Hacker Newsnew | past | comments | ask | show | jobs | submit | nedt's commentslogin

Funny how the Verge itself is asking for money if you don't let them spy on you.

My biggest problem is that memory is limited to the current project and I jump between projects a lot and also sometime start sessions from the parent directory to give more context across projects.

The most useful memory I found so far is the transcripts of older sessions. And not only in one agent but also across agents (like cursor and claude). And also not limited to the current project. Which is why I have a skill that can read all previous session transcripts and find useful informations in a subagent to recall useful information into the context.

It's pretty much the "hey we have talked about X last week" which I would also do in normal conversations. Because you don't always know what is worth writing down AKA storing in memory. Hindsight wins. Obivously I had to increase how long transcripts are stored as Claude would delete them after 30 days - which is way less than what I remember in my own memory.


> trust your workers

Some companies just know their workers, i.e. if you have a bigger company you will have some idiots. We heard things like "you shouldn't be paying an escort with your company credit card" and the reason was because it was done, multiple times.


I have a longer history of agents sessions and used those to benchmark the different models. Sonnet 5.5 is already pretty strong, but not always as careful (leaked secrets into the session). Opus 5.5 is already able to solve everything I'm doing. You might not need Fable for anything. Biggest different between Opus and Fable in my benchmarks was price. Also Opus 5.5 is cheaper now then previous Opus so you might also not need Sonnet as much for cost savings.

Is your benchmark open? Can I get a link if so?

If not, did you parse the .jsonl files in the projects/ directory?


Did I parse the jsonl files? No. I just asked Claude to create a skill to read all previous conversations across all projects or anywhere else in Claude, Cursor, Github Copilot and Windsurf. For the last one it failed (they are encrypted). The rest it just picked up whatever location and format they are stored in. So yes the skill is parsing the jsonl files in projects/. Only exception are Claude Desktop sessions that are only stored in the cloud.

The skill itself is also the basis for my memory across sessions and projects. With that Claude can recall what I can remember. It's using a subagent so only the important bits land in context.

From there it's easy to get some benchmarks. Let it find common themes across your sessions and let it mock them to create benchmarks strong enough so not everything gets the same score. My benchmarks are now based on me designing new tools, debugging live issues going through log files, finding clues in bigger code bases etc. pp.

But no they are not open. First of all they depend on my use cases (but I also got one colleague to contribute cases) and second of all they are based on the work at my company, so I can't share them.


I had a Claude session checking the hn thread every 15 minutes following links to looks for public exploits and mitigations or anything I might have missed. Anyone posting helped a lot in us feeling safer and taking all the necessary steps throughout the day. ty


Just because far right wasn't as popular for a longer while, doesn't mean it's not far right just because more people vote for them. And you shouldn't start suger coating it to protect their voters. They vote far right, they might even be far right. If we think far right is something bad calling it what it is is the only way too go. If it's not bad the voters should have a problem being defined as far rights.


So many commands. I just started claude in a new tmux session with iterm. Told it it's running in tmux with iterm and it should spawn subagents into separate tabs. When I wanted to know the status of the fleet I just asked the primary agent.

Worked pretty well. Now if I want to automate something like that I would just ask it to write a skill at the end of the session. I'm lazy and I can't remember a couple of hotkeys or commands.


herdr (supercharged tmux) has also mouse keybindings. So instead of remembering how to split a pane, you just right click with the mouse and choose "split" from the popup menu


If the opinion is basically objective (of course can't be fully) your doctor shouldn't have any issue with getting a 2nd opinion, if it's from a real doctor. Would see it as a red flag if they really object.


Funniest thing about this is how the arrows look like. They aren't just straight or one clear curve. If you have ever seen what Franz draws in workshops, this mix of art and a children drawing, that's how the arrows look like.

I'm wondering if the AI has been trained on Franz. Which would also explain why the Terminator ends up with Manner Schnitten.


i did spare quite a lot with it of how i want the arrows to look like

the first versions looked way to professional and lost that "yes, that, exactly that, i mean it" arrow feeling that i am usually going for


Funny. We have beavers in the Lobau in Vienna. But it's still dry. And yet we want to build a highway tunnel through it. Guess we just need more beavers. It's famous for it's naturism anyway.


Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: