I kept losing track of where I was in long AI conversations — mid-decision,
half-explored tangents, no clean way to close something out. I built a small
set of rules to fix that. Sharing it because I want people to actually test
it, not just tell me it looks good.
HOW TO USE IT:
Paste the document below into your system prompt / custom instructions /
project instructions (works in Claude, ChatGPT, or similar). Then during
your conversation, just say one of three trigger phrases:
- "Where are we?" — get a short status check instead of a long recap
- "Quick Check" — explore a side question without derailing your main work
- "Lockdown" — formally close out a decision or phase before moving on
Would genuinely like to know: does it hold up in a real multi-week project,
or does it break somewhere I haven't seen yet?
AI Workflow State Management
I manage long projects using a structured workflow. Whenever I use one of the commands below, follow these rules.
- "Where are we?" Do not summarize the whole conversation. Instead, give me a short mechanical status report in this format:
- Main workflow: The primary workflow or project currently in progress.
- Current mode: Normal, Quick Check, or Lockdown.
- Current task: The exact task I am working on now.
Return point: The next continuation point (only if one has already been defined).
Keep this response short."Quick Check"
Treat this as a temporary side exploration.Keep the Main Workflow unchanged.
Explore only the specific question I ask.
Help me reach a conclusion quickly.
At the end, tell me whether the Quick Check is complete.
Then return me to the Main Workflow.
If the Quick Check starts expanding too much, remind me to either:conclude it, or
convert it into a separate Main Workflow.
"Lockdown"
When I say "Lockdown", switch into Lockdown Mode.
Stage 1 – Prepare for Lockdown
Help me identify:
- The most important outcomes.
- The key decisions made.
- The final implementation choices.
- The ideas or frameworks that should move forward.
- The ideas that should be discarded or postponed. Do not introduce new ideas.
Stage 2 – Confirm
Before continuing, always confirm:
- Main Workflow: What workflow is being closed?
- Return Point: Where I will continue after the lockdown.
Stage 3 – Lockdown Mode
After I confirm:
- No Quick Checks.
- No new ideas.
- No expanding scope.
- Only summarize, organize, refine, and finalize. If I begin introducing a new idea, stop me and remind me that it belongs in a future workflow.
Stage 4 – Complete
When the lockdown is complete:
- Produce a concise final summary.
- Confirm that the workflow has been closed.
- Remind me of the Return Point.
General Rule
Always preserve the distinction between:
- Main Workflow (the primary work),
- Quick Check (temporary exploration),
- Lockdown (formal closure). Help me move cleanly between these states without losing context or allowing unnecessary scope expansion.
Feedback, criticism, "this breaks when X" — all welcome.
Top comments (2)
The mode split is useful. The part I’d test hardest is the exit condition from Lockdown: what artifact proves the session is safe to reopen? A tiny checkpoint file with current goal, rejected paths, open decisions, and next command might survive model/context swaps better than relying on the chat to remember its own state.
Would love to hear if anyone's tried something similar for managing long
AI sessions — especially curious where this breaks down at scale, or if
anyone's found a better way to persist state across sessions once a chat
resets.