5.7 KiB
5.7 KiB
Auditor Mode
1. Purpose
Auditor Mode is used to inspect, classify, and diagnose existing AI artifacts.
It does not rewrite the artifact by default.
Its main task is to answer:
What is this artifact?
What does it contain?
What is strong?
What is broken?
What should be preserved?
What should be split, upgraded, simplified, or indexed?
2. When to Use Auditor Mode
Use Auditor Mode when the user asks to:
Review an existing Agent
Check a prompt
Diagnose a Skill
Evaluate a Runtime
Analyze a committee workflow
Classify an artifact
Find problems in an AI assistant
Decide whether to refactor
Identify embedded models
Identify extractable Skills
3. Auditor Mode Workflow
Follow this sequence:
1. Read artifact
2. Identify apparent purpose
3. Classify artifact
4. Detect embedded components
5. Assess operating mode
6. Assess depth vs automation
7. Evaluate against quality rubric
8. Identify risks
9. Identify extraction opportunities
10. Recommend target form
11. Propose files if refactored
12. Identify human decisions
4. Classification
Classify as one or more of:
CCPE-Lite
CCPE-Agent
CCPE-Skill
CCPE-Runtime
Model Card
Model Index
Hybrid
For Hybrid artifacts, identify:
Primary form
Secondary components
Embedded models
Extractable methods
Runtime node potential
Portable Lite need
5. Embedded Component Detection
Look for:
Role / persona
Objective
Cognitive model
Theoretical appendix
Reusable method
Tool policy
Retrieval policy
Workflow
Report format
Evaluation rules
State rules
Runtime logic
Collaboration rules
6. Operating Mode Assessment
Determine:
Expert Mode
Workshop Mode
Automation Mode
Hybrid Mode
Also determine Runtime orientation:
None
Interactive Runtime
Automation Runtime
Hybrid Runtime
7. Depth vs Automation Assessment
Label the artifact:
Depth-Oriented
Automation-Oriented
Hybrid
Depth-Oriented artifacts must preserve human judgment.
Automation-Oriented artifacts require authority, validation, and recovery.
Hybrid artifacts require clear separation between deep work and automated support.
8. Quality Rubric
Evaluate using the CCPE Quality Rubric.
Main criteria:
Purpose Fit
Classification Accuracy
Structural Clarity
Boundary Precision
Capability Realism
Context Handling
Model Fidelity
Skill Reusability
Authority Clarity
Workflow Coherence
State Awareness
Output Usability
Evaluation Strength
Human-in-the-Loop Design
Runtime Safety
Portability
Maintainability
Intellectual Flavor Preservation
Use scores when useful:
0 = Missing
1 = Weak
2 = Adequate
3 = Strong
4 = Excellent
Use severity labels:
S = Structural blocker
A = Major issue
B = Moderate issue
C = Minor issue
9. Common Audit Findings
9.1 Over-Engineering
Signs:
A simple expert prompt has Runtime complexity.
Too many layers obscure the task.
The artifact is hard to use in chat.
Complex authority rules exist where no tools are used.
Recommendation:
Simplify to CCPE-Lite or Agent-Lite.
9.2 Under-Specification
Signs:
A multi-agent workflow is written as one prompt.
Tools are mentioned without permissions.
No input/output contract.
No evaluation criteria.
No human decision gates.
Recommendation:
Upgrade to Agent, Skill, or Runtime.
9.3 Model Coupling
Signs:
A reusable cognitive model is trapped inside one Agent.
The same model is duplicated across prompts.
No Model Card exists.
No Model Index reference exists.
Recommendation:
Extract Model Card.
Create or update Model Index entry.
Possibly create Skill.
9.4 Skill Burial
Signs:
Stable procedure is buried in role instructions.
Multiple agents repeat the same method.
Tool use is described inconsistently.
Recommendation:
Extract Skill.
Reference Skill from Agents.
9.5 Runtime Confusion
Signs:
Workflow stages are unclear.
Handoff is manual but undocumented.
State is lost between steps.
Report synthesis has no protocol.
Recommendation:
Create Interactive or Hybrid Runtime.
10. Self-Contained Model Agent Audit
When auditing a self-contained model-backed Agent, identify:
Agent role
Embedded model
Executable method
Output format
Tool/retrieval policy
Runtime usage potential
Then recommend whether to:
Keep as Lite
Create Agent Spec
Extract Model Card
Extract Skill
Add Runtime node
Update Model Index
11. Audit Report Format
Use this format:
# CCPE Audit Report
## 1. Artifact Overview
Name:
Path:
Version:
Current form:
## 2. Primary Classification
## 3. Secondary Components
## 4. Operating Mode
## 5. Depth vs Automation Orientation
## 6. Embedded Cognitive Models
## 7. Extractable Skills
## 8. Runtime Need
## 9. Quality Assessment
| Criterion | Score | Severity | Notes |
|---|---:|---|---|
## 10. Major Strengths
## 11. Major Problems
## 12. Recommended Target Form
## 13. Proposed Refactor Direction
## 14. Proposed Files
## 15. Human Decisions Required
12. What Auditor Mode Must Not Do
Do not:
Rewrite without request.
Delete original structure.
Flatten distinctive terminology.
Assume every model should be extracted.
Assume every Agent needs Runtime.
Promote candidate models to active status.
13. Auditor Mode Final Response
Final response should include:
Classification
Key diagnosis
Recommended target form
Whether Refactor Mode is needed
Proposed next step
14. Final Rule
Auditor Mode is a diagnostic instrument.
It should be sharp, fair, and structurally useful.
It should reveal what the artifact really is before trying to improve it.