#NewProfilePic

Bugs fixed & fully tested. Claude Code prompt generation for Ralph loops that cache knowledge each iteration, reducing loops + new plugin for templates that creates tests first outside of the model, so the AI won’t code to “pass the test” #AI #ralphloop #claudecode

Boris Cherny Running parallel Claude Code instances on the same repo — the biggest pain point isn't model capability, it's coordination. Agents step on each other's files, context drifts, merge conflicts pile up. Would love to see a built-in session awareness layer: 'agent A is editing

In 4 mins, I’ll prove that 2 plugins based around modular templates feeding a loop with tmp caching in a sandbox creates a system that continuously learns. No long prompts, no context window worry, no custom agents. Meet the #AnvilMethod #ClaudeCode #AI github.com/kt2saint-sec/a…

Asked for proof? Ok. See how much I needed to prompt for a fully tested API key vault for fun. Zero project agents, only 2 VERY custom plugins..and Claude 4.6 extended thinking couldn’t find a fault. Meet the #AnvilMethod #ClaudeCode #OpenSource github.com/kt2saint-sec/a…

EVALUATOR -> CHALLENGER → PROXY → RESEARCHER → TESTER → BUILDER → CODE REVIEWER → SIMPLIFIER Every phase is gated. Every 8 specs Boris checks for architectural drift and can rewrite the plan. Final integration runs ALL tests simultaneously. Test driven, cheap, SELF LEARNING

Running out of context w/Claude? new sandbox blocks Bash except Read in /tmp, which saved $ running in RAM. Also, same port as ollama MCP by default. Resolved. 4 plugins achieve TDD, Agent Teams in RAM, local think LLM access via plugin github.com/kt2saint-sec/a… Claude Code Community

When you build your own plugin that mixes agent teams, Boris Adversarial Challenges, a HiTL Proxy, a Tester, and a Builder who operates sandboxed looping until tests pass, and monitoring it all without tmux. Life is good.

Built a new skill: /diagnose It answers one key question after a test fails: Is the bug in the code, or is the test stale? This doesn’t change your code, it coordinates what needs to be fixed, including if code is correct and GitHub CI is wrong. github.com/kt2saint-sec/a…

I made these Claude Code skills as a full pipeline: plan, execute fast, challenge, TDD, verify results. 8 modular commands cover PRDs, MVP scope, stack planning, cost analysis, build loops, second opinions, and test diagnosis. #ClaudeCode #LLMs #Claude github.com/kt2saint-sec/c…

Welcome to 100k+ context local abliterated model, with thinking, 100% local. Turboquant is a game changer. Other local models can easily reach 400k context. Benchmarked a few, ROCm 7.2.1 compatible. github.com/kt2saint-sec/T…

Turboquant will definitely be something everyone will need to look at. Running 27b parameter qwen 3.5 abliterated with thinking not only works, but scores show it beats big tech models in many areas. #Turboquant #LocalAI #ClaudeCode #OpenAI #Codex #Ollama

Watching new Claude Code users get mad is almost a hobby now. It’s like watching someone build a cheap IKEA set without reading the instructions, and then seeing them yell at the box & screwdriver. #claude #claudecode #AI #vibecoding #LLMs

🚨 Breaking: Alibaba just killed the browser automation stack. **page-agent.js** — a GUI agent that lives directly inside your webpage. No Selenium. No Puppeteer. No Chrome extension. No Python backend. Just one script tag. It reads your DOM as text (no screenshots, no

Meet #MultiClawProtocol: Self Improving Agents, Simple #Hermesagent “brain” orchestrating a few, or teams of #openclaw agents. -Works with local models (text+vision+image for orchestration & navigation) -3D model, Terminal, API keys, sandboxing. v1.0 avail on GitHub 12pm est.

Anthropic gives me $200 over people upset at 3rd party support, specifically OpenClaw/Hermes-Agent. GUI Drag & Drop, 18 preset agents, fully customizable, 3D graphs, terminal/telegram integration, logs..just BYOK & Curl, no n8n headache. github.com/kt2saint-sec/m…

Someone just dropped a 9-layer production AI architecture and it's the most honest breakdown I've seen. services/ - RAG pipeline, semantic cache, memory, query rewriter, router. Not one file. Five. agents/ - document grader, decomposer, adaptive router. Self-correcting by