The 4-File AI Agent Setup That Keeps Every Client Install Clean

Why most AI agent setups fail

The typical AI agent install comes with a folder full of documentation — 20 or more files covering every edge case, process, and client preference. None of it gets read. The agent sounds generic on every run because the context is buried under complexity it never surfaces.

The fix isn't better documentation. It's less of it. Four files, each with one job, all under 300 lines total. The agent reads them every run. The human reviews them every quarter. When something changes, you update one file. That's the system.

Hypd uses this structure as the starting point for every AI agent deployment in client work.

The four files and what each one does

Each file has a single word that describes its job. If you can't name a file's job in one word, the file is wrong.

• CLIENT.md — Context: who the client is, their vocabulary, their named systems, their business-specific rules. Under 100 lines. Filled in on day one. Never grows large.
• PLAYBOOK.md — Workflow: every automated workflow written in verbs only. Each one: trigger, agent steps, escalation condition, human approval point. Four lines per workflow, under 80 lines total.
• RULES.md — Audit: what the agent must never do. Universal rules plus client-specific restrictions. The compliance file.
• JOURNAL.md — Memory: what the agent has learned over time. The only file that grows. Updated after each run with corrections, preferences, and edge cases.

What goes in each file

CLIENT.md is the context that turns a generic AI response into something that sounds like a senior employee who knows the client. It contains the client's business name, industry, vocabulary they actually use, named systems and tools, and any rules specific to their business. If CLIENT.md grows beyond 100 lines, it's becoming documentation — cut it back.

PLAYBOOK.md is the workflow engine. Each workflow follows the same four-line format: trigger, agent steps, escalate-if condition, human-approves point. A clinic intake reads: 'Greet. Identify caller. Ask date of birth. Confirm insurance. Escalate if unclear.' Built during the first-week audit, reviewed quarterly.

RULES.md is the guardrail. Two sections: universal rules that apply to every install, and client-specific rules that reflect their legal, compliance, or brand requirements.

JOURNAL.md starts empty and earns its content. After each run, the agent adds what it learned — a correction, a preference, a situation it hadn't seen before.

Claude Code prompt to generate your four files

Use this prompt inside Claude Code to generate a customized version of all four files for a new client.

promptCopy
I'm setting up a 4-file AI agent configuration for a new client. Help me build all four files.

Here's what I know about this client:
- Business name and industry: [fill in]
- What they sell or deliver: [fill in]
- Tools and systems they use: [fill in]
- Key vocabulary or terms specific to their business: [fill in]
- The main workflows we're automating: [fill in — e.g. lead follow-up, appointment booking, report generation]
- Things the agent must never do: [fill in]

Generate:
1. CLIENT.md — under 100 lines, focused on context Claude needs every run
2. PLAYBOOK.md — each workflow in 4-line trigger/steps/escalate/approve format, under 80 lines total
3. RULES.md — universal rules section + client-specific restrictions
4. JOURNAL.md — empty template with the right headings for logging what the agent learns

Format each as a separate code block I can copy into individual files.

The maintenance rhythm

The system only works if the files stay current. CLIENT.md and RULES.md change when the client changes something — update them the same day. PLAYBOOK.md gets reviewed quarterly. JOURNAL.md grows on its own after each run.

• Day one: fill in CLIENT.md and RULES.md with the client, run PLAYBOOK.md prompt for the workflow audit
• Week one: refine PLAYBOOK.md as edge cases surface during real runs
• Ongoing: JOURNAL.md updates automatically after each agent run
• Quarterly: review all four files, cut stale content, update changed workflows

Frequently asked questions

• Why only four files? More files means more context the agent has to search through — and more documentation that goes stale without anyone noticing. Four files with clear jobs means every run starts with accurate, readable context.
• What if a client has complex workflows that don't fit four lines each? Break complex workflows into smaller sub-workflows, each with its own four-line entry. If a workflow genuinely needs more than four lines, it's doing more than one thing — split it.
• Can this system work with agents other than Claude Code? Yes. The file structure is model-agnostic. The same four files work as context for any AI agent that reads files at the start of a session — Cursor, Codex, or any custom agent built on a foundation model.