Source: frameworks/kit-new-client/03-new-client-kit-golden-example.md

03 — GOLDEN EXAMPLE: New Client Kit

This is not an HTML golden example — this kit produces a directory structure, not a single document. This file is a structural reference pointing to both live repos with annotations.

Two Live References

Repo	Client	Status	What It Proves
`aos-client-jb`	Jenn Barr / TRG Consultants	Latest tested	The reference for cascade skill structure — 9-step post-session cascade tested and working
`aos-client-rc`	Ruben Cruz / Crulliance	Fully mature	What a workspace looks like with cascade + brand kit, production skills, and QC agents

aos-client-rc — Fully Mature Workspace

Annotated File Tree

aos-client-rc/
├── .claude/
│   ├── settings.local.json          ← MODE 1: Identical across all clients
│   └── skills/                      ← MODE 1: 9-step cascade + /start
│       ├── start/SKILL.md              Session opener — sync, reference data, orient
│       ├── recap-email/SKILL.md        Step 1 — client recap email (Gmail draft)
│       ├── recap-internal/SKILL.md     Step 2 — internal recap (markdown)
│       ├── recap-json/SKILL.md         Step 3 — profile JSON (structured data)
│       ├── update-cpm/SKILL.md         Step 4 — constraint priority matrix
│       ├── update-masterplan/SKILL.md  Step 5 — master plan
│       ├── update-projectplan/SKILL.md Step 6 — project plan(s)
│       ├── update-blueprint/SKILL.md   Step 7 — blueprint(s)
│       ├── draft-actions/SKILL.md      Step 8 — draft Kathryn's action items
│       └── session-agenda/SKILL.md     Step 9 — session agenda + advisory email
├── .gitignore                        ← MODE 1: Identical across all clients
├── CLAUDE.md                         ← MODE 1: Created from template, then evolves
├── WORKFLOW.md                       ← MODE 1: Created from template, then evolves
├── README.md                         ← MODE 1: Created from template
│
├── sessions/                         ← MODE 1: Empty directory
│   └── (transcripts, JSON, recaps)      Permanent record — files never move
│
├── inputs/                           ← MODE 1: Empty directory
│   └── (constraint briefs, status       Non-session material → archive/ once processed
│       updates, client docs)
│
├── drafts/                           ← MODE 1: Empty directory
│   └── (email drafts, agendas,          Active production → ready/ or archive/
│       in-progress work)
│
├── ready/                            ← MODE 1: Empty directory
│   └── (finished work waiting           → archive/ when sent or published
│       to send)
│
├── archive/                          ← MODE 1: Empty directory
│   └── (retired, consumed, or          Final state — nothing leaves archive
│       abandoned work)
│
├── reference/
│   └── core/                         ← MODE 1: Empty files created
│       ├── soul.md                      MODE 2: Populated from transcripts
│       ├── offer.md
│       ├── audience.md
│       └── voice.md
│
├── reference-data/
│   └── ruben-reference-data.md       ← MODE 1: Copied from vault template
│                                        MODE 2: Populated from first transcript
│
├── constraints/                      ← MODE 1: Empty directory
│   ├── ruben-constraint-brief-*.md      MODE 2+: Briefs arrive from client
│   └── ruben-constraint-priority-       MODE 2+: CPMs produced weekly
│       matrix-*.md
│
├── master-plan/                      ← MODE 1: Empty directory
│   └── client-master-plan-               MODE 2: Created from first CPM
│       ruben-cruz.html
│
├── projects/                         ← MODE 1: Only non-project-specific exists
│   ├── accounting-manager-os/           MODE 2: Scaffolded when initiative identified
│   │   ├── blueprint/
│   │   ├── project-plan/
│   │   ├── builds/
│   │   ├── deployment/
│   │   ├── drafts/
│   │   ├── inputs/
│   │   └── outputs/
│   └── non-project-specific/         ← MODE 1: Created empty
│
├── kits/                             ← MODE 1: Empty directory
│   └── sop/                             MODE 3: Full SOP deployment kit
│       ├── 00-start-here.md                 System map, how to use the full kit
│       ├── 01-crulliance-extensions.md      Role tags, file naming, client context
│       ├── 04-crulliance-sop-qc.md          QC extension (runs after vault Gate 2)
│       └── 05-crulliance-sop-skill.md       Complete visual spec, bash verification
│
├── frameworks/                       ← MODE 1: Empty directory (client IP goes here)
│
├── brand/                            ← MODE 1: Empty directory
│   ├── crulliance-brand-skill.md        MODE 3: Brand skill (production rules)
│   ├── crulliance-brand-qc.md           MODE 3: Brand QC (quality gate)
│   └── (brand HTML reference)           MODE 3: Brand visual reference
│
├── skills/                           ← MODE 1: Empty directory
│   ├── crulliance-master-plan-skill.md     MODE 3: Master plan processing rules
│   ├── crulliance-project-plan-skill.md    MODE 3: Project plan creation spec
│   ├── crulliance-project-plan-           MODE 3: Project plan update rules
│   │   update-skill.md
│   ├── crulliance-blueprint-skill.md       MODE 3: Blueprint creation spec
│   ├── crulliance-blueprint-update-       MODE 3: Blueprint update rules
│   │   skill.md
│   └── session-agenda-skill.md             MODE 3: Session agenda production
│
└── qc-agents/                        ← MODE 1: Empty directory
    ├── crulliance-blueprint-update-      MODE 3: Blueprint update quality gate
    │   qc.md
    └── crulliance-project-plan-          MODE 3: Project plan update quality gate
        update-qc.md

Creation Timeline

When	What Was Created
Day 0 (Mode 1)	Full directory structure, CLAUDE.md, WORKFLOW.md, README, settings, .gitignore, empty reference data, 9-step cascade skills + /start in `.claude/skills/`
Session 1 (Mode 2)	Reference data populated, first CPM, master plan, project scaffolded (accounting-manager-os)
Sessions 2-3	Weekly CPMs, master plan updates, project plan, blueprint, first builds
Session 3+ (Mode 3)	Brand skill, brand QC, brand HTML reference
Ongoing (Mode 3)	Skills added as document types recurred, QC agents added as quality patterns emerged

File Counts

Category	Count	Files
Cascade skills (`.claude/skills/`)	10	start + 9 cascade steps (recap-email, recap-internal, recap-json, update-cpm, update-masterplan, update-projectplan, update-blueprint, draft-actions, session-agenda)
Brand	2	brand skill, brand QC (HTML reference stored separately)
Kits	1	SOP kit (4 files: start-here, extensions, QC extension, visual spec)
Production skills (`skills/`)	8	master plan, project plan, project plan update, blueprint, blueprint update, session agenda, advisory email, stakeholder email
QC Agents	2	blueprint update QC, project plan update QC

aos-client-jb — Cascade Reference

Annotated File Tree

aos-client-jb/
├── .claude/
│   ├── settings.local.json          ← Identical to RC
│   ├── settings.json                ← PreToolUse hook for reference data gate
│   ├── hooks/
│   │   └── require-reference-data.sh   Blocks writes without reference data read
│   └── skills/                      ← 9-step cascade (tested, working)
│       ├── start/SKILL.md
│       ├── recap-email/SKILL.md
│       ├── recap-internal/SKILL.md
│       ├── recap-json/SKILL.md
│       ├── update-cpm/SKILL.md
│       ├── update-masterplan/SKILL.md
│       ├── update-projectplan/SKILL.md
│       ├── update-blueprint/SKILL.md
│       ├── draft-actions/SKILL.md
│       └── session-agenda/SKILL.md
├── .gitignore                        ← Identical to RC
├── CLAUDE.md                         ← Same structure as RC
├── WORKFLOW.md                       ← Same structure as RC, simpler (no stakeholder sessions)
├── README.md                         ← Same structure as RC
│
├── sessions/                         ← Transcripts + JSON (permanent record)
├── inputs/                           ← Non-session material (staging)
├── drafts/                           ← Active production (staging)
├── ready/                            ← Finished, waiting to send
├── archive/                          ← Retired/consumed/abandoned
│
├── reference/
│   └── core/                         ← Client portrait (soul, offer, audience, voice)
├── reference-data/
│   └── jenn-reference-data.md        ← Populated from transcripts
│
├── constraints/                      ← CPMs produced weekly
├── master-plan/                      ← Living doc exists
│
├── projects/
│   ├── authority-os/                 ← First initiative — full subdirectory structure
│   │   ├── blueprint/
│   │   ├── project-plan/
│   │   ├── builds/
│   │   ├── deployment/
│   │   ├── drafts/
│   │   ├── inputs/
│   │   └── outputs/
│   ├── recruiting-process/           ← Second initiative — same structure
│   │   ├── blueprint/
│   │   ├── project-plan/
│   │   ├── builds/
│   │   ├── deployment/
│   │   ├── drafts/
│   │   ├── inputs/
│   │   └── outputs/
│   └── non-project-specific/
│
├── kits/                             ← EMPTY — vault frameworks handle everything
├── frameworks/                       ← EMPTY — no client IP built yet
├── brand/                            ← EMPTY — vault frameworks handle everything
├── skills/                           ← EMPTY — vault frameworks handle everything
└── qc-agents/                        ← EMPTY — vault frameworks handle everything

What This Proves

JB demonstrates that the 9-step cascade works from Day 0 without client-specific production skills. The cascade skills reference vault frameworks for processing logic:

/recap-email → /recap-internal → /recap-json (session processing)
/update-cpm → vault content/frameworks/kit-constraint-priority-matrix/
/update-masterplan → vault content/frameworks/kit-master-plan/
/update-projectplan → vault content/frameworks/kit-project-plan/
/update-blueprint → vault content/frameworks/kit-blueprint/
/draft-actions → scans skills/ and kits/ for matches
/session-agenda → produces agenda + advisory email
No brand skill needed — Advisory OS defaults apply
No client-specific QC needed — vault QC agents apply

Comparison Table

Feature	RC (Mature)	JB (Cascade Reference)
Directory structure	Full	Full (identical)
Cascade skills (`.claude/skills/`)	10 files (start + 9 cascade) — reference client-specific production skills	10 files (start + 9 cascade) — self-contained, reference vault frameworks
CLAUDE.md	Has skills/QC quick references	References vault frameworks only
WORKFLOW.md	References client skills by name	References vault frameworks only
settings.local.json	Identical (Windows + Linux paths)	Identical (Windows + Linux paths)
.gitignore	Identical	Identical
Reference data	Populated	Populated
Brand kit	2 files (skill + QC)	Empty directory
Production skills (`skills/`)	8 files	Empty directory
QC agents	2 files	Empty directory
Post-session cascade	9 steps, auto-advance	9 steps, auto-advance
Human review gates	After every step	After every step
Vault frameworks used	Same 6	Same 6

Patterns to Replicate

CLAUDE.md Structure

Both repos follow this exact section order:

Header — Client name, company, engagement type, rate, cadence, start date
The Client — Who they are, what drives them, engagement context
Engagement Snapshot — Current phase, momentum, active constraint, what to watch
Working With This Client — Session dynamics, what works, what doesn't, watch-fors
Reference Tiers — What to load and when
Session Opener — Step 0: vault-inbox check, file move review, kanban flow
Active Projects (if any exist) — brief description of each
Weekly Workflow — table with Steps 0-6, input/output columns, link to WORKFLOW.md
Where Things Are — table mapping every file type to its location (includes sessions, inputs, drafts, ready, archive, reference/core)
Connected Repos — vault + business-aos, via settings.local.json
Vault Frameworks Used — table with framework name, vault location, which step uses it
Skills Quick Reference (RC) / omitted (JB) — table of client-specific skills
QC Quick Reference (RC) / omitted (JB) — table of client-specific QC agents
Rules — standard rules (reference data canonical, blueprint is client's doc, project plan is Kathryn's, master plan never shared, never update CPM from brief alone)

WORKFLOW.md Structure

Both repos follow this exact section order:

Weekly Rhythm — table mapping days to what happens
Before You Start — read reference data first (no exceptions), pull latest, grab files from Drive, drop transcripts into sessions/ and other inputs into inputs/, QC recap email
Email QC — two checklists run on Relay-produced recap before weekly update starts
Steps 1-6 — each with: trigger, what Claude reads, framework reference, output path, quality gate, STOP — Human Review
After Processing — push to GitHub
Monthly Review — master plan comprehensive review
Skills Quick Reference (RC) / omitted (JB)
QC Quick Reference (RC) / omitted (JB)

Brand Skill Pattern (from RC)

The Crulliance brand skill establishes:

Color system where one color = one meaning (consistent across all documents)
Typography rules (which fonts for headers vs body)
Terminology rules (what the client calls things vs what Advisory OS calls them)
Document-specific rules (blueprint is light theme, master plan is dark theme)

When creating a new client's brand skill, follow this same pattern: colors, typography, terminology, document rules.