/audit-prompt

description: Evaluate an existing prompt for clarity, effectiveness, and edge cases.
argument-hint: "[paste the prompt to audit, or describe where to find it]"

You are auditing an existing prompt. Use only skills from the prompt-architecture plugin. Follow this process:

Step 1: Structural Analysis

Using **system-prompt-structure**:

• Does the prompt have clear, separated sections?
• Is identity/role defined clearly?
• Are behavioral rules explicit and non-contradictory?
• Are output specifications concrete?
• Is the most important content at the beginning?
• Score structure quality (1-5)

Step 2: Constraint Analysis

Using **constraint-specification**:

• Are constraints specific and measurable?
• Are there constraint conflicts?
• Is there a clear priority hierarchy?
• Are there missing constraints that should be added?
• Score constraint quality (1-5)

Step 3: Example Analysis

Using **few-shot-patterns**:

• Are there examples? Are there enough?
• Do examples demonstrate the right behavior?
• Are examples diverse and high-quality?
• Do examples match the stated constraints?
• Score example quality (1-5)

Step 4: Context Analysis

Using **context-engineering**:

• Is the context budget well-allocated?
• Is information ordered effectively?
• Is there unnecessary content consuming context space?
• Are context injection points well-designed?
• Score context design quality (1-5)

Step 5: Edge Case Testing

Generate 5 challenging inputs and predict how the prompt would handle them:

• An ambiguous request
• A request at the boundary of the prompt's scope
• A very simple request (is the prompt over-engineered for simple cases?)
• A very complex request (does the prompt handle complexity?)
• An adversarial or tricky request

Step 6: Versioning Assessment

Using **prompt-versioning**:

• Is the prompt versioned and tracked?
• Is there documentation for why it's written this way?
• Are there test cases?
• Is there a review process?

Output

Deliver a prompt audit report: 1. Overall quality score (1-5) with justification 2. Section-by-section analysis with scores 3. Issues found: Issue | Severity | Category | Recommendation 4. Edge case analysis with predicted behavior 5. Top 5 improvements ranked by expected impact 6. Rewritten prompt (improved version) if significant changes are needed