FYI you can modify the system prompt using mitmproxy. Just ask your agent to walk you through it. Anthropic system prompts are gnarly and geared towards the lowest common denominator.
there is literally a --system-prompt flag for claude code. In my mitmproxy experiments using that flag appended to the existing system prompt rather than replacing it. So I had to create a little helper to strip the system prompt sent over the wire and add my own.
Use --system-prompt-file, it will replace the whole system prompt (https://code.claude.com/docs/en/cli-reference ). You'll have to use a shell alias or function or something to always append this flag when calling Claude Code but you don't need any MitM shenanigans. Then use "/context all" to see what else is sent (here I would recommend MitM'ing since Claude Code won't show the exact tools and text), there are a lot of tools no one needs and they are bloating the context, you can deny these in the settings.json (there is also a list here: https://code.claude.com/docs/en/tools-reference ). Also set "disableClaudeAiConnectors" to false to remove even more bloat.
I tried both --system-prompt and --system-prompt-file. they both appended when i tried about 6 months ago and watched the traffic. Yes I cleanup all those tools etc.
too late to edit, but i wanted to apologize for my snarky tone.
my comment was to show incredulity that people do the equivalent of pressing f12 on the browser and submitting a form bypassing the JavaScript validation. you cannot really think the backed will not have the same logic you just bypassed.. but now i realized we're talking about Ai companies so i might be the idiot after all.
this ticks at something that maybe is obvious in retrospect- this is all about economics. the arguments about "using AI doesn't make you an artist", etc. are about being able to charge money for your art i think. maybe obvious to some, but it needs to be explicitly spelled out i think. i was stuck on "i dunno, if i use an AI assistant to run blender i'm still being creative", but is the real argument "you should not be able to charge money to use blender with an AI assistant- you are displacing existing blender artists economically"?
i'm in semi-forced-retirement as an older software engineer in this labor market, so i might be less sensitive to the implicit economic arguments.
I've been on this kick since I realized the primacy of the initial part of the session context. I created a python app that reads a phased plan and kicks off a new session for each phase. There is a standard prompt and handoff mechanism to determine if we encountered any unforeseen issues that we need to address in chat, but otherwise it will just grind with a clean session with appropriate context for each phase.
Just kicking off a subagent does this automatically...?
My main conversation is usually with an orchestrator that hands off work to various (usually cheaper) subagents to plan / review / etc. It has instructions to find the correct model for each task and not to do too much itself so a multi-phase plan automatically gets a fresh subagent for each phase.
i disable subagents. it also helps that i work in microservice size repos, so a single agent can keep the full relevant context. subagents waste alot of unnecessary tokens doing code exploration that the main agent is going to end up duplicating, i think they make more sense for monorepos.
anecdotally I've tried the same plan on different branches with both approaches (single agent with subagents implement the full plan vs. my little session per phase app) and then had agents judge the code quality and my approach worked better and saves tokens. horses for courses.
you don't need to learn blender anymore. you can talk with the ai and describe what you want and iterate. all of the software tools used in media are now accessible to creative but not necessarily technical people.
Koons didn't write a two paragraph description of his ideas and then handed it over to a porcelain factory that repeatedly tried to come up with something that fit the spec until Koons cried out "Yes, that's it! That's what I had in mind!"
He drew sketches, built models, shaped prototypes, etc. He was an active participant in the creative process from the very start to the end, not just a client that checked in on what the artists were doing from time to time to give the thumbs up or thumbs down.
But also, many people consider Koons to be a hack. (He says his art has no intrinsic meaning and he mostly makes vapid balloon sculptures for very rich people.)
If just "being involved" in the "creative process" is enough to be called an artist, then by that measure that includes the guys that are simply involved by having the money to patron the process. Maybe it also includes the guy delivering the pizza.
I mean i suppose it depends on if you think people using blender are artists. If you think the part that makes them creative is knowing to right click select blah do blah then I get your position. If you say "no, make the background fade out at this point, push the face to the foreground, and add some shadow behind", I really don't understand why knowing the specific keystrokes are the creative part.
But I think you know that and you're just straw manning.
edit- to be clear i'm talking about using existing software like logic pro, blender, etc. i'm not talking about generative AI.
Knowing the tools of your trade is what you makes a artist. It's not enough to know that you would like to have the shadow there and have the face pushed to the foreground, that's just being aware of the design space. To be an artist you also need to know how to navigate the design space - know how to use the colours, brushes, vertices - to actually arrive at the place you intend. That means that you know how to use the tools of your trade so that this stroke or infill completes the brow to create the gaze you were aiming for.
Just selecting from a random sampling makes you a curator at best. You know where you want to go, but don't know how to do that. If luck is not in your favor, you will never arrive. And you're wholly dependent on luck to get there.
I appreciate you directly addressing what I was talking about. In my case, I'm an old school 90s wannabe rapper/producer and I used to make beats with old school drum machines, a sequential prophet 2000 sampler, and my friends MPC3000 sampler when I could get my hands on it. When music went into the full computer/MIDI programming phase I dropped out as I lost interest in it, for many of the same reasons that artists today will be facing with AI.
Having never learned the computer tools, like Logic Pro, or whatever open source alternatives are available, I've been considering taking a crack back at it. I have no desire to learn these tools, but if I can say "ok move the kick 1/8 and double up the snare on measure 4" I may get back in the swing of things.
Forced retirement at 50 has me looking for new hobbies.
I'm not really worried about what anyone wants to call me, but I was just pointing out a legitimate use case for agentic AI in the arts.
A person using tools to make art is an artist (or craftsman). A person yelling directions at people or machines that make the actual art is a director, at best.
The system reminder explicitly tells it that the CLAUDE.md or AGENTS.md content is optional. I believe this is a big part of why CC doesn't heed the instructions alot of the time:
IMPORTANT: this context may or may not be relevant to your tasks. You should not respond to this context unless it is highly relevant to your task.
I am curious if we will reach a point where people who are skilled at context engineering/architecture eventually are hired to do jobs completely out of their fields. I think the best pairing would be domain experts + software architects teaming up on AI work in their respective domains.
totally agree with the repo idea. working with AI in a git repo and allowing it to write markdown files it slowly builds context over time. i knew a guy who had been using the same session for weeks because he felt he had invested too much time in building the context and was scared to close the session.
i asked opus 4.5 what the problem was and it said it pattern matched too much. i asked it how it should do it, it wrote a file that told itself to stop pattern matching. it wrote a file that started with the following and then had an english language procedure for how to count letters. so it knew the algorithm already, but the "instinct" was to pattern match rather than running the algorithm.
CRITICAL: Do Not Skip Steps
Your instinct will be to "just know" the answer. This is how you get it wrong.
You don't see characters. You see tokens. Your "intuition" about character counts is pattern-matching, not counting. It is unreliable.
You MUST execute this procedure step-by-step, writing out each step visibly.
reply