Prompt Details
Model
(claude-4-8-opus)
Token size
572
Example input
[DESCRIBE YOUR FILES OR PROJECT STRUCTURE, AND ANY OFF-TOPIC BLEED YOU HAVE SEEN]: A 30KB notes.md with project ideas, meeting notes, and personal goals; AI mixes personal goals into project proposals.
Example output
AUDIT
File: notes.md
Size: ~7,500 tokens (30KB)
Themes found: project ideas (40%), meeting notes (35%), personal goals (25%)
High-gravity tokens: "earn $10K/month", "get promoted by June"
Pollution risk: HIGH
Diagnosis: Personal goals use aspirational, emotional language that scores high in attention. Loaded next to project ideas, those tokens pull the tone and content of project proposals toward your personal ambitions.
RESTRUCTURE
Before: notes.md (3 themes mixed, HIGH)
After:
notes/
project-ideas.md - ideas and specs only
meeting-notes.md - meeting summaries only
personal-goals.md - personal items only
Loading strategy:
Writing a project proposal -> project-ideas.md + meeting-notes.md
Personal planning -> personal-goals.md ONLY
Never load personal-goals.md during work tasks.
HIGH-GRAVITY QUARANTINE
- "earn $10K/month" -> personal-goals.md ONLY
- "get promoted by June" -> personal-goals.md ONLY
LOADING PROTOCOL
Task | Load | NEVER load
Project proposal | project-ideas, meeting-notes | personal-goals
Personal planning | personal-goals | project-ideas
Fix the split first; it removes the bleed with zero prompt changes.
By purchasing this prompt, you agree to our terms of service
CLAUDE-4-8-OPUS
Your AI keeps mentioning things you told it to ignore — because "ignore this" does not work; a model attends to every token it loads. This prompt makes Claude a Context Pollution Auditor: give it your file structure and any off-topic bleed, and it finds the pollution, spots high-attention tokens (numbers, claims, phrases) that leak across topics, and returns a concrete restructure: one theme per file, plus a table of which files to read per task. An audit for YOUR project, not a tutorial.
...more
Added 1 week ago
