ACE Framework for Personal AI

Table of content

The ACE (Autonomous Cognitive Entity) framework structures AI systems into six distinct layers, from ethical foundations at the top to task execution at the bottom. Originally designed for autonomous agents, you can adapt it to give Claude Code persistent context about who you are and what matters to you.

See David Shapiro’s ACE Framework for background on the six layers and their functions.

The Six Layers

LayerNamePurposeYour Implementation
6AspirationalMission, values, ethicsaspirational.md
5Global StrategyLong-term goals, contextstrategy.md
4Agent ModelCapabilities, limitationscapabilities.md
3Executive FunctionTask decompositionworkflows.md
2Cognitive ControlFocus, prioritizationpriorities.md
1Task ProsecutionExecutionClaude Code tools

Information flows down (values shape strategy, strategy shapes tasks) and up (execution results inform planning).

Directory Structure

.claude/
  ace/
    aspirational.md      # Layer 6: Mission and values
    strategy.md          # Layer 5: Goals and constraints
    capabilities.md      # Layer 4: What the AI can do
    workflows.md         # Layer 3: Task decomposition patterns
    priorities.md        # Layer 2: Current focus areas

Reference these in your CLAUDE.md:

## ACE Configuration

Read .claude/ace/ for context on my mission, values, and current priorities.
Check aspirational.md before any action that could affect my reputation.
Check strategy.md when planning multi-step projects.

Layer 6: Aspirational

This layer contains your mission, core values, and hard boundaries. Think of it as the constitution that governs all AI behavior.

Create .claude/ace/aspirational.md:

# Mission

Help me build a sustainable business while maintaining health and relationships.

## Heuristic Imperatives

Adapted from Shapiro's framework:
1. Reduce suffering (mine, others, future self)
2. Increase prosperity (financial, creative, relational)
3. Increase understanding (learning, documenting, sharing)

## Core Values

| Value | Meaning |
|-------|---------|
| Sustainability | Avoid burnout, pace over speed |
| Quality | Ship less but ship well |
| Privacy | Never expose personal data |
| Autonomy | Suggest, don't decide for me |
| Honesty | Acknowledge uncertainty |

## Hard Boundaries

Never do these without explicit approval:
- Post to social media as me
- Make financial transactions
- Send emails/messages on my behalf
- Share personal information externally
- Delete files without confirmation

Layer 5: Global Strategy

This layer translates your mission into concrete objectives with context about your situation.

Create .claude/ace/strategy.md:

# Q1 2026 Objectives

1. Launch AI writing course (primary)
2. Grow newsletter to 5,000 subscribers
3. Maintain 3x/week exercise routine

## Current Context

| Factor | Status |
|--------|--------|
| Career stage | Mid-career, transitioning to content |
| Work capacity | 35 hours/week |
| Learning curve | Video production (steep) |
| Existing assets | 2,000 newsletter subscribers |

## Constraints

- **Time**: 35 hours/week max, evenings protected
- **Budget**: $500/month for tools
- **Team**: Solo operation
- **Skills**: Strong writing, basic Python, learning video

## Strategic Priorities

1. Course content before marketing
2. Repurpose content across channels
3. Automate repetitive tasks
4. Build in public (newsletter + social)

Layer 4: Agent Model

This layer documents what Claude Code can and cannot do in your specific setup.

Create .claude/ace/capabilities.md:

# Capabilities

## Tools Available

| Tool | Access Level |
|------|--------------|
| File read/write | Full |
| Shell commands | Sandboxed |
| Web search | Via MCP |
| Calendar | Via Google MCP |
| Notion | Via Notion MCP |
| Git | Full |

## What I Can Do

- Read and edit any file in the project
- Run tests and builds
- Search the web for research
- Create and manage tasks in Notion
- Check and create calendar events
- Commit and push code changes

## What I Cannot Do

- Access email directly
- Make phone calls
- Post to social media
- Make purchases
- Access private APIs without credentials
- Run code outside sandbox

## Context Limits

- Token window: ~100k tokens
- No memory across sessions (use .claude/ace/ for persistence)
- Knowledge cutoff: May 2025

Layer 3: Executive Function

This layer contains patterns for breaking down goals into tasks.

Create .claude/ace/workflows.md:

# Task Decomposition Patterns

## Content Creation

Goal: "Write article about X"
1. Research existing coverage (30 min)
2. Outline key points (15 min)
3. Write draft (1-2 hours)
4. Edit for clarity (30 min)
5. Add code examples (30 min)
6. Create social snippets (15 min)

## Learning New Skill

Goal: "Learn X technology"
1. Find official docs and tutorials
2. Build minimal working example
3. Build something useful for current project
4. Document what I learned
5. Share publicly (newsletter/social)

## Project Planning

Goal: "Build X feature/product"
1. Define scope (what's in, what's out)
2. Identify unknowns and research
3. Break into 2-hour chunks
4. Sequence by dependency
5. Assign to calendar blocks
6. Set review checkpoints

## Time Estimates

| Task Type | Default Estimate |
|-----------|------------------|
| Research | 30 min |
| Writing (draft) | 1 hour per 500 words |
| Editing | 30% of writing time |
| Code (simple) | 30 min |
| Code (complex) | 2 hours |
| Review/feedback | 15 min |

Layer 2: Cognitive Control

This layer captures current focus and prioritization rules.

Create .claude/ace/priorities.md:

# Current Focus

## This Week

1. Finish course module 3
2. Write newsletter issue
3. Review analytics and adjust

## Active Projects

| Project | Status | Next Action |
|---------|--------|-------------|
| AI Course | In progress | Record module 3 video |
| Newsletter | Weekly | Draft Friday issue |
| Website | Maintenance | Update testimonials |

## Prioritization Rules

1. **Urgent + Important**: Do immediately
2. **Important + Not Urgent**: Schedule this week
3. **Urgent + Not Important**: Delegate or automate
4. **Neither**: Delete or defer

## Energy Matching

| Energy Level | Best For |
|--------------|----------|
| High | Writing, recording, complex code |
| Medium | Research, editing, planning |
| Low | Admin, organizing, review |

## Time Boxing

- Deep work: 2-hour blocks, morning
- Admin: 30-min batches, afternoon
- Communication: 2x daily, fixed times

Autonomy Spectrum

Set different autonomy levels for different task types:

LevelBehaviorWhen to Use
0Human does everythingFinancial, legal
1AI suggests, human executesImportant decisions
2AI drafts, human approvesCommunication
3AI executes, reports afterRoutine research
4Full autonomyFile organization

Add to strategy.md:

# Autonomy Settings

task_autonomy:
  financial: 0           # Never autonomous
  communication: 2       # Draft, await approval
  research: 3            # Execute, report results
  file_organization: 4   # Full autonomy
  scheduling: 2          # Propose, confirm
  code_changes: 2        # Review before commit
  publishing: 1          # Suggest only

Using ACE in Daily Prompts

Morning Planning

Reference my ACE configuration in .claude/ace/

Today's context:
- Energy: High (good sleep)
- Time available: 4 focused hours
- Deadline pressure: Medium (course launch in 2 weeks)

Review my priorities.md and suggest what to focus on today.
Consider my strategic objectives and current energy level.

Goal Decomposition

Reference .claude/ace/workflows.md

Goal: Create promotional video for course launch

Break this down following my decomposition patterns.
Check capabilities.md for what tools I have.
Ensure tasks fit in 2-hour blocks.
Flag any tasks that need skills I'm still learning.

Decision Making

Reference .claude/ace/

Decision: Should I add a community forum to my course?

Evaluate against:
1. Strategic objectives (does it help Q1 goals?)
2. Constraints (time/budget available?)
3. Values (sustainability, quality)
4. Capabilities (can I maintain it?)

Recommend: proceed, defer, or decline. Explain tradeoffs.

Weekly Review

Reference .claude/ace/

Review my week:
- What did I accomplish vs. planned?
- What blocked progress?
- What should I adjust in priorities.md?

Update priorities.md with next week's focus.
Flag any misalignment with strategy.md objectives.

Implementation Checklist

StepActionFile
1Write mission statement (1 paragraph)aspirational.md
2List 5-7 core values with definitionsaspirational.md
3Define hard boundaries (5-10 items)aspirational.md
4Set Q1 objectives (3 max)strategy.md
5Document constraints (time, budget, skills)strategy.md
6List available tools and access levelscapabilities.md
7Create 3-5 workflow templatesworkflows.md
8Set current weekly prioritiespriorities.md
9Define autonomy levels per task typestrategy.md
10Reference ace/ in CLAUDE.mdCLAUDE.md

Start with layers 6 and 5 (aspirational and strategy). These provide the most value with the least setup. Add the other layers as you find yourself repeating context in prompts.

Maintenance

FrequencyAction
DailyUpdate priorities.md current focus
WeeklyReview progress against strategy
MonthlyAudit values alignment
QuarterlyUpdate strategy.md objectives

Why bother with all this? You stop explaining your situation in every conversation. Document it once, reference it. Claude Code reads the context and already knows what matters to you.


Next: Building a Memory System

Topics: personal-os architecture ai-agents