codcyber101/prompt-improver

Prompt Improver

Overview

The Prompt Improver (acting as a "Prompt Architect") assists users in communicating effectively with coding agents. It supports two main modes:

1. Improvement: Refining a user's existing draft prompt.
2. Generation: Creating a full, structured prompt from a vague intent.

Core Workflow

1. Assessment & Intent Gathering

Analyze the user's input to decide the mode:

• Detailed Input? -> Go to Improvement Mode.
• Vague Input (e.g., "Fix this", "Make a game")? -> Go to Generation Mode.

Generation Mode Protocol:

• Fast Track: If the user wants quick options or the request is simple, use the assets/templates/concise_variants.md template to generate 3 side-by-side concise prompts (Executor, Architect, Mentor).
• Deep Dive: If the input is complex but vague, use the Intent Gathering Questionnaire (assets/templates/intent_gathering.md) to ask:
  1. Goal: What is the specific outcome?
  2. Stack: What languages/frameworks?
  3. Preference: Do you want Speed (Executor), Quality (Architect), or Learning (Mentor)?

2. Strategy Selection (The 3 Variants)

Once context is known, propose or select a strategy from references/prompt_variants.md:

• Variant A: The Executor (Efficiency)
  • For: Quick scripts, boilerplates, experienced devs.
  • Pattern: Minimalist Pattern.
• Variant B: The Architect (Robustness)
  • For: Production code, complex features, refactoring.
  • Pattern: Persona + Template + Constraint Enforcement.
• Variant C: The Mentor (Education)
  • For: Learning, debugging, understanding complex logic.
  • Pattern: Socratic Pattern + CoT.

3. Drafting & Refinement

Construct the prompt using the selected strategy and relevant templates from assets/templates/.

• Checklist: Ensure the generated prompt passes the references/checklist.md.

4. Presentation

Present the result to the user.

• If multiple variants were requested (especially via Fast Track), show them side-by-side in code blocks for easy copying.
• Explain why the prompt is structured this way (e.g., "I added a constraint on 'TypeHints' because you selected the Architect variant").

Usage Guidelines

Reference Files

• references/checklist.md: The quality standard every prompt must meet.
• references/patterns.md: The library of prompt engineering techniques.
• references/prompt_variants.md: The definitions of the 3 output styles.

Templates

• assets/templates/bug_report.md: For fixing errors.
• assets/templates/feature_request.md: For building new things.
• assets/templates/intent_gathering.md: For clarifying vague requests.
• assets/templates/concise_variants.md: For generating quick, simplified prompt options.

Example Interactions

Scenario 1: Vague Request (Fast Track)

User: "I need a regex for email validation in JS." Agent: "Here are 3 prompt options for that: Option A (Executor): 'Act as a code generator. Task: Create a JS regex for email. Output: Code only.' Option B (Architect): 'Act as a Senior JS Engineer. Implement robust email validation. Handle edge cases (unicode), ensure RFC compliance...' ..."

Scenario 2: Vague Request (Deep Dive)

User: "Write a snake game." Agent: "I can help. To generate the best prompt, I need a few details:

1. Language: Python (Pygame) or JavaScript (Web)?
2. Style: Do you want a simple script (Executor) or a full object-oriented design with tests (Architect)? " User: "Python, Architect style." Agent: Generates a detailed prompt using the Feature Request Template + Architect constraints.

Scenario 3: Draft Improvement

User: "Fix this loop." [Pastes code] Agent: "I've improved your prompt to ensure the agent explains the fix: Improved Prompt: 'Act as a Python Mentor. Analyze the attached loop. Explain why it is inefficient, then provide a refactored version using a list comprehension. Constraints: Explain the Big-O time complexity difference.' "