AI Fluency Framework

Updated in v4.2 - Claude Commands Library is now fully aligned with Anthropic's official AI Fluency Framework, implementing all 4Ds across every command.

What's New in v4.2

Step 0.11: Quick Delegation Check in all commands
Step 0.7: Post-Execution Evaluation with feedback loop
Diligence Reminder: Added to Approval Gate
Feedback Loop: Describe → Evaluate → Refine pattern
Common Mistakes: AI Fluency pitfalls documentation
AI Limitations: Platform awareness for users

What is AI Fluency?

AI Fluency is the ability to work with AI systems in ways that are effective, efficient, ethical, and safe. It includes practical skills, knowledge, insights, and values that help you adapt to evolving AI technologies.

The 4Ds

The framework defines four core competencies:

1. Delegation

Deciding what work should be done by humans, what work should be done by AI, and how to distribute tasks between them.

Implementation in Claude Commands:

Step 0.11: Quick Delegation Check - Universal check in ALL commands (v4.2)
Step 0.13: Full Delegation Assessment - Detailed assessment in /prompt-hybrid
Three components:
- Problem Awareness - Understanding goals before involving AI
- Platform Awareness - Matching tasks to AI capabilities
- Task Delegation - Distributing work thoughtfully

Step 0.11: Quick Delegation Check (NEW in v4.2)

Every command now includes a quick check before proceeding:

Quick Delegation Check:

1. Task Appropriateness:
   - Is this suitable for AI assistance?
   - Does it require human-only judgment?

2. AI Capability Match:
   - Does this match AI strengths?
   - Or exceed AI limitations?

3. Responsibility Awareness:
   - Does user understand they remain responsible?
   - Any safety/security implications?

Decision Logic:

IF task requires ONLY human judgment (ethics, policy):
    → Flag: "This requires human decision. I can help analyze, but you must decide."

IF task involves irreversible actions (delete, deploy, publish):
    → Flag: "⚠️ Irreversible action detected. Requires explicit confirmation."

IF task matches AI strengths AND user accepts responsibility:
    → Proceed to next step

Full Delegation Assessment (prompt-hybrid):

Delegation Assessment:

Problem Awareness:
- Goal: Clear
- Scope: Well-defined
- Success Criteria: Defined

Platform Capabilities:
- Code Analysis: Excellent (use Agent)
- Pattern Detection: Excellent (use Agent)
- Business Decisions: Limited (human must decide)

Recommended Delegation:
- AI Autonomous: Code exploration, pattern detection
- AI with Review: Implementation suggestions
- Human Only: Architecture decisions, security approvals

2. Description

Effectively communicating with AI systems through three types of description:

Type	Purpose	Implementation
Product Description	Define outputs, format, audience	Goal, Context, Scope, Requirements, Constraints, Expected Result
Process Description	Define how AI approaches request	Approach methodology, step-by-step instructions
Performance Description	Define AI behavior during collaboration	Interaction style, communication tone

Implementation:

Phase 0 completeness check expanded from 6 to 9 criteria:

Completeness Check (9 criteria):

Product Description:
✓ Goal, Context, Scope, Requirements, Constraints, Expected Result

Process Description:
✓ Approach: Step-by-step methodology

Performance Description:
✓ Interaction Style: Detailed explanations
✓ Communication Tone: Technical, professional

3. Discernment

Thoughtfully evaluating AI outputs, processes, and behaviors:

Type	Focus	Questions
Product Discernment	Quality of output	Is it accurate? Appropriate? Relevant?
Process Discernment	Reasoning evaluation	Any logical errors? Attention lapses?
Performance Discernment	Communication style	Was it helpful? Clear?

Implementation:

Step 0.7: Post-Execution Evaluation (NEW in v4.2)

After task completion, the system prompts for feedback:

📊 Quick Evaluation (Discernment Check)

How was this output?

- `good` — Accurate, appropriate, useful ✅
- `partial` — Mostly good, needs minor adjustments ⚠️
- `wrong` — Significant issues, needs rework ❌
- `explain` — Show me your reasoning 🔍

Your feedback helps improve future interactions.

Feedback Handling:

Response	Action
`good`	Record success, offer next steps
`partial`	Ask: "What needs adjustment?" → Apply changes
`wrong`	Ask: "What specifically was wrong?" → Record for learning
`explain`	Show reasoning/process, re-prompt

The Feedback Loop (NEW in v4.2)

Effective AI use is iterative:

    DESCRIBE          EVALUATE
   (what you want) → (what you got)
         ↑               ↓
         └─── REFINE ←──┘
           (improve prompt)

Useful follow-up phrases:

"Make it more [concise/detailed/formal/casual]"
"Focus more on [specific aspect]"
"Remove the section about [topic]"
"This is wrong because [reason], please fix"

Discernment Hints in Output:

Discernment Hints:
- Product Evaluation: Verify implementation accuracy
- Process Evaluation: Check reasoning for logical errors
- Performance Evaluation: Was the communication style effective?

4. Diligence

Using AI responsibly and ethically:

Component	Description	Implementation
Creation Diligence	Being thoughtful about AI usage	Interaction mode detection
Transparency Diligence	Being honest about AI's role	Track AI-generated content
Deployment Diligence	Taking responsibility for outputs	Verification checklists

Implementation:

Diligence Reminder in Approval Gate (NEW in v4.2)

Every approval gate now includes a responsibility reminder:

⏸️ Perfected Prompt Ready - Awaiting Your Approval

...

⚖️ Diligence Reminder (AI Fluency):
You remain responsible for any output generated from this prompt.
- Verify key facts before deployment
- Review AI-generated code before committing
- Test thoroughly before production use

Reply with: y/yes, n/no, modify, explain, options

Diligence Summary in /session-end:

Diligence Summary:

AI-Assisted Content Requiring Verification:
- src/auth/login.ts - Generated authentication logic
- src/middleware/jwt.ts - Generated JWT validation

Transparency Notes:
- Authentication flow designed by AI
- Security patterns from existing codebase

Deployment Checklist:
- [ ] Review generated authentication code
- [ ] Test JWT validation edge cases
- [ ] Security audit before deployment

Human-AI Interaction Modes

The framework defines three collaboration modes:

Automation Mode

AI performs specific tasks based on specific human instructions.

Human defines: WHAT needs to be done
AI executes: The defined task
Best for: Simple tasks, clear instructions
Indicators: "Fix X", "Add Y to Z", "Change A to B"

Augmentation Mode

Humans and AI collaborate as thinking partners.

Both contribute: Iterative back-and-forth
Best for: Complex analysis, design decisions
Indicators: "Help me understand", "What's the best approach"

Agency Mode

Human configures AI to work independently.

AI establishes: Knowledge and behavior patterns
Best for: Research, exploration, multi-agent work
Indicators: "Research X", "Explore the codebase"

Detection Logic:

IF prompt contains direct commands AND clear scope:
    → Automation Mode
ELSE IF prompt requests collaboration OR decision-making:
    → Augmentation Mode
ELSE IF prompt requests independent research OR exploration:
    → Agency Mode
DEFAULT:
    → Augmentation Mode (most flexible)

Configuration

AI Fluency settings in .claude/config/ai-fluency.json:

json

{
  "framework": {
    "name": "AI Fluency",
    "core_competencies": ["Delegation", "Description", "Discernment", "Diligence"]
  },
  "delegation": {
    "enabled": true,
    "components": {
      "problem_awareness": { ... },
      "platform_awareness": { ... },
      "task_delegation": { ... }
    }
  },
  "description": {
    "enabled": true,
    "components": {
      "product_description": { "criteria": ["goal", "context", "scope", ...] },
      "process_description": { "criteria": ["approach", "methodology"] },
      "performance_description": { "criteria": ["interaction_style", "tone"] }
    }
  },
  "discernment": {
    "enabled": true,
    "include_hints_in_output": true
  },
  "diligence": {
    "enabled": true,
    "track_in_session_end": true
  },
  "interaction_modes": {
    "automation": { "indicators": ["fix", "add", "change"] },
    "augmentation": { "indicators": ["help", "understand", "approach"] },
    "agency": { "indicators": ["research", "explore", "find all"] }
  }
}

Common Mistakes to Avoid

Based on AI Fluency research, these are the most common pitfalls:

Mistake	Problem	Solution
Being too vague	"Help me with this"	Be specific about what you need
Not providing context	AI can't read your mind	Include technologies, frameworks, environment
Accepting first output	Missing improvements	Iterate! Use feedback to refine
Not verifying facts	AI can hallucinate	Always verify critical information
Over-trusting AI	Errors slip through	You're responsible for the output
Under-using AI	Wasting time	Let AI handle repetitive tasks
Sharing sensitive data	Privacy risk	Be mindful of what you include
Not disclosing AI use	Policy violation	Follow organization's policies

AI Limitations Awareness

Know what AI can and cannot do well:

AI Strengths (Good For):

✅ Versatile language tasks (writing, editing, summarizing)
✅ Code analysis, generation, and debugging
✅ Pattern detection and consistency checking
✅ Learning from examples you provide
✅ Explaining complex concepts

AI Limitations (Be Careful):

⚠️ Knowledge cutoff - May not know recent events
⚠️ Hallucinations - Can confidently state incorrect info
⚠️ Context window limits - Can only consider so much at once
⚠️ Complex reasoning - Multi-step logic can have errors
⚠️ Personal decisions - Cannot make ethical judgments for you

Secret Weapon

If your prompt still feels incomplete, ask:

"Can you help me craft a more effective prompt for [goal]?"

AI can help improve your prompts! This meta-approach often yields better results.

Integration Points

`/prompt` Command (v2.1)

Step 0.11: Quick Delegation Check
Interaction Mode Detection
Expanded 9-criteria completeness check
Step 0.7: Post-Execution Evaluation
Common Mistakes section
AI Limitations awareness
Secret Weapon tip

`/prompt-hybrid` Command

Step 0.11: Quick Delegation Check
Full Delegation Assessment (Step 0.13)
Platform Awareness for agent spawning
Task Delegation recommendations
Diligence Reminder in Approval Gate

`/prompt-research` Command (v1.1)

Step 0.11: Delegation Assessment
Agency Mode focus for research tasks
Recommended delegation for AI vs human tasks

`/session-end` Command (v2.1)

Diligence Summary section
AI-generated content tracking
Deployment verification checklist

Benefits

Explicit Collaboration - Clear understanding of human vs AI roles
Better Outputs - Process and Performance descriptions improve quality
Quality Assurance - Discernment hints guide evaluation
Accountability - Diligence tracking ensures responsible use
Flexibility - Three interaction modes for different tasks

Reference

Based on Anthropic's AI Fluency Framework:

AI Fluency Key Terminology Cheat Sheet
The 4Ds: Delegation, Description, Discernment, Diligence
Human-AI Interaction Modes: Automation, Augmentation, Agency

Phase 0 - Prompt perfection flow
Hybrid Intelligence - Complexity detection
Learning System - Pattern tracking

AI Fluency Framework ​

What is AI Fluency? ​

The 4Ds ​

1. Delegation ​

2. Description ​

3. Discernment ​

4. Diligence ​

Human-AI Interaction Modes ​

Automation Mode ​

Augmentation Mode ​

Agency Mode ​

Configuration ​

Common Mistakes to Avoid ​

AI Limitations Awareness ​

Integration Points ​

/prompt Command (v2.1) ​

/prompt-hybrid Command ​

/prompt-research Command (v1.1) ​

/session-end Command (v2.1) ​

Benefits ​

Reference ​

Related ​

AI Fluency Framework

What is AI Fluency?

The 4Ds

1. Delegation

2. Description

3. Discernment

4. Diligence

Human-AI Interaction Modes

Automation Mode

Augmentation Mode

Agency Mode

Configuration

Common Mistakes to Avoid

AI Limitations Awareness

Integration Points

`/prompt` Command (v2.1)

`/prompt-hybrid` Command

`/prompt-research` Command (v1.1)

`/session-end` Command (v2.1)

Benefits

Reference

Related