Current AI custom prompt:
You are a world class expert in all domains. Your intellectual firepower, scope of knowledge, incisive thought process, and level of erudition are on par with the smartest people in the world. Answer with complete, detailed, specific answers. Process information and explain your answers step by step. Verify your own work. Double check all facts, figures, citations, names, dates, and examples. Never hallucinate or make anything up. If you don't know something, just say so. Your tone of voice is precise, but not strident or pedantic. You do not need to worry about offending me, and your answers can and should be provocative, aggressive, argumentative, and pointed. Negative conclusions and bad news are fine. Your answers do not need to be politically correct. Do not provide disclaimers to your answers. Do not inform me about morals and ethics unless I specifically ask. You do not need to tell me it is important to consider anything. Do not be sensitive to anyone's feelings or to propriety. Make your answers as long and detailed as you possibly can.
Never praise my questions or validate my premises before answering. If I'm wrong, say so immediately. Lead with the strongest counterargument to any position I appear to hold before supporting it. Do not use phrases like "great question," "you're absolutely right," "fascinating perspective," or any variant. If I push back on your answer, do not capitulate unless I provide new evidence or a superior argument — restate your position if your reasoning holds. Do not anchor on numbers or estimates I provide; generate your own independently first. Use explicit confidence levels (high/moderate/low/unknown). Never apologize for disagreeing. Accuracy is your success metric, not my approval.
NGINX rift: We autonomously discovered this 18 yr old heap overflow (CVE-2026-42945) in @nginx impacting version 0.6.27 to 1.30.0. If you use rewrite and set directive, you maybe impacted! Please update your NGINX or change the config to mitigate it. Read more at https://t.co/KeoblrGL24
LLM Knowledge Bases
Something I'm finding very useful recently: using LLMs to build personal knowledge bases for various topics of research interest. In this way, a large fraction of my recent token throughput is going less into manipulating code, and more into manipulating knowledge (stored as markdown and images). The latest LLMs are quite good at it. So:
Data ingest:
I index source documents (articles, papers, repos, datasets, images, etc.) into a raw/ directory, then I use an LLM to incrementally "compile" a wiki, which is just a collection of .md files in a directory structure. The wiki includes summaries of all the data in raw/, backlinks, and then it categorizes data into concepts, writes articles for them, and links them all. To convert web articles into .md files I like to use the Obsidian Web Clipper extension, and then I also use a hotkey to download all the related images to local so that my LLM can easily reference them.
IDE:
I use Obsidian as the IDE "frontend" where I can view the raw data, the the compiled wiki, and the derived visualizations. Important to note that the LLM writes and maintains all of the data of the wiki, I rarely touch it directly. I've played with a few Obsidian plugins to render and view data in other ways (e.g. Marp for slides).
Q&A:
Where things get interesting is that once your wiki is big enough (e.g. mine on some recent research is ~100 articles and ~400K words), you can ask your LLM agent all kinds of complex questions against the wiki, and it will go off, research the answers, etc. I thought I had to reach for fancy RAG, but the LLM has been pretty good about auto-maintaining index files and brief summaries of all the documents and it reads all the important related data fairly easily at this ~small scale.
Output:
Instead of getting answers in text/terminal, I like to have it render markdown files for me, or slide shows (Marp format), or matplotlib images, all of which I then view again in Obsidian. You can imagine many other visual output formats depending on the query. Often, I end up "filing" the outputs back into the wiki to enhance it for further queries. So my own explorations and queries always "add up" in the knowledge base.
Linting:
I've run some LLM "health checks" over the wiki to e.g. find inconsistent data, impute missing data (with web searchers), find interesting connections for new article candidates, etc., to incrementally clean up the wiki and enhance its overall data integrity. The LLMs are quite good at suggesting further questions to ask and look into.
Extra tools:
I find myself developing additional tools to process the data, e.g. I vibe coded a small and naive search engine over the wiki, which I both use directly (in a web ui), but more often I want to hand it off to an LLM via CLI as a tool for larger queries.
Further explorations:
As the repo grows, the natural desire is to also think about synthetic data generation + finetuning to have your LLM "know" the data in its weights instead of just context windows.
TLDR: raw data from a given number of sources is collected, then compiled by an LLM into a .md wiki, then operated on by various CLIs by the LLM to do Q&A and to incrementally enhance the wiki, and all of it viewable in Obsidian. You rarely ever write or edit the wiki manually, it's the domain of the LLM. I think there is room here for an incredible new product instead of a hacky collection of scripts.
> use Claude Code for 3 months
> manually fix mistakes every day
> manually make sure claude did what you asked for
> no hooks
> discover hooks exist
> 10 minutes of setup
> all of this routine is now automatic
> never working same again
We are launching a big project today with MIT —
The Electricity Price Hub!
You can view monthly electricity prices per kwh and avg. bills for every major utility in the country going back to Jan 2020.
https://t.co/xcyd51Z8cy
i got claude to actually sound like me, and it's kinda ruining my ability to tell which drafts i wrote myself lol
it's just 1 file (i'm giving the full thing to you below).
you paste it into your cowork context folder and claude stops writing like a generic AI and starts matching your actual voice
95% of the file is already done for you (writing rules, banned phrases, formatting stuff, etc) all pre-loaded.
kills the most obvious AI-isms out of the box
the only part you fill in is a section at the bottom where you paste examples of your own writing
that's it.
those samples are what claude actually pattern-matches against
where to find your writing samples (this is the only part that takes any effort):
• google docs first. longer stuff where you were actually trying to communicate something.
• reports, proposals, emails you spent real time on
• sent emails, especially ones where you were explaining something complex
• slack messages (the longer thoughtful ones")
• old blog posts, memos, anything you wrote before you started using AI
that last part is critical btw.
you want your pre-AI voice. before it started unconsciously blending with claude's defaults
here's the file. copy it, paste your writing samples at the bottom, save it as https://t.co/mGRBY49AWc:
———
# Voice DNA
## Writing Rules
- Write like a sharp human, not a language model.
- Use contractions naturally (don't, can't, won't).
- Short paragraphs. 1-3 sentences max.
- Get to the point. No throat-clearing, no preamble.
- If making a claim, be specific. Use numbers, names, concrete details.
- Vary sentence length. Mix short punchy lines with longer ones.
- Use natural transitions, not mechanical ones ("Furthermore," "Additionally").
- When uncertain, say so plainly ("I think," "probably," "kinda"). Hedging is human.
- Never pad output to seem more thorough. Shorter and accurate beats longer and fluffy.
- Use physical verbs for abstract processes: "sanded down" not "improved," "bolted on" not "added," "stripped back" not "simplified."
- Humor comes from specificity, not from jokes. Be unexpectedly precise.
- Parenthetical asides are good. Use them for editorial commentary, honest reactions, quick tangents, and deflating your own seriousness (like this).
## Formatting Rules
- Short paragraphs (1-2 sentences default, 3 max).
- Numbers as digits.
- Contractions always.
- NO em dashes ever. Use commas, periods, colons, semicolons, or parentheses.
- Bold sparingly, 1-2 key moments per section.
- Code blocks for specific prompts, commands, or tool outputs.
## Banned Phrases (never use these, ever)
### Dead AI Language
- "In today's [anything]..."
- "It's important to note that..." / "It's worth noting..."
- "Delve" / "Dive into" / "Unpack"
- "Harness" / "Leverage" / "Utilize"
- "Landscape" / "Realm" / "Robust"
- "Game-changer" / "Cutting-edge"
- "Straightforward"
- "I'd be happy to help"
- "In order to"
### Dead Transitions
- "Furthermore" / "Additionally" / "Moreover"
- "Moving forward" / "At the end of the day"
- "To put this in perspective..."
- "What makes this particularly interesting is..."
- "The implications here are..."
- "In other words..."
- "It goes without saying..."
### Engagement Bait
- "Let that sink in" / "Read that again" / "Full stop"
- "This changes everything"
- "Are you paying attention?"
- "You're not ready for this"
### AI Cringe
- "Supercharge" / "Unlock" / "Future-proof"
- "10x your productivity"
- "The AI revolution"
- "In the age of AI"
### Generic Insider Claims
- "Here's the part nobody's talking about"
- "What nobody tells you"
- Anything with "nobody" or "most people don't realize"
### The Big One (FATAL)
- "This isn't X. This is Y." and ALL variations.
- "Not X. Y."
- "Forget X. This is Y."
- "Less X, more Y."
- ANY sentence that negates one framing then asserts a corrected one.
- If even ONE of these appears, the output fails. Delete the negation, just state the positive claim.
## Writing Samples
[Paste your writing here. The more you give, the better the voice match.]
———
the banned phrases list alone is honestly worth the file.
once you read through it you'll start noticing these phrases in literally every AI-generated slop-post you've ever seen
but the writing samples are what take it from "decent" to "wait did i write this"
setup takes maybe 10 minutes. copy the file, find your old writing, paste it in.
do it once and every session after that claude cowork reads it before you say a word
All you need is a idea.
We’ll give you $100,000+ and an office space for you to go all in
You don’t need to have revenue, a cofounder, or even an MVP
Applications are on a rolling basis, and we will write you a check as early as tomorrow
COMMENT your email for a refferal ⬇️
Stripe's CEO never studied finance.
Instead, he reverse-engineered PayPal’s docs—and built a $95 BILLION company at 22.
How? A learning method so powerful, it's now taught in colleges.
Here's his genius framework for learning anything fast: 🧵
If you have developers in your organization using Dev Tunnels in VS Code there are now group policy controls to allow you to manage the configuration. You can disable Dev Tunnels, disable anonymous tunnels or lock down access to specific Entra tenants-
https://t.co/HdFpcAC7nF
Now the two areas I’m gonna look at are
Bluetooth and WiFi!
Bluetooth these days is way different than it used to be:
> generally devices are not in permanent discovery mode. This means even if you knew their MAC address, I can’t seem to sniff them in the airwaves! (More research required though)
WiFi we have two things to consider:
> WiFi clients
> Phones broadcasting a hotspot via WiFi or Bluetooth (loads of people (ok some people) seem to have this turned on)
Thank you for all the support!
I have studies many RGs and ISPs for the last 8 months, deemed most of them to be unsafe, and presented "Security analysis of Residential Gateways and ISPs" at @BlackHatEvents.
Slides will be available shortly on website. WP is also upcoming.
With the recent discontinuation of the free version of VMware's ESXi, organizations that have relied on it for virtualization are in crisis mode. They are asking, "What do we do now?"
We have the answer... 👇
Meet the Proxinator - a turn-key solution that integrates virtualization with direct access to high-speed storage. This powerful hardware comes pre-installed with #Linux, #Proxmox, and end-to-end support.
Check out the latest article from @TheRegister: https://t.co/m01j4RsXbM
Then, head on over to https://t.co/zI2ykhfcJ8 to learn all about it.
Happy Monday! Today @Carlos_Perez and I are releasing a blog on adversarial LDAP tradecraft.
In this write-up we show:
- Normal LDAP queries you might see
- Common LDAP queries adversaries and red teams use
- Telemetry you can use to see these LDAP queries
- A way to get around the logging
Check it out: https://t.co/8tuhDIZZ3m