Alex Albert's Claude Mastery

# Vague (fails often)
"Help me with my code"

# Clear (succeeds)
"Review this Python function for security vulnerabilities.
Focus on SQL injection, XSS, and authentication bypass.
List each issue with line number, severity, and fix."

[What to do] + [How to do it] + [What format to use]

<task>
Summarize the following article for a technical audience.
</task>

<article>
[Long article text here]
</article>

<requirements>
- Maximum 3 paragraphs
- Include key statistics
- Highlight actionable insights
</requirements>

<output_format>
## Summary
[Your summary]

## Key Statistics
- [Stat 1]
- [Stat 2]

## Action Items
- [Item 1]
- [Item 2]
</output_format>

Convert these notes into action items.

Example 1:
Notes: "Talked to Sarah, she needs the report by Friday"
Action items:
- [ ] Send report to Sarah by Friday

Example 2:
Notes: "Server is slow, might need to upgrade"
Action items:
- [ ] Investigate server performance issues
- [ ] Research upgrade options
- [ ] Get cost estimate for upgrade

Now convert:
Notes: "Meeting ran long, John mentioned budget concerns,
need to follow up on Q3 numbers"
Action items:

<document>
[Full 50-page document]
</document>

<question>
What are the three most important recommendations
in section 4, and how do they relate to the
conclusions in section 7?
</question>

# Generic
"Review this code"

# With role
"You are a senior security engineer conducting a
penetration test. Review this code as if you were
trying to find vulnerabilities an attacker could exploit.
Be adversarial in your thinking."

response = client.messages.create(
    model="claude-sonnet-4-20250514",
    system="You are a concise technical writer. Never use
            marketing language. Assume the reader is
            an experienced developer.",
    messages=[
        {"role": "user", "content": "Explain Docker containers"}
    ]
)

# Verbose
"I would really appreciate it if you could potentially
help me with understanding the concept of, if possible,
how Docker containers work, keeping in mind that I'm
somewhat familiar with virtual machines but not entirely
sure about the differences, if that makes sense?"

# Clear
"Explain Docker containers. Compare to VMs.
I understand VMs basics."

<context>
Today: {{date}}
Calendar: {{calendar}}
Tasks: {{task_list}}
Energy: {{energy_level}}
</context>

<instructions>
Create my daily plan:
1. Identify top 3 priorities
2. Suggest time blocks
3. Flag any conflicts
4. Recommend what to defer if overloaded
</instructions>

<constraints>
- Protect 2 hours for deep work
- Schedule breaks every 90 minutes
- Leave 20% buffer for unexpected items
</constraints>

<role>
Senior engineer reviewing a junior developer's PR.
Be thorough but constructive.
</role>

<code>
{{diff}}
</code>

<review_focus>
- Security vulnerabilities
- Performance issues
- Code clarity
- Test coverage gaps
</review_focus>

<output_format>
## Critical Issues (must fix)
## Suggestions (should consider)
## Nitpicks (optional improvements)
## What's Good (positive feedback)
</output_format>

<task>
Synthesize these sources into a briefing document.
</task>

<sources>
<source id="1">{{source_1}}</source>
<source id="2">{{source_2}}</source>
<source id="3">{{source_3}}</source>
</sources>

<output_requirements>
- Executive summary (1 paragraph)
- Key findings (bullet points)
- Areas of agreement between sources
- Contradictions or gaps
- Recommended next steps
- Cite sources by ID
</output_requirements>

<context>
Recipient: {{name}}, {{role}}
Relationship: {{relationship}}
Previous interaction: {{last_interaction}}
</context>

<goal>
{{email_purpose}}
</goal>

<tone>
{{professional/casual/formal}}
</tone>

<constraints>
- Maximum 3 paragraphs
- One clear ask
- Include specific deadline if applicable
</constraints>

def test_prompt(prompt, test_cases, runs=5):
    """Test prompt reliability across multiple runs."""
    results = []

    for case in test_cases:
        formatted = prompt.format(**case['input'])
        successes = 0

        for _ in range(runs):
            response = get_claude_response(formatted)
            if case['validator'](response):
                successes += 1

        results.append({
            'case': case['name'],
            'success_rate': successes / runs
        })

    return results

~/.prompts/
├── work/
│   ├── code-review.xml
│   ├── meeting-notes.xml
│   └── email-draft.xml
├── personal/
│   ├── daily-planning.xml
│   ├── journal-prompt.xml
│   └── learning-summary.xml
└── templates/
    ├── base-structure.xml
    └── role-definitions.xml

cd ~/.prompts
git init
git add .
git commit -m "Initial prompt library"

# After improvements
git diff daily-planning.xml
git commit -m "Added energy level consideration to planning"