Kali Charan

[Kali Charan @KaliCharanhere]

@tobiasjames_eth Yep, I'm experimenting with combining Wayfinder and the teach skill for learning. Wayfinder plans the path, teach skill teaches it one lesson at a time.

[Kali Charan @KaliCharanhere]

@Jhaddix @mattpocockuk Cool, for me it's hybrid. First I pour my full context, then at the end I tell it to ask me if it has any questions. That's where I use grill-me 🙂.

[Kali Charan @KaliCharanhere]

Biggest takeaway from Opus 5.5: your last mistake doesn't define you, how you respond to it does. Take the feedback, refine, ship. People will come back.

#Claude #Anthropic #Buildinpublic

[Claude @claudeai]

Introducing Claude Opus 5.5, the first model in our new Claude 5.5 family.

It performs at the level of Claude Fable 5.1 for most tasks, and costs 40% less to run than Opus 5.

[Kali Charan @KaliCharanhere]

@smileyboyoff My fav suit

[Kali Charan @KaliCharanhere]

@vasloga Me and my friend in our 70s

[Kali Charan @KaliCharanhere]

@mattpocockuk 's /wayfinder and /teach skill makes the killer combo

Wayfinder creates me a full curriculum of an unknown topic week by week plan with my own destination.

Teach takes each week plan and created me the tailored lessons .

Which is super cool 😎

[Kali Charan @KaliCharanhere]

@yadavji_codes Soo accurate 😂

[Kali Charan @KaliCharanhere]

Islam wins
Ian proves

#ufc330 #mma

[Kali Charan @KaliCharanhere]

@Im_IrushiK Yes right now , literally opened x to see it is a thing . First prompt of the day my limit crossed 93%

[Kali Charan @KaliCharanhere]

Today another incident - Degraded performance on Claude Sonnet 5

#claude #claude_incident

[Kali Charan @KaliCharanhere]

Anna varaaar valividuu 💀

[Marvel Studios @MarvelStudios]

Ghost Rider enters the MCU 🔥

Ryan Gosling will star in Marvel Studios' Ghost Rider, directed by Shawn Levy. #SDCC

[Kali Charan @KaliCharanhere]

The future of AI security isn't just about protecting what models say—it's about controlling what agents can do.

[Kali Charan @KaliCharanhere]

We need to worry far more about AI agent security than chatbot security.

A RAG bot mostly retrieves information and generates text. An AI agent can take actions—access systems, call APIs, execute workflows, make decisions, and interact with real-world infrastructure.

[Kali Charan @KaliCharanhere]

As AI agents become more autonomous and capable, the attack surface grows just as fast. Prompt injection is only the beginning. Agent identity, permissions, memory, tool access, and runtime security will become critical.

[Kali Charan @KaliCharanhere]

Detection approach (planned):

Regex/keyword rules
Semantic detection
LLM-as-judge — catch multi-turn attacks that unfold slowly across a conversation

Starting the build. Will share progress as I go

[Kali Charan @KaliCharanhere]

Building an LLM SIEM 🛡️

Been thinking about this a lot: LLM apps are a new attack surface (prompt injection, jailbreaks, PII leakage) and most companies have zero visibility into it. Traditional SIEMs weren't built for this.

So I'm building one

[Kali Charan @KaliCharanhere]

Core features:

Detect prompt injection attempts
Real-time alerts to security teams
Block/prevent PII from leaking to the LLM
Giving full visibility on Llm prompts and responses

[Kali Charan @KaliCharanhere]

@Vigresh711 Goooosebumpps..

[Kali Charan @KaliCharanhere]

We got our spiderman with organic webs .PEAK 🤩

[Pop Crave @PopCrave]

Tom Holland in ‘Spider-Man: Brand New Day.’