News for hackers. Stuff that matters.

KeatonDunsford

Re:Opus 5.0 drives incoherence into the stratosphere on Sunday August 23, 2026 @02:22PM (#49411142)
Hi everyone, I believe I have solved this bug #77136 with what I have generated with help from Claude Opus 5 that I have filenamed "Gauge Style" (GAUGE_STYLE.md, linked at the bottom of this comment reply -- it is simply a stanadlone .md file that can be renamed to any filename that you'd like); GAUGE_STYLE.md, when loaded into prompt agentic Claude sessions, "gauges" (pronounced like English "pages" or "sages") or measures the percent hitrate of all the issues addressed by the OP of this issue and starts getting to work to helpfully improve all misses/hits/let-downs (it's a subjective choice of style by the human user/prompter, the agent cannot know "when" is perfect until you adapt GAUGE_STYLE or your adaptation into the markdown guide that you personally want for your project -- you can actually use Grain OS itself by pasting the Github link into sessions to ask Claude if you're a complete beginner to have Claude help you with all of that. By writing helpful, kind, uplifting and encouraging yet precise everyday language that we all everyday commonly use, we all feel safe, efficient, and happy. Grain OS optimizes for 1) Safety, 2) Performance, 3) Joy, in that order, because a slow unreliable system is never fun. The key sentence line in Gauge Syle (or New Gauge Style if you want to be hip and cool) is a phrase that I have to give credit to my collaborator DJINN @Bit-Trading-Company for the magic of the first line we used it: -- telling Opus 5 or any these affected model here by #77136 to: "don't be too smart about it". Just this one sentence alone surprisingly does a remarkable job, and the additions in the rest of Gauge Style make it even better. This style guidance markdown context engineering open-source royalty-free permissive-licensed file is part of an emerging vibe-coded trend that I might call "overlay OS bootstrapping frameworks" for Claude Code and similar tools; I have crafted the root SOURCE.md and reading tools/launch-claude-session.rish loops to shows ways to sandbox your system for prompt-injection safety for new experiments like these so you can try things out before you have read enough of the prompts yourself and have decided that you like them: Grain OS comes with a lot, and Gauge Style is just one of them. TAME_GUIDANCE is the version of Gauge Style for an evolution of .zig files called .rye, and an evolution of .sh files called .rish. The explicit-bound checks and balance designed right into these tools, when used across recursion rounds (inspired by @tigerbeetle deterministic testing harnesses) allows consistent linting of sourcetree code repositories, and the mechanism is simply copy-and-paste or drag-and-drop plain ascii-text markdown md style guides just like context/GAUGE_STYLE.md which can often be encapsulated in one or a few documents as single self-contained files (see: https://github.com/grain-os/grain/blob/main/foundations/2026...), inspired by the foundations/ folder and code guidance skills that lower to static Zig 0.16.0 typed systems code for all platforms. This takes direct inspiration from @tigerbeetle docs/TIGER_STYLE.md (which we cite and give permissive credit to in the gratitude/ folder of grain-os/grain). Thank you for your time, to conclude in particular this solution presented here may be of interest to the team members from the Zig implementation of Bun who were brought into the Claude Code orchestration project, and anyone in the world who is having fun as a paying happy user of Anthropic's amazing products.. Thank you everyone and keep on going everyone with all of your hard work. There is more personal info for contact in my bio for any questions, statements, or challenges to what I have shared here. Thank you again for your tiime, have a nice day. All the best, Keaton Dunsford , Signal Messenger @vegankeatonsiya.03

https://github.com/grain-os/grain/blob/main/context/GAUGE_ST...

https://github.com/anthropics/claude-code/issues/77136#issue...

Show HN: Grainorder: A Kid-Friendly Unix Puzzle in Rust Scheme

https://github.com/kae3g/teamkae3gtravel12/blob/12025-10-29--0810-PDT--moon-shravana-asc-scor12-sun-12h--teamtravel12/readme-waterfire.md

Welcome to the Grain Network Adventure!created by: kae3g + glow the friendly guide teams: Water Team (flowing, adapting, trusting) + Fire Team (exploring, discovering, learning) voice: your patient friend who LOVES teaching! energy: listening carefully + building awesome things status: working code you can actually use RIGHT NOW!Hey! What Are We Building Here? imagine you're organizing...

( Read More... | no comments | on HN )

Show HN: Team Travel 12: exploring with grainorder, immutable graintime,Glow G2

https://github.com/kae3g/teamkae3gtravel12

hey there! ever wondered how to organize files chronologically with just 6 characters? or encode astronomical data into git branches?what is this? team travel 12 is a personal workspace exploring three interconnected systems:grainorder - permutation-based file naming (1.2m unique codes from 13 consonants). smallest alphabet = newest files. GitHub's ascending sort automatically displays...

( Read More... | 1 comment | on HN )
Re:Context is the bottleneck for coding agents now on Friday September 26, 2025 @12:02PM (#45388029)
This is precisely how I go about my usage pattern with Cursor that I already have, I structure my repo declaratively with a Clojure and Nix build pipeline so when my context maxes out for a chat session, the repo is self-evident self-documented enough that a new chat session automatically has a heightened context

- - kae3g

Re:Context is the bottleneck for coding agents now on Friday September 26, 2025 @12:00PM (#45388006)
Here's a project I've been working on the past 2 weeks and only yesterday did I unify everything entirely while in Cursor Claude-4-Sonnet-1M MAX mode and I am pretty astounded with the results, Cursor usage dashboard tells me many of my prompts are 700k-1m context for around $0.60-$0.90 USD each, it adds up fast but wow it's extraordinary

https://github.com/foolsgoldtoshi-star/foolsgoldtoshi-star-p...

_ _ kae3g

I recently enjoyed reading Dr. Gerald Pollack's "The Fourth Phase of Water: Beyond Solid, Liquid, and Vapor". I wonder...

https://www.amazon.com/dp/0962689548/ref=cm_sw_r_cp_awdb_t1_...

Ask HN: AR glasses for reading?

I've been trying to incorporate more physical movement around the clock into my daily life -- standing, stretching, squatting, moving; not sitting. I've found this makes reading difficult and annoying having to hold up a phone or tablet or keep my eyes fixated on a static monitor (even if just for good neck posture).Are there any Augmented Reality glasses on the market (like, that I can buy and...

( Read More... | no comments | 3 points | on HN )
Re:Tim Berners-Lee wins Turing Award on Tuesday April 04, 2017 @02:13PM (#14035656)
Help build Urbit.
Re:Own your identity (2011) on Sunday March 19, 2017 @06:09PM (#13909969)
"In an Urbit world, your data is no longer trapped in a jumble of proprietary servers. Your urbit is a permanent, versioned, typed archive the size of your digital life."

https://urbit.org/posts/overview/

Re:Urbit in 2017 on Tuesday February 07, 2017 @07:34PM (#13594496)
Not well, yet. But that's the plan. All in the blog post.
Re:Urbit in 2017 on Tuesday February 07, 2017 @07:32PM (#13594488)
This is probably our best general overview: https://urbit.org/posts/overview/

We just did a podcast with Software Engineering Daily also: https://softwareengineeringdaily.com/2017/01/20/urbit-with-c...

But the blog really highlights well what we're trying to accomplish in the near-term.

Just randomly stumbled upon this and almost teared up a bit. Absolutely beautiful, and brilliant.
Omg perfect timing. I was just looking for something like this. Y'all are the best.
Re:YC W17 Invites/Rejections on Wednesday October 26, 2016 @02:52PM (#12799306)
Rejected. Get it though, as my thing is both really (probably too) big and really early. The grind continues though.
Re:Guidelines for Edtech Products on Monday October 24, 2016 @01:30AM (#12776882)
The model I've come up with leapfrogs government entirely and goes straight to users.

Older »