Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Changes in the system prompt between Claude Opus 4.6 and 4.7

simonwillison.net

Post reply on HN
Print view View original

181–190 of 240 posts

Re: Changes in the system prompt between Claude Opus 4.6 and 4.7

#181

Post by jrvarela56 »

dockerd wrote: post #179
jrvarela56 wrote: post #169

The past month made me realize I needed to make my codebase usable by other agents. I was mainly using Claude Code. I audited the codebase and identified the points where I was coupling to it and made a refactor so that I can use either codex, gemini or claude. Here are a few changes: 1. AGENTS.md by default across the codebase, a script makes sure CLAUDE.md symlink present wherever there's an AGENTS.md file 2. Skill…

How do you share the context/progress of goal across agents?

I implemented a client for each so that the session history is easy to extract regarding the agent (somewhat related to progress of goal).

Context: AGENTS.md is standard across all; and subdirectories have their AGENTS.md so in a way this is a tree of instructions. Skills are also standard so it's a bunch of indexable .md files that all agents can use.

Re: Changes in the system prompt between Claude Opus 4.6 and 4.7

#182

Post by Lucasoato »

jrvarela56 wrote: post #169

The past month made me realize I needed to make my codebase usable by other agents. I was mainly using Claude Code. I audited the codebase and identified the points where I was coupling to it and made a refactor so that I can use either codex, gemini or claude. Here are a few changes: 1. AGENTS.md by default across the codebase, a script makes sure CLAUDE.md symlink present wherever there's an AGENTS.md file 2. Skill…

Have you got any advice in making agents from different providers work together?

In Claude, I’ve seen cases in which spawning subagents from Gemini and Codex would raise strange permission errors (even if they don’t happen with other cli commands!), making claude silently continue impersonating the other agent. Only by thoroughly checking I was able to understand that actually the agent I wanted failed.

Re: Changes in the system prompt between Claude Opus 4.6 and 4.7

#183

Post by jrvarela56 »

Lucasoato wrote: post #182
jrvarela56 wrote: post #169

The past month made me realize I needed to make my codebase usable by other agents. I was mainly using Claude Code. I audited the codebase and identified the points where I was coupling to it and made a refactor so that I can use either codex, gemini or claude. Here are a few changes: 1. AGENTS.md by default across the codebase, a script makes sure CLAUDE.md symlink present wherever there's an AGENTS.md file 2. Skill…

Have you got any advice in making agents from different providers work together? In Claude, I’ve seen cases in which spawning subagents from Gemini and Codex would raise strange permission errors (even if they don’t happen with other cli commands!), making claude silently continue impersonating the other agent. Only by thoroughly checking I was able to understand that actually the agent I wanted failed.

Not sure if you mean 1) sub-agent definitions (similar to skills in Claude Code) or 2) CLI scripts that use other coding agents (eg claude calling gemini via cli).

For (1) I'm trying to come up with a simple enough definition that can be 'llm compiled' into each format. Permissions format requires something like this two and putting these together some more debugging.

(2) the only one I've played with is `claude -p` and it seems to work for fairly complex stuff, but I run it with `--dangerously-skip-permissions`

Re: Changes in the system prompt between Claude Opus 4.6 and 4.7

#184

Post by zdragnar »

otabdeveloper4 wrote: post #153
zozbot234 wrote: post #79

Earlier quoted context omitted.

That part of the system prompt is just stating that telling someone who has an actual eating disorder to start counting calories or micro-manage their eating in other ways (a suggestion that the model might well give to an average person for the sake of clear argument, which would then be understood sensibly and taken with a grain of salt) is likely to make them worse off, not better off. This seems like a common-sen…

> This seems like a common-sense addition. Mm, yes. Let's add mitigation for every possible psychological disorder under the sun to my Python coding context. Very common-sense.

It's what you get when you create sycophant-as-a-service. It will, by design, feed all of your worst fears and desires.

LLMs aren't AGI, and I'd go further and say they aren't AI, but admitting it is snake oil doesn't sell subscriptions.

Re: Changes in the system prompt between Claude Opus 4.6 and 4.7

#185

Post by sigmoid10 »

aesthesia wrote: post #132
sigmoid10 wrote: post #14

Earlier quoted context omitted.

Yeah that was the original idea of system prompts. Change global behaviour without retraining and with higher authority than users. But this has slowly turned into a complete mess, at least for Anthropic. I'd love to see OpenAI's and Google's system prompts for comparison though. Would be interesting to know if they are just more compute rich or more efficient.

Leaked/extracted system prompts for other chat models, particularly ChatGPT, are often around this size. Here's GPT-5.4: https://github.com/asgeirtj/system_prompts_leaks/blob/main/O...

Thanks, but that kind of confirms my belief. wc counts ~15k words in there. That may technically be the same order of magnitude, but it is only a quarter of Claude's and less than 2% of the context limit. So a lot more steering is baked into the model weights than into the prompt compared to Claude.

Re: Changes in the system prompt between Claude Opus 4.6 and 4.7

#186

Post by ashdksnndck »

eastbound wrote: post #71
naasking wrote: post #30

Earlier quoted context omitted.

Seriously, when you're conversing with a person would you prefer they start rambling on their own interpretation or would you prefer they ask you to clarify? The latter seems pretty natural and obvious. Edit: That said, it's entirely possible that large and sophisticated LLMs can invent some pretty bizarre but technically possible interpretations, so maybe this is to curb that tendency.

—So what would theoretically happen if we flipped that big red switch? —Claude Code: FLIPS THE SWITCH, does not answer the question. Claude does that in React, constantly starting a wrong refactor. I’ve been using Claude for 4 weeks only, but for the last 10 days I’m getting anger issues at the new nerfing.

I sometimes prompt with leading questions where I actually want Claude to understand what I’m implying and go ahead and do it. That’s just part of my communication style. I suppose I’m the part of the distribution that ruins things for you.

Re: Changes in the system prompt between Claude Opus 4.6 and 4.7

#187

Post by rob74 »

majormajor wrote: post #147
embedding-shape wrote: post #6

> The new section includes: When a request leaves minor details unspecified, the person typically wants Claude to make a reasonable attempt now, not to be interviewed first. Uff, I've tried stuff like these in my prompts, and the results are never good, I much prefer the agent to prompt me upfront to resolve that before it "attempts" whatever it wants, kind of surprised to see that they added that

I wonder if they're optimizing for metrics that look superficially-worse if the system asks questions about ambiguity early. I've had times where those questions tell me "ah, shit, this isn't the right path at all" and that abandoned session probably shows up in their usage stats. What would be much harder to get from the usage stats are "would I have been happier if I had to review a much bigger blob of output to re…

Or they're optimizing for increased revenue? If Claude goes down a completely wrong path because it just assumes it knows what you want rather than asking you, and you have to undo everything and start again, that obviously uses much more tokens than if you would have been able to clarify the misunderstanding early on.

Re: Changes in the system prompt between Claude Opus 4.6 and 4.7

#188

Post by sigmoid10 »

pests wrote: post #136
sigmoid10 wrote: post #21

Earlier quoted context omitted.

You can cache K and V matrices, but for such huge matrices you'll still pay a ton of compute to calculate attention in the end even if the user just adds a five word question.

The state of the system can be cached after the system prompt is calculated and all new chats start from that state. O(n^2) is not great but apparently its fine at these context lengths and I'm sure this is a factor in their minimum prompt cost. Advances like grouped query or multi head attention or sparse attention will eventually get rid of that exponential, hopefully.

That's not how it works. The system prompt doesn't "get calculated first" or anything. You combine it with the user prompt and then run the generation for the first new token on that thing, which basically boils down to one huge matmul that runs in parallel. So you can literally just cache a part of the input matrices for the first step and then you'll very quickly run into n^2 complexity.
Post reply on HN