# Rules for ultra context, memories, lessons, scratchpad with plan and act modes

**URL:** <https://forum.cursor.com/t/rules-for-ultra-context-memories-lessons-scratchpad-with-plan-and-act-modes/48792>\
**Category:** Built for Cursor\
**Created:** [February 7, 2025, 12:21pm UTC](https://forum.cursor.com/t/rules-for-ultra-context-memories-lessons-scratchpad-with-plan-and-act-modes/48792 "2025-02-07T12:21:13Z")\
**Posts on this page:** 1\
**Showing post:** 21

<div class="post-metadata">

**Author:** ![bulletninja](https://avatars.discourse-cdn.com/v4/letter/b/bbce88/32.png) [@bulletninja](https://forum.cursor.com/u/bulletninja)\
**Post date:** [February 8, 2025, 9:17pm UTC](https://forum.cursor.com/t/rules-for-ultra-context-memories-lessons-scratchpad-with-plan-and-act-modes/48792/21 "2025-02-08T21:17:45Z")

</div>

That’s really interesting! I had previously thought about creating an ontology (see [here](https://en.wikipedia.org/wiki/Ontology_(information_science))) using JSON schemas to define certain concepts. For example, some aspects of your architecture could be represented as clearly defined schemas that you can reference later. However, I wasn’t systematic enough in my approach back then.

Now I’m wondering if this more dynamic, learning-oriented perspective could extend with more than just a summarization, as @bmadcode suggested. Perhaps it could lead to a more complete consolidation that includes stricter definitions using schemas or something similar. In theory, this would allow the agent to dynamically choose how to use knowledge, while at the same time providing us rigorous validators for free. I believe that some core, “firm” knowledge might really benefit from being represented in this more rigid format.

---

_[View the full topic](https://forum.cursor.com/t/rules-for-ultra-context-memories-lessons-scratchpad-with-plan-and-act-modes/48792)._