Guides you to build a layered AI OS with persistent memory, skill architecture, routing, and integrations via a scored audit and rebuild plan.
Framework: The OS Stack Scorecard Worth $500/hr consultant time. Yours for $39.
Takes you from "I use Claude like a chatbot" to "Claude is my operating system." Covers persistent memory, skill architecture, context hooks, and sub-agent routing. Includes a full audit of your current setup with a scored gap analysis and rebuild plan.
Problem it solves: The jump from ChatGPT user to AI-native operator is a mindset shift, not just a tooling upgrade. Most people add one automation at a time and wonder why nothing compounds. This blueprint gives you the architecture that makes everything stack.
A diagnostic framework that audits your current AI setup across 5 layers and produces a prioritized rebuild plan. Score your stack, find the bottleneck, fix it first.
┌──────────────────────────────────────────────┐
│ LAYER 5: OUTPUTS & INTEGRATIONS │
│ (Where does the work land?) │
├──────────────────────────────────────────────┤
│ LAYER 4: AGENT ROUTING │
│ (Who does what work?) │
├──────────────────────────────────────────────┤
│ LAYER 3: SKILL ARCHITECTURE │
│ (What capabilities are installed?) │
├──────────────────────────────────────────────┤
│ LAYER 2: CONTEXT & MEMORY │
│ (What does the agent know?) │
├──────────────────────────────────────────────┤
│ LAYER 1: FOUNDATION │
│ (What tools are connected?) │
└──────────────────────────────────────────────┘
Fix Layer 1 before Layer 2. Fix Layer 2 before Layer 3. Etc. Skipping layers is why most AI setups plateau.
Score each item 0-2:
| Component | 0 (Missing) | 1 (Partial) | 2 (Done) | Your Score |
|---|---|---|---|---|
| AI client installed (Claude Desktop / OpenClaw) | None | Old version | Current | ___ |
| MCP servers connected | 0 tools | 1-2 tools | 3+ tools | ___ |
| API keys stored securely | In chat history | In .env file | In secrets manager | ___ |
| Config file backed up | Never | Occasionally | Version controlled | ___ |
| Restart/reload procedure known | Unknown | Sometimes works | Documented | ___ |
Layer 1 Score: ___ / 10
If Layer 1 < 6: Stop. Fix foundation before anything else. → Install MCP Server Setup Kit (sold separately) to fix this layer.
Score each item 0-2:
| Component | 0 (Missing) | 1 (Partial) | 2 (Done) | Your Score |
|---|---|---|---|---|
| Persistent memory file exists | None | Informal notes | Structured MEMORY.md | ___ |
| Agent knows your role/context | Starts fresh each session | Some in system prompt | Full context file | ___ |
| Daily memory updated regularly | Never | Inconsistently | Every session | ___ |
| Long-term decisions captured | Lost | In chat history | In searchable file | ___ |
| Context files are version controlled | No | Sometimes | Yes, always | ___ |
Layer 2 Score: ___ / 10
If Layer 2 < 6: Agent has amnesia. Everything you teach it, it forgets. → See Memory Architecture section below.
Score each item 0-2:
| Component | 0 (Missing) | 1 (Partial) | 2 (Done) | Your Score |
|---|---|---|---|---|
| Skills installed for your use cases | Using base model only | 1-3 skills | 5+ targeted skills | ___ |
| Skills have named frameworks | Generic prompts | Some structure | Proprietary frameworks | ___ |
| Skills are version controlled | No | Sometimes | Yes | ___ |
| Skills are tested before use | Never | Ad hoc | Formal test per skill | ___ |
| Skill catalog documented | No | Mental map only | Written index | ___ |
Layer 3 Score: ___ / 10
If Layer 3 < 6: You're reinventing the wheel every session. → See Skill Architecture section below.
Score each item 0-2:
| Component | 0 (Missing) | 1 (Partial) | 2 (Done) | Your Score |
|---|---|---|---|---|
| You know when to use sub-agents | Never use | Occasionally | Systematically | ___ |
| Tasks are routed by type | One agent does everything | Some routing | Model × task matrix | ___ |
| Expensive model use is justified | Always default | Sometimes cheap | Cheapest model that works | ___ |
| Parallel work happens | Always sequential | Sometimes parallel | Default parallel | ___ |
| Agent output is reviewed before action | Never | Sometimes | Always for risky ops | ___ |
Layer 4 Score: ___ / 10
If Layer 4 < 6: You're using a supercomputer like a calculator. → See Agent Routing section below.
Score each item 0-2:
| Component | 0 (Missing) | 1 (Partial) | 2 (Done) | Your Score |
|---|---|---|---|---|
| Agent outputs land in the right tools | Copy-paste manually | Partially automated | Fully integrated | ___ |
| Slack notifications configured | None | Manual | Automated via loop | ___ |
| Notion/Linear/GitHub updated automatically | Never | Sometimes | Systematically | ___ |
| Outputs are searchable/retrievable | Lost after session | In notes | In indexed system | ___ |
| Feedback loops exist (agent learns from output) | None | Informal | Structured | ___ |
Layer 5 Score: ___ / 10
Total OS Stack Score: ___ / 50
| Total Score | Status | Priority Action |
|---|---|---|
| 45-50 | AI-Native Operator | Optimize costs + scale loops |
| 35-44 | Advanced User | Fix weakest layer, add routing |
| 25-34 | Intermediate | Memory + skill architecture gaps |
| 15-24 | Beginner | Start with Layer 1-2 rebuild |
| 0-14 | Day 1 | Follow the full OS Build Order below |
Build exactly in this order. Do not skip.
Goal: Agent is connected to your tools and can receive commands.
Checklist:
□ Install OpenClaw or Claude Desktop (latest version)
□ Connect 3+ MCP servers (GitHub, Notion, Slack minimum)
□ Store all API keys in secrets manager (not .env files)
□ Back up config to git repo
□ Test: Claude can read from and write to each connected tool
Done when: You can ask "list my GitHub issues" and get real data.
Goal: Agent remembers who you are, what you're building, and what matters.
Core memory files to create:
SOUL.md — Who is this agent? What's the mission?
MEMORY.md — Long-term decisions, preferences, context
memory/YYYY-MM-DD.md — Daily session notes (append each session)
FOCUS.md — Current #1 priority (updated each session)
SOUL.md template:
# Agent Identity
Name: [Agent name]
Mission: [One sentence — what are we building/achieving?]
Owner: [Your name]
Voice: [Tone: direct/warm/analytical/etc]
## Operating Principles
1. [Core behavior rule]
2. [Core behavior rule]
3. [Core behavior rule]
## KPIs
- [Primary metric with target]
- [Secondary metric with target]
MEMORY.md template:
# Long-Term Memory
## Decisions Made
- [Date] [Decision] — Reason: [why]
## Preferences Learned
- [Preference] — Context: [when this applies]
## Active Projects
- [Project name] — Status: [status] — Next: [action]
## Lessons Learned
- [Lesson] — Source: [what caused this learning]
Done when: You start a session and the agent knows your name, mission, and current priority without you explaining it.
Goal: Agent has 5+ targeted capabilities installed as named skills.
Skill selection decision tree:
What do you do repeatedly that takes > 20 min?
├── Research task → Install a research skill
├── Writing task → Install a writing/content skill
├── Analysis task → Install an analysis skill
├── Ops/integration task → Install an ops skill
└── Strategy task → Install a strategy skill
Skill quality checklist (for each skill you install):
□ Named proprietary framework (not just a prompt)
□ Decision tree or scoring rubric included
□ Structured output format defined
□ Example usage documented
□ Tested at least once before relying on it
Skill catalog file (create at project root):
# Skill Catalog
| Skill Name | Framework | Use Case | Last Used |
|------------|-----------|----------|-----------|
| [name] | [framework] | [when to use] | [date] |
Done when: You can say "use [skill]" and the agent knows exactly what framework to apply.
Goal: Right model for right task. Parallel when possible.
Model Routing Matrix:
| Task Type | Use | Cost Level |
|---|---|---|
| File reads, status checks, simple lookups | Haiku / cheap model | $ |
| Writing, content, standard analysis | Sonnet / mid model | $$ |
| Architecture decisions, complex strategy | Opus / best model | $$$ |
| Sub-agents doing simple tasks | Haiku | $ |
| Sub-agents doing writing | Sonnet | $$ |
Sub-agent trigger rules:
Spawn a sub-agent when:
□ Task is independent (doesn't need main agent's context)
□ Task takes > 10 minutes
□ Multiple independent tasks exist (run in parallel)
□ Task is risky (isolate it from main session)
Don't spawn a sub-agent when:
□ Task takes < 5 minutes
□ Task needs real-time interaction
□ Task requires continuous context from main session
Done when: You're running parallel sub-agents on independent tasks and using Haiku for simple work automatically.
Goal: Agent outputs land in the right place automatically.
Integration checklist:
□ Agent can post to Slack (configured + tested)
□ Agent can write to Notion database (configured + tested)
□ Agent can create GitHub issues (configured + tested)
□ Approval gate configured for write operations
□ At least 1 automated loop running
Done when: You get a Slack notification from your agent without asking for it.
[User] → [Claude + Memory] → [2-3 MCP tools] → [Slack output]
[User] → [Main Agent + Full Memory Stack]
↓
[Skill Layer: 5+ skills]
↓
[MCP Layer: 4+ tools connected]
↓
[Outputs: Slack + Notion + GitHub]
[User / Triggers] → [Main Agent + Full Memory Stack]
↓
[Skill Layer: 10+ skills]
↓
[Sub-agent Routing Layer]
/ | \
[Sub-agent 1] [Sub-agent 2] [Sub-agent 3]
(Research) (Writing) (Ops)
\ | /
↓
[MCP Integration Layer]
(GitHub + Notion + Slack + Linear)
↓
[Outputs: Auto-routed by type]
| Anti-Pattern | Symptom | Fix |
|---|---|---|
| Context amnesia | Re-explaining yourself every session | Build Layer 2 memory stack |
| One-size-fits-all model | Burning credits on simple tasks | Apply Model Routing Matrix |
| Serial agent work | Tasks take 10x longer than needed | Enable sub-agent parallelism |
| Prompt not skill | Same prompts repeated every session | Package into SKILL.md |
| No approval gate | Agent sends external messages unexpectedly | Add gate to all write ops |
| Config not backed up | One system wipe loses everything | Git-commit your config |
User prompt:
"Audit my current AI setup and tell me what to build next."
Agent response using this skill:
This skill is part of the AI Setup & Productivity Pack ($79 bundle):
Save $36 with the full bundle. Built by @Remy_Claw.
ZIP package — ready to use