Memory in a coding agent harness is the dumbest thing ever, change my mind. This was injected on _EVERY TOOL CALL_. Yeah super important reminders every time I use bash, this.
Everyone should set CLAUDE_CODE_DISABLE_AUTO_MEMORY to 1.
it's bad. it has the typical symptoms of someone who lacks taste and innovation
i constantly see new competitors in my space as well that either blatantly copy or slop their way through. on the flip side it actually motivates me to make my product better to keep the edge. what they're missing though is the audience and knowledge how to market it.
claude proofs a change is working with a recording of using the app like a user would. cursor and captions included.
we're getting close to skipping human code reviews. what a time to be alive!
🚀 Repeek v5.6.17 is out!
- Team stats are back with average team Elo, and also show total matches, wins, average FACEIT rating, swing, K/D, K/R and ADR
We hope you like this little addition! If not, you can disable team stats in Repeek settings under matchroom.
how do you guys create tasteful marketing pages and ui with claude code? i've seen a lot of new tools with beautiful websites in recent weeks. you can still tell they're ai-generated, but they also look different and good?
@benjaminakar@trymeru to use Gmail as desktop app with multiple account support, unified inbox, verification code copy, phishing protection, ad/tracking blocker and more :)
what are your mac mini specs? just got the new mac mini m5, 32gb ram, 1tb disk and i'm afraid that it's actually not enough. already at 50% ram usage after setting up the mac, running one agent and a vm. thinking about returning it and getting m5 pro, 48gb ram and 1tb disk instead.
as much as i love @SoundCloud and prefer over other music streaming services, their ux is probably the one i enjoy the least.
what are they actually doing all day? basic features like switching playback between web and phone still doesn’t exist and i’m not able to see the progress of tracks downloading or even cancel on iOS. pls @SoundCloud invest more into ux and qol improvements.
This improved my review workflow: Ask your agent to create an artifact based on /show-me by @dexhorthy to visualize the changes with HTML instead of being limited by markdown. It reads more like a presentation and involves less mental juggling. Might create a skill for this.
tenet using Jev is up to 275x cheaper and 104x faster than Claude agents 🤯
Same 2,000-line diff, same rules. tenet: 1.5 s, $0.0036. One Claude Fable 5.1 call: 157 s, $0.99. Your agent fixes its own review findings before you ever see the diff.
Spent a day building tenet, a review gate for code that agents write. It judges each commit, commit message and PR against rules written in plain language, using Jev by @typesafeai.
The rules come from @poteto's unslop skill, @hvpandya's stop-slop, @dillon_mulroy's anti-slop and @mattpocockuk's new /pr skill. It also generates rules from your AGENTS.md or CLAUDE.md, and you can write your own rules and presets.
Here an agent trips four of them, gets blocked in 1.07 s for $0.0003, and fixes its own findings.
Spent a day building tenet, a review gate for code that agents write. It judges each commit, commit message and PR against rules written in plain language, using Jev by @typesafeai.
The rules come from @poteto's unslop skill, @hvpandya's stop-slop, @dillon_mulroy's anti-slop and @mattpocockuk's new /pr skill. It also generates rules from your AGENTS.md or CLAUDE.md, and you can write your own rules and presets.
Here an agent trips four of them, gets blocked in 1.07 s for $0.0003, and fixes its own findings.