Client Setup Guide

How to create client profiles, brand overlays, and capabilities with ACR.

No coding required. If you can fill out a form, you can build these.

What You’re Building (Big Picture)

You build ONCE (generic):        You build PER CLIENT:
├── email-copywriting            ├── acme-brand (overlay)
├── campaign-strategy            ├── globex-brand (overlay)
├── red-team-review              ├── initech-brand (overlay)
├── social-content               └── etc.
└── visual-creative

The generic capabilities know HOW to do the work. The client overlays know WHO you’re doing it for.

Step 1: Create a Client Overlay

This is the most common thing you’ll do. Every new client gets one.

Create the folder

clients/
└── acme-nonprofit/
    ├── capability.yaml    ← the manifest (copy the template below)
    ├── index.txt          ← one line about this client
    ├── summary.md         ← key facts at a glance
    └── standard.md        ← everything the AI needs to know

index.txt (one line — that’s it)

acme-nonprofit: Acme Foundation — youth education nonprofit, warm/invitational voice.

summary.md (the cheat sheet)

Write this like you’re briefing a new team member in 60 seconds:

# Acme Foundation — Quick Reference

- **Full Name:** Acme Foundation
- **Sector:** Youth education / community development
- **Audience:** Parents, educators, community donors
- **Voice:** Warm, invitational, community-focused
- **Don't:** Guilt-based appeals, political messaging
- **Do:** Say "partner" not "donor", emphasize impact
- **Current campaigns:** Spring enrollment drive
- **Key contact:** Jamie (marketing lead)

standard.md (the full brief)

This is where you put everything the AI needs. Write it like a brand guide meets an account brief:

# Acme Foundation — Full Client Profile

## Organization
Acme Foundation is a youth education nonprofit...

## Brand Voice
- Tone: [describe it like you'd tell a new copywriter]
- Examples of good copy: [paste real examples that nailed it]
- Examples of bad copy: [paste things that missed the mark]

## Audience Segments
- Segment 1: Parents (30-50)
  - What motivates them: [...]
  - What turns them off: [...]
- Segment 2: Educators
  - ...

## Red Lines (things to NEVER do)
- Never use guilt-based fundraising language
- Never take political sides
- Never make claims about outcomes without data

## Past Campaign Performance
- Spring 2025 email series: [what worked, what didn't]
- Year-end 2025: [results, learnings]

## Technical Details
- Email platform: [whatever they use]
- Donation platform: [Stripe, etc.]
- Tracking: [any specific UTM conventions]

## Key People
- Jamie: [role, preferences, communication style]
- Alex: [role, what they care about]

capability.yaml (copy this template, fill in the blanks)

name: acme-brand
type: capability
version: 1.0.0
description: "Acme Foundation — brand voice, audience, campaign history, and org-specific rules."

provides:
  - client-profile
  - acme-nonprofit

requires:
  tools: []
  capabilities: []
  context: []

budget:
  index: 15
  summary: 200
  standard: 1200

activation:
  triggers:
    - type: pattern
      match: "(?i)(acme|acme.foundation)"
  trigger_logic: OR
  co_activates: []
  conflicts: []

permissions:
  data:
    client_data: read-only

behavioral:
  core: |
    When working on Acme Foundation content:
    - Use warm, invitational tone
    - Say "partner" not "donor"
    - Never use guilt-based appeals
    - Stay politically neutral
    - Emphasize measurable impact
  overlays: []

state_schema:
  version: 1
  max_size_tokens: 100
  fields:
    - name: active_campaign
      type: string
    - name: last_approval_status
      type: string

verification:
  checklist:
    - "Content matches brand voice"
    - "No red line violations"
    - "Audience segment is appropriate"
  completion_signal: content_approved

That’s it. Four files per client. The AI reads them when that client comes up and immediately knows who they are, how they talk, and what not to do.

Step 2: Add Red Team Rules Per Client

You can either add a red team section to the client’s standard.md, or create a separate capability. For most clients, just add a section:

Add to the client’s standard.md:

## Red Team Rules (Client-Specific)

When reviewing content for Acme Foundation, flag:
- [ ] Any guilt-based language ("you must," "how can you ignore")
- [ ] Political messaging or partisan language
- [ ] Unverified outcome claims
- [ ] Urgency that feels manufactured
- [ ] Anything that wouldn't feel natural in a parent-teacher conversation

The generic red team capability will pick these up automatically when the client overlay is mounted.

Step 3: Validate Your Work

After creating a client overlay, check it:

acr validate clients/acme-nonprofit/

You want to see: ✅ Valid

If something’s wrong, it’ll tell you exactly what to fix.

Quick Reference: What Goes Where

Thing you want to capture	Where it goes
Client’s brand voice	`standard.md` → Brand Voice section
Colors, fonts, logos	`standard.md` → Brand Assets section (or link to Figma)
Past campaign results	`standard.md` → Past Campaign Performance
Things to NEVER do	`standard.md` → Red Lines section
Things the red team should catch	`standard.md` → Red Team Rules section
Key people and their preferences	`standard.md` → Key People section
Tracking codes / UTM conventions	`standard.md` → Technical Details section
“We learned X from last campaign”	`standard.md` → update Past Campaign Performance
Quick client context for the team	`summary.md`

Updating Client Profiles

As you learn more about a client, just update their files:

After a meeting: Add notes to standard.md (new preferences, feedback, learnings)
After a campaign: Update Past Campaign Performance with results
New red line discovered: Add it to the Red Team Rules section
Contact change: Update Key People

The AI picks up changes immediately on the next run. No deployment, no code, just edit the file.

Creating a New Generic Capability

Less common, but straightforward:

Create the folder with the four files (capability.yaml, index.txt, summary.md, standard.md)
Write the standard.md as if you’re training a really smart new hire — tell them exactly how to do the task
Fill in the capability.yaml using the template above
Run acr validate to check your work

The key question for standard.md: “If I hired a smart person and handed them just this document, could they do the job?” If yes, it’s good enough.

Common Mistakes

Mistake	Fix
Writing vague brand voice (“be professional”)	Be specific: “Write like you’re texting a friend who cares about this cause — casual but sincere”
Forgetting to include real examples	Paste actual good/bad copy samples. AI learns from examples better than rules.
Making standard.md too long	Keep it under 1,500 tokens (~3 pages). If you need more, use deep.md
Not updating after campaigns	Set a reminder: after every campaign wraps, spend 10 minutes updating the profile
Writing red team rules as suggestions	Write them as hard rules: “FLAG if…” not “Consider whether…”

Need Help?

“I don’t know what to put in a section” → Leave it blank with a TODO. Something is better than nothing.
“How do I check token count?” → Run acr budget clients/acme-nonprofit/
“I want to test it before going live” → Run a test pipeline with your overlay.
“The YAML syntax is confusing” → Copy an existing client’s capability.yaml and just change the names and descriptions. Don’t write YAML from scratch.