Skip to content

aPoWall/pos-stack-assessment

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

2 Commits
 
 
 
 
 
 
 
 

Repository files navigation

POS Stack Assessment

Interactive tool for assessing and building your Personal Operating System — an AI agent architecture built on Obsidian vault + Claude Code + naming conventions.

Based on methodology from AI Mindset POS Sprint (March 2026) and AI-Native Organizations Sprint.

What's Inside

File Description
playground.html Interactive stack assessment — 6 layers, 24 blocks, implementation prompts
enhanced-prd.md Enhanced PRD template with sprint methodology baked in

Quick Start

  1. Open playground.html in any browser
  2. Select a preset or start from Blank Slate
  3. Click any block to assess status and get implementation prompts
  4. Copy prompts into Claude Code to build each piece

Architecture

POS is built on 6 layers, bottom-up:

L6 Governance      evals, context protocol, audit, reflect
L5 People & Org    shared vault, role cards, team skills
L4 Skills          atomic workflows, pipelines, scheduled agents
L3 Orchestration   MCP servers, hooks, sub-agents
L2 Data & Context  CLAUDE.md, naming convention, vault, memory
L1 Infrastructure  Claude Code, terminal, Obsidian, voice input

Core Principles

  • Convention = Agreement — naming convention is an agreement between you and the agent. Clearer agreement = more autonomy you can delegate
  • File name is API — agents find files by pattern-matching names. {project} {type} description – YYYY-MM-DD.md
  • Start with one skill, not Jarvis — automate one painful process end-to-end before expanding
  • POS = Architecture of Attention — CLAUDE.md = attention rules, Skills = attention automation, Obsidian = attention memory, MCP = attention channels

Unix Analogy

Unix POS What it is
/etc/ CLAUDE.md Config read on every session start
script.sh Skill One command triggers a chain of actions
kernel module MCP server Driver connecting to external service
cron job Scheduled agent Timed execution of a skill

MVP Tiers

Tier What you need Time
BASE CLAUDE.md + naming convention + 1 skill 1 evening
MID + vault + 2-3 MCP + 2-3 skills 1 week
ADVANCED + dashboard + pipelines + scheduled agents 2-4 weeks

The Playground

Two types of prompts (by design):

  • Block Prompt (right panel) — specific implementation prompt for the selected block. Copy → paste into Claude Code → get that piece built
  • Gap Analysis (bottom panel) — prioritized plan based on all weak blocks. Copy → paste → get a step-by-step implementation order

Presets

Preset Score Use case
Blank Slate 0/108 Starting from zero
BASE MVP ~30/108 Minimum viable POS
MID POS ~60/108 Routine automation
Full POS ~90/108 Complete system
Andrey's PRD ~16/108 Assessment from a real PRD

L0-L3 Autonomy Map

Before building skills, map your processes:

Level Description Example
L0 Human does, AI suggests Partnership negotiations
L1 AI drafts, human approves /draft-message → review
L2 AI executes, human monitors /meeting-summary → auto-save
L3 Full autopilot, escalates on exception Morning brief, status monitoring

Exercise: list 5-7 repeating processes → score current level → identify target → automate biggest gaps first.

References

Methodology

Concept Source
Context → Harness Engineering Epsilla: The Third Evolution (2026)
What is Harness Engineering harness-engineering.ai
State of Context Engineering 2026 Towards AI
Autonomy Levels for AI Agents arXiv paper (Feng, McDonald, Zhang)
5 Levels of AI Autonomy Turian.ai
Autonomy Levels for Agentic AI Cloud Security Alliance
Spec-Driven Development DevOpsTales: Spec-Driven vs. Vibe Coding

Claude Code as OS

Article Author
Claude Code to AI OS Blueprint Jan Lucas Andmann
I Turned Claude Code Into an Operating System Delanoe Pirard
Claude Code Agent Skills prg.sh
Agent Skills Complete Guide The Prompt Index
Self-Learning AI Skill System MindStudio

Obsidian + AI

Article Author
AI-Native Obsidian Vault Setup Guide Chase Adams
Obsidian Vaults & Claude Code: A Second Brain Kenneth Reitz
PKM with Obsidian and AI Eric J. Ma

MCP Servers

Guide Source
Complete Guide to MCP Servers Claude Directory
MCP Servers: Connect, Configure, Use Builder.io
Essential MCP Servers Claude Code Learning Hub
MCP Setup Step-by-Step AgentPatch

Tools

Tool URL Purpose
Claude Code npm install -g @anthropic-ai/claude-code AI agent CLI
Obsidian obsidian.md Vault viewer
grammY grammy.dev Telegram bot framework
Krisp krisp.ai Call transcription
Raycast raycast.com Quick access launcher

AI Mindset

Resource URL
Substack aimindsetspace.substack.com
Knowledge Base base.aimindset.org
Instagram @ai_mind_set_

No Personal Data

This tool stores nothing. All assessment is local in the browser. No cookies, no tracking, no external requests (except Google Fonts for typography). Safe to share and use publicly.

License

MIT

About

Interactive POS (Personal Operating System) stack assessment tool — 6 layers, 24 blocks, implementation prompts. Based on AI Mindset POS Sprint methodology.

Topics

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors

Languages