Context hygiene for serious ChatGPT users.
Context Lab is an inspectable prompt workflow for auditing, cleaning, and operationalizing ChatGPT context.
It is built for ChatGPT users, operators, founders, executives, consultants, creators, researchers, students, and knowledge workers who want better personalized output without stale or over-saved memory.
It helps users build a cleaner operating profile, decide what should and should not be saved to memory, and turn that context into reusable workflows for strategy, writing, research, planning, and execution.
It is not a prompt pack. It is not a memory dumping system. It is not a claim that ChatGPT knows you accurately. It is a context hygiene workflow.
If you want the full workflow, open a new ChatGPT chat and paste Prompt 01:
If you only want to clean up memory, use the fast path:
1 → 4 → 7
I want to run Context Lab. Start with Prompt 01: Context Audit. Audit your current context of me and my work, separate confirmed facts from assumptions and unknowns, and identify what should be kept, corrected, clarified, or ignored.
For a concrete walkthrough, see the End-to-End Demo.
Most ChatGPT personalization fails because the model's context is incomplete, stale, tactical, over-saved, or unsupported.
Bad context creates bad advice:
- fake personalization
- stale assumptions
- overconfident recommendations
- generic strategy
- bloated memory
- tactical details treated as durable facts
Context Lab forces the model to separate confirmed facts, reasonable inferences, unknowns, stale assumptions, and high-risk guesses before relying on them.
Memory is for stable context.
Working documents are for live operating state.
Do not save tactical details, active deals, current priorities, project status, or temporary plans as durable memory.
After the full workflow, you should have:
- A strict audit of what ChatGPT thinks it knows about you
- A corrected operating profile
- A conservative memory cleanup plan
- A clear separation between durable memory and live working context
- A reusable AI workflow library
- Master prompts for future conversations
- A memory execution confirmation or Memory Execution Sheet
- Context Audit
- Gap Interview
- Operating Profile
- Memory Cleanup
- AI Workflow Plan
- Master Prompt
- Memory Execution or Handoff
If you only want to clean up ChatGPT memory, use this shorter path:
- Run Context Audit
- Run Memory Cleanup
- Run Memory Execution or Handoff
Use the full seven-step workflow when you want a complete operating profile, reusable workflows, and master prompts.
Open a new ChatGPT conversation and paste the prompt in prompts/01-context-audit.md.
After each response, continue through the seven prompts in order. Correct the model when it guesses, flatters, or treats stale context as current. Prompt 07 is capability-dependent: it either applies the approved cleanup plan where durable saved memory management is available or produces a Memory Execution Sheet for manual use or a memory-enabled chat.
For a more detailed walkthrough, see Quickstart.
Prompt 07 does not assume every ChatGPT chat can directly modify saved memory. It checks capability first, then either executes the approved cleanup plan or produces a Memory Execution Sheet.
- Context audit prompt
- Gap interview prompt
- Operating profile builder
- Memory cleanup workflow
- AI workflow planner
- Reusable master prompt generator
- Memory execution or handoff prompt for approved memory cleanup plans
- Templates for saved context documents
- Fictionalized before/after memory cleanup examples
- Optional role-specific examples
- End-to-End Demo
- Markdown-only workflow
- No automated memory API
- Direct memory execution depends on the current chat's capabilities
- No model-specific validation across ChatGPT, Claude, Gemini, or other models
- No CLI yet
- Output quality depends on how honestly the user corrects the model
- Sensitive context still requires user judgment
prompts/: the core seven-step prompt sequencetemplates/: reusable blank context documentsdocs/: principles, memory rules, fast paths, and failure modesexamples/: fictionalized example outputs for adaptation, including the End-to-End Demo.github/: issue templates for bugs and workflow suggestions
- Harness Lab: inspectable YAML and Markdown runbooks for AI workflows.
- LLM Council: multi-model decision support with anonymous peer review.
- Clip Lab: local-first long-form to short-form video pipeline.
MIT