template, all four prompts and the verification checklist are as given in the source. The author’s byline, the pointer to their longer paid guide, and the closing follow prompt are not carried across. Nothing here needed external verification — there are no version numbers, star counts or pricing claims to check, which is unusual for this shelf and worth saying plainly rather than implying a check happened.