Deep-dive into the three quality tools — how they work, their scoring systems, and how to get the most from them.

• Prompt Linter
• Prompt Optimizer
• Smart Recommender

CLI: prompt-lib lint Workshop: Tools tab → Linter Source: src/linter.js

The Prompt Linter analyzes any prompt against 14 quality rules and produces a score from 0 to 100, a letter grade (A–F), and prioritized suggestions for improvement.

score = (sum of passed rule weights / total weight) × 100

Total weight: 100 (sum of all 14 rule weights: 10+12+8+10+8+8+5+7+7+7+5+5+3+5)

Input prompt:

Can you please help me write something good about dogs? Please make it nice
and interesting. Please include some good information.

Lint result:

Score: 15/100 (Grade: F)
Rules: 2/14 passed | 23 words

✅ Passing:
   • Not excessively long
   • Specific language

💡 Suggestions to improve:
   → State the task explicitly: "Your task is to…"
   → Add a role definition like "You are a senior [role] expert…"
   → Specify the output format: "Respond in [format]"
   → Add context: "Context: …" or "Given the following information: …"
   → Add constraints: "Rules: …" or "Never do X" / "Always do Y"
   → Your prompt is quite short. Add more detail about what you want.
   → Add examples of expected input/output to guide the AI.
   → Organize your prompt with clear sections using headers or labels.
   → Specify the target audience: "This is for [audience]…"
   → Define the tone: "Use a [professional/casual/technical] tone."
   → Reduce "please" usage — direct instructions work better with AI.
   → Add a quality check: "Before responding, verify that…"

• Add a role definition (10 points): "You are a senior content writer"
• State the task clearly (12 points): "Your task is to write a 500-word blog post"
• Specify output format (10 points): "Respond in markdown with headings"
• Add constraints (8 points): "Never make up statistics"
• Add context (8 points): "Context: This is for a pet adoption website"
• Write at least 50 words (8 points)
• Include examples (7 points)
• Use section headers (7 points)

The linter auto-detects prompt types and adjusts rule weights:

For the complete weight multiplier table, see INFRASTRUCTURE.md.

CLI: prompt-lib optimize Workshop: Tools tab → Optimizer Source: src/optimizer.js

The Prompt Optimizer rewrites prompts using content-aware analysis. It detects the domain, audience, and intent, then applies targeted improvements specific to the prompt's actual content.

The optimizer processes your prompt through these steps in order:

Input prompt
    │
    ├── 1. Domain detection (7 domains)
    ├── 2. Filler removal (redundant phrases)
    ├── 3. Politeness reduction (excessive "please")
    ├── 4. Weak verb strengthening (hedging → direct)
    ├── 5. Vague language replacement (specific terms)
    ├── 6. Compound task decomposition (multi-task → steps)
    ├── 7. Domain-specific role injection
    ├── 8. Audience detection and tone setting
    ├── 9. Constraints addition (domain-specific)
    ├── 10. Output format specification
    ├── 11. Examples placeholder
    └── 12. Quality verification step
    │
    ▼
Optimized prompt + before/after scores + change log

The optimizer scans for keywords to identify one of 7 domains:

Each domain provides:

• A specific expert role (instead of generic "helpful assistant")
• Domain-relevant constraints
• Appropriate output format
• Domain-specific quality checks
• Example placeholders

The optimizer strips these filler phrases:

• basically, essentially, actually, literally, honestly
• obviously, clearly, simply put, in other words
• as you know, it goes without saying, needless to say
• I think, I believe, I feel like, in my opinion

These hedging patterns are removed or replaced with direct instructions:

Before (Score: 35/100, Grade: F):

Can you please help me write a good blog post about machine learning?
Please make it interesting and cover the basics. I want you to maybe
include some good examples if possible. Thanks!

After (Score: 88/100, Grade: B):

You are an experienced content strategist and professional writer with
expertise in clear, engaging communication.

TASK:
Write a blog post about machine learning. Cover the basics. Include
insightful and thought-provoking examples.

TARGET AUDIENCE:
This is for beginners.

TONE:
Be clear, engaging, and well-paced.

OUTPUT FORMAT:
Respond in well-structured prose with clear headings, short paragraphs,
and a logical narrative flow. Use markdown formatting.

RULES:
- Use clear, concise language appropriate to the target audience
- Maintain a consistent tone and voice throughout
- Structure content with logical flow and smooth transitions
- Support claims with specific details or examples

Before responding, verify that the writing is clear, free of jargon,
logically structured, and engaging for the target audience.

Changes made:

• Removed filler words and redundant phrases
• Replaced weak/hedging language with direct instructions
• Replaced vague language with specific terms
• Added domain-specific role (writing)
• Organized prompt into clear labeled sections
• Added writing-specific output format
• Added writing-specific quality constraints
• Added quality verification step

After optimization, toggle between:

• Optimized — clean optimized prompt
• Diff view — line-by-line color-coded comparison (🟢 added, 🔴 removed, unmarked = unchanged)

CLI: prompt-lib recommend <query> Workshop: Tools tab → Recommender Source: src/recommender.js

The Smart Recommender analyzes your natural-language description and suggests the best prompts from the library, including an optimal combination.

User description
    │
    ├── 1. Term extraction (split into searchable words)
    ├── 2. Intent detection (map to 8 intent categories)
    ├── 3. Prompt scoring (match terms against all prompts)
    ├── 4. Intent bonus (boost prompts in matching categories)
    └── 5. Combo building (system prompt + framework + template)
    │
    ▼
Top 8 matches + suggested combination

For each prompt in the library, the recommender calculates a relevance score:

Prompts in categories that match the detected intent get a bonus.

The recommender returns:

1. Top 8 matching prompts — ranked by relevance score
2. Suggested combo — the best combination of:
   • 🧠 System prompt — highest-scoring system prompt
   • 🔧 Framework — highest-scoring framework
   • 📝 Template — highest-scoring domain template
3. Categorized results — top 3 system prompts, top 3 frameworks, top 5 templates

Query: "I need to build a REST API with authentication"

Output:

Suggested combination:

  🧠 System prompt:  Coding Assistant (coding-assistant)
  🔧 Framework:      Chain-of-Thought (chain-of-thought)
  📝 Template:       API Design (api-design)

Top matching prompts:

  API Design
    slug: api-design | category: development | score: 68
  System Design
    slug: system-design | category: development | score: 43
  Coding Assistant
    slug: coding-assistant | category: system-prompts | score: 38
  ...

• Be descriptive: "I need to write marketing emails for a SaaS product" > "email help"
• Include domain keywords: "data analysis SQL dashboard" triggers the data intent
• Mention the end goal: "create a landing page that converts" > "make a web page"
• Combine contexts: "code review for Python REST API" triggers both coding and development

Navigation: ← AI Models Guide  |  API & Playground →