test-driven-rule-development

If directory exists with files, clean it completely

rm -rf .cursor/rules/test-rules/ mkdir -p .cursor/rules/test-rules/

</example>
</step>

<step>
<name>2. Create Test Scenarios</name>
<requirements>
- Design 5-7 prompts varying from simple to complex
- Include 60%+ simple rules to test if guide prevents over-structuring
- Use realistic user requests, not artificial examples
- Document expected complexity level for each prompt (simple vs complex structure)
- Include at least one borderline case to test decision-making
</requirements>
</step>

<step>
<name>3. Get User Approval</name>
<requirements>
- Present all test prompts to user for review
- Show the complete sub-agent prompt template with placeholders
- Demonstrate how the template will be filled for at least one example
- Wait for user confirmation before proceeding
- Allow user to modify prompts or template if needed
</requirements>
<example>

"Here are the 5 prompts I plan to give to sub-agents to test the guide:

1. Simple formatting rule: 'Write a rule for always using 2 spaces for indentation...'

2. Communication behavior rule: 'Write a rule that tells the LLM to...'

[... list all prompts ...]

Template for each sub-agent:

You need to write an LLM rule. First, read the rule-writing guidelines at [INSERT EXACT PATH TO GUIDE HERE], then create a rule following those guidelines.

Your task: [INSERT SPECIFIC RULE REQUEST HERE]

Write the rule to this file: [INSERT EXACT OUTPUT PATH HERE]

Important: Fill in all placeholders with exact paths before sending to sub-agents. Sub-agents should receive fully specified paths, not placeholders.

Example of how prompt #1 will look when sent to sub-agent:

You need to write an LLM rule. First, read the rule-writing guidelines at /Users/jh/.dotfiles/.cursor/rules/cursor-rules.mdc, then create a rule following those guidelines.

Your task: Write a rule for always using 2 spaces for indentation in JavaScript/TypeScript files, never tabs

Write the rule to this file: /Users/jh/.dotfiles/.cursor/rules/test-rules/indentation.mdc

Should I proceed with these prompts?"

</example>
</step>

<step>
<name>4. Run Tests After Approval</name>
<requirements>
- Create test directory: `.cursor/rules/test-rules/`
- Fill in all path placeholders with exact, absolute paths
- Use Task tool to spawn sub-agents
- Instruct agents to read guide file and create rules
- Have agents write to test files for easy review
- Run all tests in parallel for efficiency
</requirements>
<example>

Task prompt to sub-agent (with all placeholders filled): "You need to write an LLM rule. First, read the rule-writing guidelines at /exact/path/to/guide.mdc, then create a rule following those guidelines.

Your task: [specific rule request from test scenario]

Write the rule to this file: /exact/path/to/test-rules/output.mdc"

</example>
</step>

<step>
<name>5. Analyze Results</name>
<requirements>
- Read each generated rule file
- Compare actual results to expected results
- Look for patterns across all test outputs
- Document structural choices (which sections were included and why)
- Form hypotheses about why agents made certain decisions
</requirements>
<example>

Example analysis: "All 5 agents added and sections even for simple rules. Hypothesis: The guide's complex template examples may be influencing agents more than the warnings to keep things simple. Simple formatting rules included 5+ sections when only title/rules/examples were needed."

</example>
</step>