- 29/10/2025 - The Freedom Hub café (in waterloo)
	- BugBot, PR, 
	- Chat takes up full screen in v2.0… no more code shown if you choose to
	- Rules in Cursor
		○ Get consistent results from the models
		○ They are non-deterministic. Submitting 5 same prompts 5 times and you get different outputs
		○ You can add a Rule to Cursor which get attached to every Prompt in Cursor
			§ eg. Rule: This codebase is written in Python 3.11.
			§ Add the language youre using, the DB
			§ Ask the model if it needs help… then add rules
			§ Variables should be camelcase
			§ Use spaces instead of tabs
			§ Instead of making assumptions ask me questions to clarify
		○ Good idea to have the prompt to audit files and apply the rules to it…
	- Prompt engineering
	- Context engineering
		○ This is the short-term memory for the model
		○ In the Chat pane if you create a new tab it’s a new context window
		○ Every new tab Cursor will submit some info to the new tab to provide it context
		○ It’s a good idea to fill the new tab with useful context before asking it to begin coding
		○ "please explore the codebase" is a good way to build context in a new chat tab/window
		○ 3 dot menu has a Chat Export button so you can pull out an existing chat and have it summarise in the next 
		○ "I want to continue working this tomorrow but lets pretend tonight I had amnesia and forget everything. write a prompt for tomorrow so the chat is able to pick up where it left off"
	- Use new tabs for new topics. Try to keep it tight
		○ The larger the context window gets the lower the accuracy of the model.
		○ The more you jump around to different topics the harder the model is able to keep up
	- Claude 4.5-sonnet is the best for coding
	- GPT-5 is decent for coding. About as good as Claude, but needs more guidance.



Privacy of Cursor. What info is stored or used?


Workshop with Nick from Cursor (LA, CA, USA)
	- User LLM to discuss app
	- SPEC  > CHECKLIST > BUILD
		○ You can use this using Plan Mode in Cursor
		○ Itll walk you through like a wizard. Q&A
		○ Once submitted itll crearte  a plan and .md
		○ Along with a task list
		○ can take up to 20 mins to think and build
		○ Try to remain specific. If you want to include tests and pipeline etc… do it in a separate plan window
		
	- Get it to write a list in point form about what to build
	- Get it to check off tasks as it goes
	- BEst-of-N
		○ You can have multiple models run the same prompt and compare them and choose one
		○ Can be exxy
		○ Use an agent to looks at each of the model outputs to review them
	- Browser Use
		○ Agent be connected to a browser in Cursor
		○ Integration with Console in the browser (Dev tools)
	- Voice Input
		○ Manage agents with voice
	- Hooks
		○ Custom script can be hooked into Cursor
		○ You can run a script before running a prompt (for example)
	- USE PRIVACY MODE!
	

Workshop - Context Engineering
Presented by Tabish Fission
	- RESEARCH > PLAN> IMPLEMENT
		○ This is what we did prior to Plan Mode coming out
		○ Kiro (amazon). This is an IDE which has wide mode and spec mode. Cursor is doing this now
			§ Spec Mode - You describe the requirements to the LLM
		○ GitHub now has spec-kit, which is the same/similar thing
		○ OpenSpec - an open source tool which does this too
			§ New version recently released
	- OpenSpec was created by the presenter
		
		
	- What is a Spec?
		○ The what? the why? and the how?
		○ Its easier to iterate on spec than it is to iterate on code. So make sure you get the SPECS right!
	- OpenSpec Demo
		○ website: openspec.dev
		○ AGENTS.md - instructions for the Agent in Cursor
		○ project.md - described the project. Architectural patterns and other things
		○ specs/openspec-proposal.md - 
		○ specs/openspec-apply.md
		○ specs/openspec-(??)
		○ contents of specs folder is updated as the project is built and work is done
		○ USING OPENSPEC: In cursor chat window you type: /openspec-proposal "make a Snake style game in Python"
			§ This builds out the spec files which can then be used to by Cursor to build the code
			§ 
		○ When you are ready to implement you can use: /openspec-apply "implement the specs are you’ve created"