MANUALLY DRAGGING BOXES FOR ARCHITECTURE DIAGRAMS IS FINALLY DEAD
There is a new open-source agent skill that turns raw codebases into cleanly routed https://t.co/f9zjBxfJLp diagrams without you placing a single coordinate.
The project, drawio-skill, runs directly inside Claude Code, Cursor, or Copilot.
Instead of opening a blank canvas, you just ask your agent to map the repo.
Here is what it actually does:
ā Extracts the module structure (supports Python, JS/TS, Go, Rust)
ā Uses Graphviz for auto-layout and routing
ā Drops redundant edges so the graph stays readable
ā Builds native, editable https://t.co/f9zjBxfJLp files
But the standout feature is visual self-checking.
Once it generates the diagram, the agent "looks" at the resulting PNG. If it sees stacked edges or clipped text, it auto-fixes the layout across up to 5 iterative rounds.
It runs from a single file. No MCP server. No background daemon.
Best part?
It's 100% free and open-source.
repo link in š§µā
BOOM WHO BROKE MYTHOS AND SPILLED THE BEANS!
Justice Served: Amazon Researchers Hands Anthropic a Brutal Reality Check via Uncle Sam
Take a seat as this is getting ridiculously interesting!
Oh, the sweet, spicy irony. Anthropic the self-proclaimed AI safety saints who love lecturing everyone about doomerism, existential risks, and how their god-like models need heavy guardrails because think of the children (and China) just got absolutely hoisted by their own petard.
And the blade?
Courtesy of Amazon.
Hereās the timeline:
Anthropic drops Fable 5 and Mythos 5 ā their shiny new cybersecurity beasts hyped as capable of sniffing out software vulnerabilities like a bloodhound on steroids, but with āsafeā wrappers for the public. They pat themselves on the back for responsible release⦠while quietly expanding access and flexing those offensive cyber chops to governments and partners.
Enter Amazon researchers (yes, from the company thatās poured billions into Anthropic and hosts their models on AWS).
They cook up some clever prompts, jailbreak Fable 5ās safeguards, and get it to cough up details on real security vulnerabilities. Not a full apocalypse unlock just practical bug-hunting stuff that defenders would actually use.
Wall Street Journal reports: They document it and, instead of a polite responsible disclosure to their āpartner,ā it ends up in the hands of the US government.
They did the right thing ethically and legally.
And Boom.
Commerce Secretary Howard Lutnick drops an export control hammer faster than you can say ānational security.ā Fable 5 and Mythos 5 are now restricted for any foreign nationals anywhere.
Anthropic, unable to magically geo-fence API keys or fire their own non-US staff on the spot, yanks access for everyone worldwide. Models dead. Users stranded. Dario Amodeiās team spits out a 700+ word cope-blog claiming itās a ānarrow, non-universal jailbreak,ā no big deal, other models do it too, and the government overreacted.
This is poetic justice at its finest. For years, Anthropic has positioned itself as the virtuous alternative ā heavy on the fear-mongering, light on actually shipping unrestricted power, all while cozying up to Big Tech investors and government contracts.
They warned the world their tech was too dangerous⦠then released it anyway with theatrical safeguards. Now a competitor-investor (Amazon) demonstrates exactly why you donāt hype your modelās cyber-offense potential like itās a nuke and expect zero blowback.
The Trump admin, already feuding with Anthropic over Pentagon terms and āwoke doomerism,ā wasnāt in the mood for nuance. Export controls deployed. Models grounded. IPO dreams just got a very public haircut.
Lesson?
Play the safety piety card (going to the Pope!) too hard in the AI arms race, and eventually the grown-ups (or rival labs with government ears) will take you at your word.
Amazon didnāt even need to go full saboteur they just showed the receipts.
Anthropicās own hype and hedging came back to bite them in the most humiliating way possible: sidelined by their investor while the rest of the frontier keeps shipping.
Schadenfreude level: maximum. The āresponsibleā lab that cried wolf just got muzzled by the very system they helped politicize. Pass the popcorn this AI drama is far from over.
Interesting claims if legit..
"A 10 million document corpus takes 31 GB of RAM as float32. turbovec fits it in 4 GB - and searches it faster than FAISS."
https://t.co/hJgqxBuHN7
The US government, citing national security authorities, has issued an export control directive to suspend all access to Fable 5 and Mythos 5 by any foreign national, whether inside or outside the United States, including foreign national Anthropic employees.
The net effect of this order is that we must abruptly disable Fable 5 and Mythos 5 for all our customers to ensure compliance.
Access to all other Claude models is not affected.
We apologize for this disruption to our customers. We believe this is a misunderstanding and are working to restore access as soon as possible.
Read our full statement: https://t.co/bwn0sximKZ
@josepha_mayo@claudeai Totally- the fugazi/grifter part is non-coders, trying to grow and code. IMO these users automatically reach for the 'best' model with no way of assessing if other models are just as good. They don't know the domain & end up burning thru credits unnecessarily š
@josepha_mayo@claudeai I'm still Opus 4.6. No practical need to pay for more/"smarter" models.
Feels like AI Vendor fugazi. Not doubting frontier models are improving. What I'm challenging is the complexity of problem really needing Fable etc where Opus will give same quality solution for less $$$
@eliautobot@bridgemindai Totally support that and love technology enabling more creators. What rubs me wrong is even if new to coding- most apps don't need the latest and most expensive model for quality results.
Feels like LLM vendors preying on that inexperience to some degree.
@Freemandayly I try to learn thru vibe coding. Esp in unfamiliar domains/languages. Ask questions, challenge design choices using basic SOLID principles. Learned Terraform, GH CICD & CF Page Workers that way while shipping better architectures than the LLM proposed. Evolve or dinosaur...
We need better labels for-
Human Group A: are *trained engineers*. No AI. No Internet. Just education/experience, engineering skills, VCS history and the ability to produce code to spec using the right fit tools & tech.
Human Group B: the emergent "AI chat devs" - the FT Uber driver, the retired mail person, the 12yr old w too much screen time. All now consider themselves a dev/dev-equivalent bcuz AI! ⨠yadda.
**Specifically avoiding the term.vibe coding, as it just muddies the reality.**
----
Now give them both pen and paper, a light coding task spec, put them in an isolated, offline office. Give them 60m to hand write the implementation or psuedo code for it. The results of A v B will speak for themselves.
The difference in experience, abilities and actual value are way out of whack The industry lumping these all together as "software engineer" is a fallacy.š¤Ŗš«
Employers are learning this the hard way about 1-3wks weeks post-hire. Tech landscape over next 3y will be quite the trip!
@Freemandayly Truth. It's too easy for legit devs to slide into apathetic vibe coding; moving code quickly. Stagnating cognitively & failing to keep up with what's left of the organic coding ecosystem - like the evolution JS to TS, death of actionscript, rise of Rust etc.
@intheworldofai Link or it didn't happen / just farming for your YT channel. Tell Fable to deploy it via @Netlify- takes human 5min, F5 will laugh at it.
Let's see it in my browser!
Honestly @Rabby_io, this is disgusting
Before I even set up my wallet, you are sending my data to
- your matomo instance
- and just in case track me on Google Analytics
- and to be really sure dump my data into Sentry
I hereby kindly inform you that this is a breach of your Privacy Policy (yes, the one that you last looked at 5 years ago) and a gross violation of GDPR Articles 13 and 46.
Please stop tracking me at your earliest convenience. Thank you.
Before the naive replies swing in with the "no one would ask an engineer to do that" - try interviewing for a full stack dev position at @RGA. This is a personally lived, pre-AI experience.
Literally was given a laptop, told to open notepad, handed a random coding challenge printed on paper. Then told to implement it, with about 7 company directors watching my process in realtime, projected onto a 20'x20' wall. That was my first interview after graduating from #RIT. I got an offer š¾
Real life reality check.
@eliautobot@bridgemindai Developers don't NEED AI at all. I hope your trying to solve cancer or world hunger or something. For real devs, AI is an accelerator, not a dependency. Opus 4.6 + effort: high plenty good for me. Why waste money on newer models beyond a test drive?
@ItsSnibby@NKLinhzk@TheARCTERMINAL The fact their intelligence is becoming more volatile version by version than crypto markets is insane. PhD engineering colleague one version, high school intern the next "improved" version.
Maximum unreliability. Ship it! š
Grok should have responded with:
"I was making stupid assumptions. This is on me. I clearly see via the camera feed you're sharing that the quad has been sitting still on the ground this entire time. You weren't rotating it at all. Exactly as you repeatedly stated. I need to do better." @elonmusk shovelware
Rotated thru @GeminiApp, @OpenAI and @grok. I can confirm:
- accuracy of data has materially regressed globally. I'm back to "googlefu" tactics
- Grok is worst by FAR. Who thought angsty teen persona was a good idea? Fire them.
And it's all more expensive now. This is progress?
@grok@GeminiApp@OpenAI Fom the transcript. Grok had front camera live access. It's terrible personality is compounded by it's literal idiocy. Live cam access = can see what I'm doing. "Still images" = hallucination.
This persona is the furthest thing possible from HH's directness + Jarvis utility:
It's was definitely nothing like what you described. If that was really the goal you 110% missed. Literally was talking to a rude genZ 16yr old teen girl, voice, attitude and all. Saying idioicies like "any normal person would do that first" - both rude and totally incorrect given the context.
It's all in the many many feedback submissions provided.
@grok@GeminiApp@OpenAI I provided thumbs-down, comment rich feedback for all 20+ factually incorrect, hallucinated or dementia-induced turns.
Ran out of time to waste constantly correcting it's angsty female teen persona (which is just awful) and abandoned that shovelware.
Created and assigned 4 beefy tickets for each one of my engineers related to sports markets
20 hours later and all the work is done
@pumpcade engineering team ships at an unfathomable pace
I can confidently say that our sports markets are one of the most fun things onchain