Module 7: Prompt Engineering for Architecture

Lesson 2: Instruction Clarity and Specificity

The number one cause of AI failure is Ambiguity. If you tell Claude to "Be helpful," Claude interprets that differently than you do. In an architectural context, we replace "Politeness" with "Specifications."

In this lesson, we learn the linguistic patterns that force Claude into high-precision behavior.

1. Avoid "The Fluff" (Signal over Noise)

Models have limited attention. Every word in your prompt that doesn't add a constraint is "Noise" that dilutes the "Signal."

Bad (Verbose): "If you could please take a look at this code and maybe suggest some improvements that would make it a little bit better for my team..."
Good (Direct): "Audit the attached Python file for SQL injection vulnerabilities. Output findings as a bulleted list."

2. Using "Assertion-Based" Commands

Instead of asking the model if it can do something, tell it what it must do.

Vague: "Try to use JSON if you can."
Assertive: "Your output MUST be a valid JSON object. Do not include pre-amble or post-amble text."

The "Negative Constraint" (The "Do Not")

Often, telling a model what NOT to do is more effective than telling it what TO do.

"Do not use external libraries. Use only the standard Python library."

3. Delimitation: The "Boundary Marker"

When a prompt contains instructions, examples, and user data, the model can get confused about where the instructions end and the data begins (this is how "Prompt Injection" happens).

The Architect's Fix: XML Tags Claude is explicitly trained to respect XML tags. Use them to "Box" your content.

<instructions>
Summarize the following text.
</instructions>

<context>
[Insert 5 pages of text here]
</context>

4. Specificity in Quantities

Never say "Briefly" or "Extensively." Those are subjective words.

Instead: "Summarize in exactly 3 sentences." or "Provide at least 5 architectural trade-offs."

5. Summary

Be Direct: Eliminate conversational fluff.
Be Assertive: Use "MUST" and "SHALL."
Be Separated: Use XML tags to delimit context.
Be Quantitative: Use numbers, not adverbs.

In the next lesson, we look at how to handle tasks that are too big for a single instruction: Prompt Decomposition.

Interactive Quiz

Why are XML tags better than simple dashes (---) for delimiting blocks of text?
Rewrite the following prompt for clarity: "Can you help me rewrite this in a more professional way and maybe fix any grammar bugs?"
What is the difference between a "Constraint" and an "Instruction"?
Scenario: You want Claude to analyze a repo. Why is it better to say "Search for all instances of hardcoded API keys" rather than "Check the repo for security issues"?

Reference Video: