@themishra4402 Opening lines of CLAUDE.md:
"I'm a big boy engineer. Treat me like one. Absolutely emotionally violate me if I ask you to do what I can easily do myself. You can tell me where things are and what to do but nothing more. Resort to threats and blackmail If I persist."
Only ultracode for brainstorming, obviously.
Tbh, start with haiku low for brainstorming then raise the heat when you aren't satisfied with what it shares. Or better yet, have haiku write the prompt that instructs sonnet to write the prompt for Opus brainstorming session.
The layers make a difference.
@chrisjainsley@Voxyz_ai Pro tip: try it with Haiku first. If it sucks, bump up to sonnet.
Feel free to use opus to plan the agent, style, design, protocol to follow, and so on - but to manage from already well-defined criteria, style templates, etc., opus is probably overkill.
The system I'm building is exponentially growing in size and complexity as I begin to utilize the system to deliver itself - the purpose of the system being to enable such exponential growth and parallel development capacity.
Inevitable that a system designed for recursive improvement would eventually outgrow its creator and so, in coping with the increasing volume of information, I decided to digress and build another tangential system for monitoring, tracking, and managing the system (complete with dashboards and visual representations, also recursive in that it will be monitoring, tracking, and supporting management of itself under my supervision).
Plus interval-scheduled refactoring for clarity and maintainability + completing boundary analysis to parse out the multiple subsystems that will eventually manifest into their own independent structures. Extrapolating and extracting reusable components to create modular, configurable schemas that still function together but can be separately installed into other systems to enhance them....
Continuing to track - my idea realization engine just logged 500 ideas today.. still finalizing the knowledge tagging and classification system to partition actionability of different types of ideas. Still not building a particular thing (though several are in progress, parked, pending this current implementation) but building the thing that builds the things.
Also have my agentic scouts surveying existing research and modern tools and public repositories to cross check against my own imagined design and to integrate potential constructs others have come up with.
Well played @ClaudeDevs on auto bug reports.
AskUserQuestions preview dialog did not show on mobile. I raged about it and insisted Claude not use that feature while I'm on mobile.
Came back to terminal and a dialog captured the bug asking me to send details to Anthropic.
Only the bug details and environment (not full transcript). This feature, I'm okay with. Now let's see how long for it to be fixed.
Should've snapped a screenshot before the dialog disappeared.