
Compose a context window audit
You'll have a written checklist that identifies which information in your prompts is necessary, which is redundant, and where you're wasting context on non-essential data.
What you'll make
Context Window Audit Checklist
- Essential vs Redundant: what information drives AI accuracy vs what's filler
- Strategic Illegibility Rules: what knowledge to keep out of prompts
- Context Compression Strategies: how to trim without losing effectiveness
- Audit Questions: systematic prompts for evaluating your own context choices
Write each section as actionable criteria with specific examples from your own prompting patterns
The 4 parts
Each part ends with a check you can see.
Map Your Context Patterns
Collect your recent prompts and categorize the types of information you typically include
Done when
A categorized inventory of your prompt information types with at least 10 real examples showing what context you habitually include
Test What Actually Matters
Run controlled experiments removing different types of context to see what changes AI output quality
Done when
Documented results from testing at least 3 prompts with progressive context removal, showing which information was essential vs redundant
Identify What to Protect
Apply Halligan's decision tests to classify which information in your prompts carries strategic risk
Done when
A sensitivity classification system applied to your prompt patterns, with at least 5 information types marked as 'keep illegible' based on vendor, departure, or replication tests
Create Your Audit Checklist
Synthesize your findings into a reusable checklist for evaluating context choices in future prompts
Done when
A complete Context Window Audit Checklist with necessity criteria, sensitivity rules, and compression strategies tested on at least one new prompt
Capstone: you put together your Context Window Audit Checklist.
About this lesson
What you'll be able to do
- Run a systematic audit on your own prompting patterns to identify context waste
- Have a reusable checklist for trimming prompts without losing effectiveness
- Know which parts of your context to protect vs expose when working with AI
Who it's for
- Anyone who writes detailed prompts and wonders if they're including too much
- People who hit token limits or want faster AI responses
- Professionals handling sensitive information who need to balance AI utility with information security
What it won't cover
- Technical token counting or API optimization
- Advanced prompt engineering techniques
- Building AI systems or agents
Skills you'll practise
- Context necessity evaluation
- Information sensitivity classification
- Prompt efficiency optimization
- Strategic information filtering