3371

More importantly, they can stuck at reasoning loop!

Maybe look up "China deflation"

Unfortunately it's much less likely to happen for their managed models because surveillance is built-in in every public-facing service since day-one.

Interestingly the latest report from Anthropic explained that they may be just routing requests to managed models to Anthropic.

This raise a question: why do open source models sometimes identify themself as Anthropic's models. I recall seeing some plausible theories in the past but I can't recall.

codegraph + context mode are all I need.

"He was born, lost his father at 13, was orphaned at 20, had 4 children, and died at 90 of a parasitic infection, with 3 surviving children, 10 grandchildren, and 15 great-grandchildren."

This 3727 BCE Korean boy died at 90.

You may be interested in my No3371/projex repo, which is a whole solution based in this idea, I wonder how much do people do differently in this direction

This makes we wonder if anyone have tried to track attention to code with eye tracker

I don't think it's possible to effectively control LLM's output, unlike Human the difference between "thoughts" and "words" is very weak.

Better just clean stuff up afterward with a dedicated agent.

Yeah, making it a skill is just much more convenient. It can be a very simple skill, a one liner asking the agent to gather the context and write a handoff for another agent. My skill is unnecessarily complex if that's the only requirement.

Yes it's just another (but my own) impl of SDD

It's just good old prompting in skill format. You can check out the skill repo: no3371/projex

> Plus your handoff files becomes a useful piece of project memory that you can reference later.

I've been doing this since I started agentic development, and have a whole framework based on this; Simply put I define workflow s that output templated files for everytype of tasks that happens in development.

It's a powerful pattern I'd recommend everyone.

People believe ublock because it's reputable and not affiliated with browsers IMO.

I used this for a while then decided to build my own suites that pack individual harness and respective host state (config, plugins, skills, etc.) into an image. Works better and much flexible in my opinion.

In my own framework (just a skill), every unit of works produce a markdown document. This provides rich context over past plan/interview/implementation/etc., this + git is more than enough to not think about context and memory.

BBM: Big Brain Model

I did but I removed the English part, not because I assume people can read hiragana, but because I assume people will need to search it up anyway.

IIRC it's a popular TTS character called ずんだもん.

I am taking my limited time before next job hunt and using coding agents to create a wego board strategy game inspired by Escape From Tarkov, Advanced Wars, and PhantomBrigade.

> Why don't they carpet bomb all of the ukraine?

In Chinese we call it "wash the ground with missiles", and many Chinese and Taiwanese say this constantly as a threat or concern. It's amusing that many people think this is a practical thing to happen, and keep bringing it up.

My experience is that Codex's auto review is extremely costly, with $20 on both sides, I can run CC with auto mode for longer than with Codex's auto review enabled. Also in my own experience Claude's usage is actually bigger than Codex, but I am not sure if that's due to I stick to 5.5 with Codex while keep Sonnet as the default to orchestrate other models in CC.

This kind of services that requires the user to share a seed/code to the recipient always seems kinda awkward to me. The code is not simple/short enough to be verbally communicated; If I can send the code, I usually can just send the file.

This does kind of smell like the wrong way to use it. Not trying to self-promote here, but the experiences you shared really made me think I headed the right direction with my prompting framework ("projex" - I once made a post about it).

I straight up skip all the memory thing provided by harnesses or plugins. Most of my thread is just plan, execute, close - Each naturally produce a file - either a plan to execute, a execution log, a post-work walkthrough, and is also useful as memory and future reference.

They would be 100% lying if they have infinite budget allocated to this campaign and haven't approved all requests.

That's... not what was written there. Better read gp again slower.

In my prompting framework I have a workflow that the agent would scan all the artifacts in my closed/ folder and create a yyyymmdd-archive artifact which records all artifact name and their summaries, then just delete them. Since the framework is deeply integrated with git, the artifact can be digged up from git history via the recorded names.

Isn't that for... readability...?

Somehow made me think I should enforce a rule agents should sign their conment so it's identifiable at first glance