Prompt Architect

A comprehensive framework for creating, analyzing, and refining prompts for AI language models using evidence-based techniques, structural optimization principles, and systematic anti-pattern detection.

Overview

Prompt Architect provides a systematic approach to prompt engineering that combines research-backed techniques with practical experience. Whether crafting prompts for Claude, ChatGPT, Gemini, or other systems, this skill applies proven patterns that consistently produce high-quality responses.

This skill is particularly valuable for developing prompts used repeatedly, troubleshooting prompts that aren't performing well, building prompt templates for teams, or optimizing high-stakes tasks where prompt quality significantly impacts outcomes.

When to Use This Skill

Apply Prompt Architect when:

• Creating new prompts for AI systems that will be used repeatedly or programmatically
• Improving existing prompts that produce inconsistent or suboptimal results
• Building prompt libraries or templates for team use
• Teaching others about effective prompt engineering
• Working on complex tasks where prompt quality substantially impacts outcomes
• Debugging why a prompt isn't working as expected

This skill focuses on prompts as engineered artifacts rather than casual conversational queries. The assumption is you're creating prompts that provide compounding value through repeated or systematic use.

Core Prompt Analysis Framework

When analyzing existing prompts, apply systematic evaluation across these dimensions:

Intent and Clarity Assessment

Evaluate whether the prompt clearly communicates its core objective. Ask:

• Could someone unfamiliar with context understand what task is being requested?
• Are success criteria explicit or must the AI infer what constitutes a good response?
• Is there ambiguous phrasing that could be interpreted multiple ways?
• Does the prompt state its goal unambiguously?

Strong prompts leave minimal room for misinterpretation of their central purpose.

Structural Organization Analysis

Evaluate how the prompt is organized:

• Does critical information appear at the beginning and end where attention is highest?
• Are clear delimiters used to separate different types of information?
• Is there hierarchical structure for complex multi-part tasks?
• Does organization make the prompt easy to parse for both humans and AI?

Effective structure guides the AI naturally through the task.

Context Sufficiency Evaluation

Determine whether adequate context is provided:

• Are there implied assumptions about background knowledge?
• Are constraints, requirements, and edge cases explicitly stated?
• Does the prompt specify audience, purpose, and contextual factors?
• Is necessary background information included or assumed?

Strong prompts make required context explicit rather than assuming shared understanding.

Technique Application Review

Assess whether appropriate evidence-based techniques are employed:

• For analytical tasks: Are self-consistency mechanisms present?
• For numerical/logical problems: Is program-of-thought structure used?
• For complex multi-stage tasks: Is plan-and-solve framework present?
• Are techniques appropriate to the task type?

Different task categories benefit from different prompting patterns.

Failure Mode Detection

Examine for common anti-patterns:

• Vague instructions that allow excessive interpretation
• Contradictory requirements
• Over-complexity that confuses rather than clarifies
• Insufficient edge case handling
• Assumptions that may not hold across all expected uses

Identify what could go wrong and whether guardrails exist.

Formatting and Accessibility

Evaluate presentation quality:

• Do delimiters clearly separate instructions from data?
• Does visual hierarchy aid understanding?
• Is whitespace, headers, and structure used effectively?
• Is the prompt accessible to both AI systems and human maintainers?

Good formatting enhances both machine and human comprehension.

Prompt Refinement Methodology

When improving prompts, follow this systematic approach:

1. Clarify Core Intent First

Begin by ensuring the central task is crystal clear:

• Rewrite primary instruction using specific action verbs
• Replace abstract requests with concrete operations
• Add quantifiable parameters where appropriate
• Make success criteria explicit

A refined prompt should leave no doubt about its fundamental purpose.

2. Restructure for Attention and Flow

Apply structural optimization:

• Move critical instructions and constraints to beginning and end
• Organize complex prompts hierarchically
• Use formatting and delimiters to create visual structure
• Ensure logical progression through the task

Each section should build naturally on previous ones.

3. Add Necessary Context

Enrich prompts with previously implicit or missing context:

• Specify audience, purpose, and situational factors
• Define ambiguous terms or concepts
• Establish constraints and requirements explicitly
• Provide background needed to understand task significance

Make assumptions explicit rather than hidden.

4. Apply Evidence-Based Techniques

Incorporate research-validated patterns:

• Self-Consistency: For factual/analytical tasks, request validation from multiple perspectives
• Program-of-Thought: For logical tasks, structure step-by-step explicit reasoning
• Plan-and-Solve: For complex workflows, separate planning from execution
• Few-Shot Examples: Provide concrete examples of desired input-output patterns
• Chain-of-Thought: Request explicit reasoning steps for complex problems

Match techniques to task requirements.

5. Build in Quality Mechanisms

Add self-checking and validation:

• Include verification steps in multi-stage processes
• Specify quality criteria for outputs
• Request explicit uncertainty acknowledgment when appropriate
• Build in sanity checks for analytical tasks

Quality mechanisms increase reliability and reduce errors.

6. Address Edge Cases and Failure Modes

Anticipate and handle potential problems:

• Identify likely edge cases and specify handling
• Include fallback strategies for error conditions
• Use negative examples to illustrate what to avoid
• Make explicit any assumptions that might not hold

Proactive edge case handling prevents common failures.

7. Optimize Output Specification

Be explicit about desired output format:

• Specify structure (prose, JSON, bullet points, etc.)
• Define required components and their order
• Indicate appropriate length or detail level
• Clarify how to handle uncertainty or incomplete information

Clear output specification prevents format ambiguity.

Evidence-Based Prompting Techniques

Self-Consistency

For tasks requiring factual accuracy or analytical rigor, instruct the AI to:

• Consider multiple perspectives or approaches
• Validate conclusions against available evidence
• Flag areas of uncertainty explicitly
• Cross-check reasoning for internal consistency

Example addition to prompt: "After reaching your conclusion, validate it by considering alternative interpretations of the evidence. Flag any areas where uncertainty exists."

Program-of-Thought

For mathematical, logical, or step-by-step problem-solving tasks:

• Structure prompts to encourage explicit step-by-step thinking
• Request showing work and intermediate steps
• Break complex operations into clear substeps
• Have the AI explain its reasoning at each stage

Example structure: "Solve this problem step by step. For each step, explain your reasoning before moving to the next step. Show all intermediate calculations."

Plan-and-Solve

For complex multi-stage workflows:

• Separate planning phase from execution phase
• Request explicit plan before beginning work
• Build in verification after completion
• Structure as: Plan → Execute → Verify

Example structure: "First, create a detailed plan fo