This page maps how Claude operates across Welch Wellbeing — which mode does what, what each mode can access, how context flows between systems, and where data lives. It’s the single reference point for understanding the AI architecture so that both Lee and Claude stay aligned on structure.
Claude Modes
Three modes, three distinct roles. Each has different capabilities, memory, and system access.
The Thinking Layer — strategy, philosophy, deep conversation, decision-making
Memory
Persistent cross-session Past chat searchSystem Access
Google Drive Gmail Calendar ClickUp P: Drive C: Drive Web SearchBest For
Strategic planning, IK/Sovereign Fields thinking, member decisions, complex problem-solving, content strategy, philosophical exploration
The Production Layer — file creation, bulk operations, scheduled tasks, repetitive output
Memory
Project-scoped No cross-session memorySystem Access
C: Drive (primary) P: Drive ClickUp Web SearchBest For
Kartra HTML pages, vault file operations, Hub page builds, YouTube descriptions, WhatsApp messages, bulk content production, scheduled tasks
The Automation Layer — scripts, integrations, Empire Build pipeline
Memory
Session onlySystem Access
Full filesystem APIs (programmatic)Planned For
Empire Build L2/L3, Kartra replacement automation, PWA development, API integrations, CI/CD pipelines
Context Flow
How information moves between Claude modes and the four core systems.
System Access Matrix
What each Claude mode can reach. This is the wiring diagram.
| System | Chat | Co-Work | Code (Future) | Access Method |
|---|---|---|---|---|
| ClickUp | ✓ MCP | ✓ MCP | Planned | ClickUp MCP Connector |
| Google Drive | ✓ MCP | Limited | Planned | Drive Search + Fetch |
| Gmail | ✓ MCP | Limited | Planned | Gmail MCP Connector |
| Google Calendar | ✓ MCP | Limited | Planned | Calendar MCP Connector |
| C: Drive (Vault) | ✓ Filesystem | ✓ Native | Planned | Filesystem MCP / Direct |
| P: Drive (Output) | ✓ Filesystem | ✓ Native | Planned | Filesystem MCP / Direct |
| Kartra | ✗ Manual | ✗ Manual | L2 Target | Copy-paste HTML upload |
| Web Search | ✓ Native | ✓ Native | — | Built-in tool |
| Persistent Memory | ✓ Cross-session | Project only | ✗ None | Anthropic memory system |
| Past Chat Search | ✓ Native | ✗ None | ✗ None | Chat-only feature |
| Skill Files | ✓ /mnt/skills/ | ✓ /mnt/skills/ | — | Mounted read-only |
Skills Catalogue
Agent-readable skills encode expertise into repeatable workflows. These are the callable skills available across Claude modes, tiered by scope and composability.
Skills are markdown files that give Claude context to produce predictable output for specific tasks. They compound over time — unlike prompts, which evaporate after a conversation. The vault note skills.md is the canonical source of truth for methodology, tiering, and evolution.
◆ Tier 1 — Standard (Org-wide conventions)
◆ Tier 2 — Methodology (Practitioner craft)
◆ Tier 3 — Personal (Creative & reflective)
Workflow Chains
How skills compose — where one skill’s output becomes the next skill’s input.
Teach → youtube-description → Upload → whatsapp-class-share → Send
Insight → second-brain-note → Review → Vault → Cowork sweep
Build HTML → hub-ops → Git push → Cloudflare auto-deploy
ClickUp task → tcg-support → Review → Gmail send
Data Boundaries
What’s cloud vs local, what persists vs what’s session-scoped, what syncs where.
Co-Work Projects (Recommended)
Scoped workspaces for production work where context persistence matters most. Skills remain portable across all projects.
These are recommended project scopes for Co-Work, aligned to the departments where context loss costs the most time. Chat remains open-plan — no Chat projects.
Kartra lesson pages, YouTube descriptions, course architecture, movement library builds. Claude remembers which lesson you’re on, naming conventions, and series state.
Absence handling, new member follow-ups, taster communications, class register management. Context carries across member interactions.
L1/L2/L3 infrastructure work, Cloudflare migration, Hub development, automation pipeline. Deeply technical, high context overhead.
System Maps
Interactive maps of key platforms. Deep-dive views into how each system is wired.
| Page | Status | Data Source | Added |
|---|---|---|---|
| Hub Homepage | Active | ww-data.json + localStorage | 2026-03-14 |
| Events Register replaces Calculator + Summary | Active | ww-data.json | 2026-04-11 |
| Systems Architecture | Active | static HTML + ww-data.json | 2026-03-14 |
| Processes | Active | static HTML | 2026-03-14 |
| Four Pillars | Active | static HTML | 2026-03-26 |
| AI Session Cheat Sheet | Active | static HTML | 2026-03-28 |
| Stone Circle + 12 Stone Pages | Active | static HTML | 2026-03-29 |
| Kartra System Map | Active | static HTML | 2026-03-14 |
| ww-data.json | Active | — | 2026-04-07 |
| Season Calculator | Removed | — | 2026-03-26 |
| Season Summary | Removed | — | 2026-03-26 |
Data Conventions
Naming patterns and code conventions used across ww-data.json, ClickUp, and the vault.
Identify the movement set / curriculum. Uppercase, alphanumeric.
Identify a specific teaching sequence for a given session duration. A class uses planCode to link to its applicable plan.
Unique identifier per venue/day/time slot. Matches ClickUp tag convention.
ClickUp event register task naming convention.
Key Paths
Canonical file locations. One source of truth — no duplicates.