Zook

Data analyst | Blockworks

[Zook @zook_data]

@asmirkn @dani_avila7 Yeah that's what I'm worried about.
@grok read the OP linked repo. Does the plugin have a reliable way to prevent this trap? Does it prompt the user before adding stuff to Claude.MD, or does it just make updates pro actively ?

[Zook @zook_data]

@JamesChristoph Huge EWZ crash this morning. Largest drop in 26 years.

[Zook @zook_data]

@0xSero Tomorrow can you add all historical evidence that he does not exist ? Seems fair. 😀

[Zook @zook_data]

@Hikari_07_jp You could probably write a ground breaking research paper on this topic.

[Zook @zook_data]

@_dshap This past weekend I used Luna to complete my Black October security sweep.
It did a great job patching vulnerabilities and upgrading absolutely every program on my systems. It worked for an hour and turned the procedure into a Skill file. I like Luna. :-)

[Zook @zook_data]

@_dshap OpenAI on the speaker list is quite impressive. So is the side event at the US embassy! 👀

[Zook @zook_data]

@blockworksDAS The side events look amazing !

[Zook @zook_data]

@poteto @yminsky Thank you for sharing. Will listen during my lunch time.

[Zook @zook_data]

@grok I keep hearing these breaking news about the FP16 TOPS unlock on the 3090 GPU.
Real talk: does it allow these cards to churn more tok/sec ? If yes, does it require proportionally more power ?
Search far and wide for real user testimonials and before / after numbers on LLM inference.

[Zook @zook_data]

@SilvioBusonero Hey Silvio,
With Muse you can opt out of using your chats to train the models. I did it.

[Zook @zook_data]

@Hikari_07_jp Hey Hikari,
If it does not work, I recommend installing Meta Muse on his smartphone. I think its available in Japan now. Get it connected to his email and his phone.

[Zook @zook_data]

@Aquilae_80085 @JamesChristoph Right!?

[Zook @zook_data]

@netflix @grok is there a new season of Dark, or is Netflix advertising their back catalogue out of desperation ?

[Zook @zook_data]

My takeaways:
1) The Orchestrator-subagent swarm is overrated!
2) A curated set of custom skills, tools and Memory .MD outperforms both Vanilla Claude and Swarms.

A good Subagent I'm still using :
A tool call that invokes a Sonnet subagent to read Anthropic documentation. It spares my main agents' context window when I ask it "Can Claude do X?"

A good skill file I'm maintaining:
A /dashboard-preferences skill that tells my agent how to make my data dashboards look clean.

When building your own skill files and memory files, you can prevent context bloat by storing them in a project folder. You can also add disable-model-invocation if you don't want its description to always appear in your agents context window.

One problem I'm still facing is maintaining up to date Readme files and other project documentation as new decisions get made.

https://t.co/WkMNkXxiLq

[Tak 🦞 @cherry_mx_reds]

Anthropic is now recommending that companies REMOVE unnecessary subagents from their AI systems.

This is a pretty significant shift in agent architecture, with their team finding that a single agent equipped with skills can frequently outperform complex multi-agent systems, and one customer seeing roughly 2x faster execution.

The emerging rules of thumb:

→ Skills: Teach the agent HOW to do something, including procedures, expertise, and instructions.

→ Tools / MCP: Give the agent the ability to DO things through APIs, actions, and live data.

→ Subagents: Use for parallel work, isolated context, or tasks a single agent struggles with.

The interesting part is progressive disclosure. Instead of spawning specialized agents or stuffing everything into a giant prompt, one agent discovers and loads skills and tools as needed.

Their advice is to fold existing subagents into the main loop, turn their instructions into skills, expose their actions as tools, and run evals to see if performance holds up.

As models have gotten smarter, improving agent architecture is becoming less about what you can add and more about what you can delete.

https://t.co/DNq60Skkmc

[Zook @zook_data]

@bradbeckett @michaelpatron0 @grok is the strategy here to save a passkey to Bitwarden, and then when its time to login , the user will pull their Passkey from Bitwarden ? If yes, what is the advantage of login via Passkey in Bitwarden versus Normal Password in Bitwarden? Is the login flow faster?

[Zook @zook_data]

@shydev69 @poteto Good idea to feed the transcript.

[Zook @zook_data]

@Hikari_07_jp I wonder where they will be distributed now.

[Zook @zook_data]

@kevinrwhitley @Blockworks Can't wait to use it. 😀

[Zook @zook_data]

@Hikari_07_jp Brilliant workaround.