I Let Claude Code Control My Logged-In Chrome (Setup Guide)
A beginner's guide to letting your AI agent work in your own logged-in Chrome with free Chrome Use: four setup steps, all 8 prompts, and the honest account risks.
Stop briefing your AI from scratch. One free folder holds your background, your rules and your voice. Eight prompts build it, no code, and any AI can read it.
| Step | What it's for | Link |
|---|---|---|
| 1 | The free starter folder: eight prompts, the standard, sample answers | awp-playbook-knowledgebase-skeleton |
| 1 | Git, so the repo can be cloned | git-scm.com/downloads |
| 1 | The first prompt (setup) | steps/00-setup/prompt.md |
| 2 | All eight prompts, one per step | steps/ |
| 2 | The standard your AI reads while it builds (66 files) | reference/awp-knowledge-management-standard |
| 3 | A finished knowledge base to compare against (104 Markdown files) | solutions/final |
| 4 | The naming rule, the first fix most folders need | kb-naming-segment-convention.md |
| 5 | What moves with you when you switch AI tools | AI Memory: What You Typed Moves |
| All | The full 26-minute build, from an empty folder | The Simplest AI Second Brain: 8 Folders + Claude Code |
Every time you ask an AI for something real, you explain who it's for, how you want it to sound and what to avoid. Write that down once, in a folder of plain-text files, and hand the AI the files that matter before the task. The folder is yours, so it comes with you to whatever AI you use next.

Start your AI in the folder where you keep your projects and paste this. It's the Quick Start prompt from the repo's README:
Working directory: the folder where you keep your projects — the clone lands inside it.
Clone https://github.com/aiworkflowpro/awp-playbook-knowledgebase-skeleton and cd into it.
Run pwd and confirm it ends with awp-playbook-knowledgebase-skeleton.
Read AGENTS.md, then follow steps/00-setup/prompt.md.
Tell me when it is done, and do not start step 1 until I paste it.
Or do it by hand, then start your AI inside the repo:
git clone https://github.com/aiworkflowpro/awp-playbook-knowledgebase-skeleton
cd awp-playbook-knowledgebase-skeleton
ls steps
What you get:
steps/: the eight prompts. Read-only.reference/: the standard the prompts point to. Read-only.solutions/: a sample answer for each step, plus a finished knowledge base. Read-only.workspace/: your knowledge base. It starts empty, and git ignores what you put in it, so nothing personal gets committed back.Open each steps/NN-.../prompt.md, copy everything below the line, and paste it into your AI. Do them in order.
| Prompt | What the AI builds |
|---|---|
| 0 · Setup | The working folder and an empty workspace/ |
| 1 · The Skeleton | Eight folders, each with a CLAUDE.md signpost |
| 2 · The Person | Who you are and how you sound |
| 3 · The Business | Where your work lands: platforms and products |
| 4 · The Rules | Your first rules: naming and quality |
| 5 · The Workflows | Your first pipeline, run once for real |
| 6 · The Knowledge | Research topics, organized so they can be found again |
| 7 · The Engine | Dashboard, inbox, four imported sample files, and the standard copied into your folder |
Each prompt works the same way. The AI asks one question at a time, you answer in your own words, it writes the files, then prints the folder tree and waits. Read what it wrote before you paste the next prompt. That pause is where you catch the wrong guess.

From the full build video (7:30): the AI asks one question at a time, and you answer in your own words, here by voice.
Step 1 starts with three questions:
Each step can run in a fresh session. The AI doesn't need to remember the last one, because the first thing it reads is workspace/CLAUDE.md, the index of everything already built.
After step 7, compare your top two folder levels with solutions/07-engine/tree.txt, then run the naming check from the repo root. No output means every file follows the rule:
find workspace/research workspace/dashboard -type f -name '*.md' ! -name 'CLAUDE.md' \
| grep -vE '/([a-z0-9]+-)+[a-z0-9]+\.md$'
| Folder | What it holds |
|---|---|
owner/ |
Who you are: background, skills, decisions, how you communicate |
brand/ |
How you sound: identity, positioning, expression |
business/ |
Where your work lands: products, services, operations data |
specs/ |
Your rules: naming, writing conventions |
workflows/ |
Your pipelines: creation, publishing, research |
research/ |
What you know: topics, references, analysis |
dashboard/ |
Run results: task output and status snapshots, by month |
inbox/ |
New files land here, then get sorted into the right folder |

Six areas are shared by everyone: your background, your decisions, your voice, your rules, and what flows in and out. The rest take shape around your work. A lawyer's fill up with case files. An accountant's fill up with client records. I make content, so mine holds channels and posts.
Your folder won't look like solutions/final, and it shouldn't. Every file comes from your answers. Compare the structure, not the words.
The eight prompts are only the start. Mine began as eight empty folders. Then real work pushed on it: my file names became a mess, so I wrote a naming rule. A workflow failed, so I rewrote it and kept the better version. Those fixes grew into 32 standards and 96 workflows, and the folder now holds about 800,000 files. None of that was designed on day one. You don't need anything close to it. You need enough context for the next job.
When something works, save it in the shape the repo already uses. A rule goes in specs/:
# [Title]
## Scope
What this rule applies to.
## Rule
The constraint, in one clear sentence.
## Good example
One file name or snippet that follows the rule.
## Bad example
The same thing done wrong.
## Check
How the agent verifies it: what to look for, or what command to run.
A process that worked goes in workflows/:
# [Workflow Name]
## Trigger
When it runs: a schedule, a command, or an event.
## Steps
Numbered actions the agent can execute, in order.
## Check
How to tell the output is good. Point to a rule in specs/ if one applies.
## Lands in
Which folder the finished output goes to.
Then add a row for the new file to that folder's CLAUDE.md. A file that isn't indexed is one the next session will never find.

From the full build video (19:10): describe a job you repeat, and the agent writes it up as a workflow.
Take a video script, the example from the video. Normally you'd explain who it's for, how you want to sound, and what to avoid. With the folder, that's already written down. Give the AI the relevant files, then the new task.
In an AI that can read files, start it in the repo and point it at the index:
Read workspace/CLAUDE.md, then the files it lists for writing a video script:
who I am, how I sound, who I'm writing for, and my script rules.
Task: write a script for next week's video about <topic>.
In a chat app, attach the same files yourself. In the sample knowledge base those are owner/profile.md, brand/identity.md, brand/audience.md and specs/youtube-script-standard.md. Yours will have different names.
Switch AI tools and the same files come with you. They're plain text: you can read them, change them and back them up. The model is rented. The tools are rented. The folder is yours.

From the full build video (6:30): another agent and another model, the same folder.
specs/. Step 4 has you write one early, and the repo's naming convention is the reference.workflows/, so the next run starts from what worked.pwd, then cd into awp-playbook-knowledgebase-skeleton. Every path in the prompts is relative to that folder.tree: command not found. Install it (brew install tree on macOS, apt install tree on Debian or Ubuntu), or ask the AI to list the folder with find workspace -maxdepth 2.course/ and knowledge-base/ are now steps/ and workspace/, and the prompts in the repo already use the new paths.
Using the folder, yes. The files are plain text, so you can attach the ones a task needs to any of them. Building it with the eight prompts is different: the prompts clone a repo and write files, so they need an AI that can read and write files on your computer, such as Claude Code, Codex, Cursor or Gemini CLI.
No. You paste a prompt, answer questions in plain words, and the AI writes the files. Your job is to read what it wrote and fix anything that is wrong.
Much smaller than you think. Mine holds about 800,000 files today, but it started as eight empty folders. You need enough context for the next job, nothing more.
Built-in memory lives on the vendor's servers, shows you a short summary, and works in that one product. This folder lives on your disk, every file is one you can open and edit, and any AI that can read files can use it.

The useful part: a better instruction, a rule, or a process that worked. Rules go in specs/, processes go in workflows/, and each new file gets a row in its folder's CLAUDE.md so the next session can find it.
The skeleton, the prompts and the standards are free on GitHub. The prose is licensed CC BY 4.0, and the prompts, AGENTS.md, templates and workflow files are MIT. Whatever AI tool you run them in has its own pricing.
When I figure out how to get something real done with AI, you get the write-up — every step, including the parts that didn't work.