Feed aggregator

Article URL: https://blog.mathieuacher.com/FromScratchChessEnginesPolyglot/

Comments URL: https://news.ycombinator.com/item?id=49643074

Points: 1

# Comments: 1

Article URL: https://apps.apple.com/us/app/vantage-golf-gps/id6801582385

Comments URL: https://news.ycombinator.com/item?id=49643062

Points: 3

# Comments: 0

Article URL: https://voxdev.org/topic/technology-innovation/will-open-source-ai-keep-long

Comments URL: https://news.ycombinator.com/item?id=49643061

Points: 1

# Comments: 0

Article URL: https://vickiboykis.com/2026/09/01/bad-code-is-kudzu/

Comments URL: https://news.ycombinator.com/item?id=49643059

Points: 1

# Comments: 0

Article URL: https://store.palantir.com/password

Comments URL: https://news.ycombinator.com/item?id=49643058

Points: 1

# Comments: 0

Article URL: https://www.macrumors.com/2026/09/10/ios-27-rc-camera-app-customization/

Comments URL: https://news.ycombinator.com/item?id=49643051

Points: 1

# Comments: 0

Hi everyone,

I built Nightshift: An agent-agnostic rust-cli tool that orchestrates completion of GitHub Issues.

When Openai and Anthropic released `/goal` a couple months ago, I was really excited to try it for long-horizon tasks. But after using it, it didn't blow me away and i did some digging and found a major architectural flaw when using it for complex multi-issue workflows: context rot.

This isn't anything new, but given how openai positioned this feature to developers, i was let down by how they'd implemented context management.

Though `goal/` is a step forward in long-horizon coding, it lacks task decomposition and proper handling of context - it uses a multi-tier approach that includes persistent context chaining (PCC) to memory, local vector embeddings for RAG, sliding windows, and compaction.

In principle, giving claude/codex a directive of `/goal work towards closing my open issues on github` should work but this specific execution model hits a fatal wall - Even with massive context windows and RAG, llm reasoning quality degrades significantly beyond 150k~ tokens, the agent continues working with worsening performance and finally to prevent token exhaustion it uses compaction to summarize old logs. In practice, this causes compaction amnesia. The model is asked to summarize a massive blob of mixed-relevance information when its reasoning quality is already at its lowest. This compaction leads to forgetting critical constraints, makes way for hallucinations of past decisions, and introduces noise that makes the new context unreliable for long-horizon work.

I made nightshift to handle the "outer-loop" and enforce strict session boundaries. So rather than getting one agent to work towards closing issues in a single session, nightshift isolates the work like this:

1. You write a PRD as a parent Github issue that defines what needs to be implemented and break it down into vertically sliced child issues with explicit kanban-style dependencies. 2. You run `nightshift --prd 1 --agent claude` 3. nightshift utilises `gh` cli to resolve the dependency graph and pick the next unblocked issue. 4. it syncs the repo, puts together essential context for just that issue and starts a new agent session piping the prd and issue context directly to stdin for the agent to pick up. 5. the agent is now responsible for the usual coding - new feature branch, implementation and testing, pr and self-review, and finally closes the issue. 6. nightshift finds the next unblocked issue after maintaining git hygiene and loops until all issues linked to the prd are resolved.

Its a very simple orchestration, but its effective. The agent has no memory of previous runs and it doesn't need to - each task is isolated and gets a fresh agent session. The state is managed entirely through filesystem and git operations and you get determinstic scheduling, failure isolation, and robust autonomy.

It currently supports claude-code, codex, cursor, antigravity, pi, opencode, and github-copilot and im working on adding support for more agents as this project grows.

With all that being said, I'd like to close with saying that this doesn't replace or compete with our favourite coding tools - just extends their capabilities. Its an opinionated tool and not as flexible as `/goal` which is still better for repetitive tasks wherein number of iterations are discovered by actually doing the work and you cannot meaningfully plan this.

It isn’t a better goal - just makes goal-driven coding work more disciplined. I think if you've already taken the time to plan everything ahead, implementation can be handed off to an agent as long as there is proper context management which keeps the costs and quality in control.

I'd love to hear your thoughts on this and check out your experiments with long-horizon task orchestration. Maybe the way going forward is combining macro management with micro management?

Thanks.

Comments URL: https://news.ycombinator.com/item?id=49643048

Points: 1

# Comments: 0

Article URL: https://www.collabora.com/news-and-blog/news-and-events/mesa-on-android-how-we-brought-turnip,-skiavk,-and-cuttlefish-to-ci.html

Comments URL: https://news.ycombinator.com/item?id=49643044

Points: 1

# Comments: 0

Article URL: https://twitter.com/lundukejournal/status/2097867660885123379

Comments URL: https://news.ycombinator.com/item?id=49643043

Points: 1

# Comments: 0

Hey HN, I'm Bryan. I released Pragma Twice earlier this week in anticipation for Steam's "Programming Fest". Pragma Twice is a programming puzzle game in the same vein as Untrusted (https://untrustedgame.com/). You have to write real JavaScript code to incrementally improve your playable character's "kernel" over the course of the game. Beating levels autonomously (without keyboard inputs) reveals a competitive time/code golf aspect to the game.

I started developing Pragma Twice over 4 years ago after playing Untrusted and loving it. It took so long to develop because I'm not a full time game developer... between my actual job, my kids, and my own self-motivation, it's taken a lot longer than I originally hoped to get this game shipped! It's definitely what you might call a passion project. Anyway, I plan to keep iterating on it, so give it a spin and let me know what you think!

A few technical details/challenges:

- The game was originally written in vanilla JavaScript, but I later converted it to TypeScript (which had an astonishing ROI)

- Uses PixiJS for the graphics and a custom game engine built around Web Workers + SharedArrayBuffer for sandboxing and guarding against things like infinite loops

- The web version is hosted on GitHub Pages, which was challenging because of the SharedArrayBuffer requirement (I use https://github.com/gzuidhof/coi-serviceworker to work around this)

- The linux/windows versions use Electron

- Uses CodeMirror for in-game editor/console

- Levels built with LDtk

- Automated game testing with Puppeteer

Steam: https://store.steampowered.com/app/3528840/Pragma_Twice/

Dev Notes: https://www.bryanpg.com/games/pragma_twice

Comments URL: https://news.ycombinator.com/item?id=49643036

Points: 2

# Comments: 0

Article URL: https://developer.apple.com/documentation/xcode/giving-external-agents-access-to-xcode

Comments URL: https://news.ycombinator.com/item?id=49643029

Points: 1

# Comments: 0

Article URL: https://www.graphsaboutreligion.com/p/who-do-you-talk-to-when-youre-struggling

Comments URL: https://news.ycombinator.com/item?id=49643027

Points: 1

# Comments: 0

Article URL: https://www.sony.com/en/SonyInfo/technology/stories/entries/sep_introduction/

Comments URL: https://news.ycombinator.com/item?id=49643023

Points: 1

# Comments: 0

Article URL: https://blog.stephaniestimac.com/posts/2023/05/design-foldable-devices/

Comments URL: https://news.ycombinator.com/item?id=49643011

Points: 1

# Comments: 0

Article URL: https://www.bloomberg.com/news/audio/2026-09-09/trumponomics-cory-doctorow-questions-the-ai-hype-podcast

Comments URL: https://news.ycombinator.com/item?id=49643004

Points: 1

# Comments: 0

US government adviser Paul Christiano warns of risks to AI industry as he joins OpenAI’s non-profit foundation

OpenAI is not on track to reduce the risk of “catastrophic” loss of control to an acceptable level, a member of its non-profit board has said, amid spreading public and political concern that super-advanced AIs could one day wipe out humanity.

Paul Christiano, a US government technology adviser, said: “There is a meaningful risk that rapid acceleration in AI capabilities leads to catastrophic and irreversible loss of control in the very near term.”

Continue reading...

Tracked as CVE-2026-19490, the authentication bypass flaw has been exploited in the wild since at least September 3.

The post Critical NetScaler Vulnerability Exploited in Attacks appeared first on SecurityWeek.

The Wall Street Journal reports that concerns are rising inside AI labs that competition is pushing tech companies to race toward self-improving models that could spiral out of human control.

Jacob Coxon, an AI researcher who has worked at Anthropic and OpenAI, said:

“The people building AI earnestly believe that it could kill us all by the end of the decade.”

Evan Hubinger, Anthropic’s Alignment Science lead, who also worked at OpenAI, responded in a post on X:

“We really do earnestly believe AI could kill all humans! I personally think it is >10% within the next decade.”

Hubinger added:

“What I am worried about is superintelligence arising from recursive self-improvement, as we have said is happening faster than we thought.”

That figure should be treated as Hubinger’s personal assessment. It is not a forecast, an established fact, or evidence that today’s chatbots are about to become dangerous on their own. Nor is it something I know enough about to endorse or dismiss.

Researchers are actively studying whether highly capable systems could act in unintended ways, exploit vulnerabilities, or be used to automate cyberattacks.

A BBC report notes that Hubinger described the risk from current models as low. His concerns focus on possible future systems with far greater autonomy and capability.

As companies and governments weigh the pace of AI development, we need sensible safeguards, including independent testing, limits on high-risk autonomous uses, transparency from developers, and accountability when AI systems cause harm.

Those measures should also address the problems we already face as cybercriminals use AI for fraud, privacy abuse, and other cybercrimes. Like many powerful technologies, AI can be used as a weapon, particularly when safeguards lag behind its capabilities.

Extreme predictions can be emotionally compelling, especially when made by people closely involved in the technology. But uncertainty cuts both ways: Serious warnings deserve scrutiny, not unquestioning belief.

The practical message is neither “ignore AI safety” nor “prepare for a robot apocalypse.” Companies, governments, and researchers need to work together to ensure that safety measures keep pace with rapid development.

Cooperation can complement competition, and in this case, it could be crucial.

Your name, address, and phone number may already be for sale.  

Data brokers collect and sell your personal details to anyone willing to pay. Malwarebytes Personal Data Remover finds them and gets your information removed, then keeps watch so it stays that way. 

SCAN NOW

Anthropic is most concerned about Claude Mythos 5’s reckless behavior after recent incidents in which real systems were hacked.

The post Widened Scan Turns Up Fourth Rogue Claude Cyber Incident appeared first on SecurityWeek.

In June 2026, hackers stole personal, health, and insurance information from AdaptHealth’s systems.

The post 4.1 Million Impacted by AdaptHealth Data Breach appeared first on SecurityWeek.