aeo-spec-validator

AEO is a set of Vercel Skills that enables Claude Code to execute development tasks autonomously with built-in safeguards:

• Confidence scoring - Knows when to proceed vs when to ask
• Spec validation - Ensures tasks are well-defined before starting
• Automated QA - Internal code reviewer with veto power
• Failure patterns - Recognizes and auto-fixes common errors
• Smart escalation - Human-AI interface with clear options
• Cost tracking - Token usage and budget management
• Architecture protection - Detects circular deps and layer violations

```bash

# Prerequisites: PAI and Ralph must be installed first

npx skills add snarktank/ralph

npx skills add laurence/aeo-skills

```

```bash

# In your project

claude

# Activate autonomous mode

/aeo

# Describe your task

"Add user authentication with email verification"

# AEO will:

# 1. Validate the spec

# 2. Calculate confidence

# 3. Execute or escalate

# 4. Run QA review

# 5. Learn from outcome

```

```

User Request → /aeo activates

↓

aeo-core calculates confidence

↓

aeo-spec-validator scores spec

↓

If confidence < 0.70 → aeo-escalation

If confidence ≥ 0.70 → Execute autonomously

↓

Ralph execution loop (on-demand calls AEO)

↓

aeo-qa-agent reviews (before commit)

↓

If QA veto → Fix → Re-review

If QA passes → Commit

↓

aeo-failure-patterns handles errors

↓

Record outcome → PAI memory

↓

Update confidence model

```

Essential Skills (auto-installed)

1. aeo-core - Confidence engine and autonomy decision making
2. aeo-spec-validator - Task definition validation
3. aeo-qa-agent - Internal code reviewer with veto power
4. aeo-failure-patterns - Error pattern recognition and auto-fix
5. aeo-escalation - Human-AI interface

Optional Skills

1. aeo-cost-governor - Token usage and budget tracking
2. aeo-architecture - Architecture analysis and protection

```bash

# Small task - AEO executes autonomously

"Add email validation to the signup form"

→ Confidence: 0.92 → Autonomous execution

# Medium task - AEO generates mini-PRD

"Create a dashboard showing user metrics"

→ Confidence: 0.78 → Advisory mode → Executes

# Large task - AEO escalates for planning

"Build a complete project management system"

→ Confidence: 0.45 → Escalates → Generates PRD → Ralph executes

# Risky task - AEO blocks

"Refactor the authentication system"

→ Confidence: 0.35 → Refuses → Requests detailed spec

```

Optional configuration files in $PAI_DIR/USER/:

```json

// ~/.claude/USER/aeo-budget.json

{

"daily_budget_usd": 10.00,

"per_task_budget_usd": 2.00,

"alert_threshold_percent": 80

}

```

```yaml

# ~/.claude/USER/aeo-layers.yaml

layers:

- name: "presentation"

may_import: ["domain", "application"]

- name: "domain"

may_import: []

- name: "application"

may_import: ["domain"]

```

AEO stores data in $PAI_DIR/MEMORY/:

• aeo-signals.jsonl - Task outcomes for confidence learning
• aeo-escalations.jsonl - Escalation decisions and patterns
• aeo-failure-patterns.json - Learned error patterns
• aeo-costs.jsonl - Token usage tracking

```bash

# Fork and clone

git clone https://github.com/YOUR_USERNAME/aeo-skills.git

cd aeo-skills

# Make changes

vim skills/aeo-core/SKILL.md

# Test locally

npx skills link .

# Publish

git push origin main

# Verify at https://skills.sh

```

MIT

• [Vercel Skills](https://github.com/vercel-labs/agent-skills)
• [Ralph](https://github.com/snarktank/ralph)
• [PAI](https://github.com/danielmiessler/Personal_A