Viewing profile — alexsmirnov

"The single worst issue: the system prompt never explicitly tells it to ignore auto-injected files like CLAUDE.md or *.rule.md." - Yes, it is! Right after the content of CLAUDE.md,…

I did run both kimi-k3 and glm-5.2 with Capital One vulnhunt [1]. No rejections, they did find the same problems on my project that I used for testing. gpt-5.6, gemini pro, and opu…

Shit happens. I had my iMac and Time machine hard drives died 2 days apart...

The problem of duplicated code described in the article, goes in a different way in reality: AI does not update 4 places in the same way, but implement them a slightly different. I…

> This happens but far less often than it used to, and the case for full autonomous agents is getting stronger, not weaker. This is that I do not see. My journey, just couple weeks…

If I pay for vending machine by corporate card on a business trip, it looks more like B2B

From the first hand: I was in the air base service regiment, Ovruch ( 30 km from reactor ), 1 year in service at the time of disaster. The average numbers have a little sense. By o…

I more like my high school math teacher explanation: "Put your one hand on dry ice, another in the boiling water. On average, you feel warm and cozy"

We usually squash feature branches before merge. To squash before rebase, I use git reset --soft $(git merge-base develop HEAD) && git commit && git rebase develop - you have to re…

Much longer than that, and was available way before an internet. I graduated STEM high school in St. Petersburg in 1981, and I had several classmates who were big funs of chemistry…

Exactly. I did create my own MCP with custom agents that combine several tools into a single one. For example, all WebSearch, WebFetch, Context7 exposed as a single "web research" …

Almost instantly, compared to my experience working for a big health care provider... I waited 6 moths for IT department to allow me install development tools on work laptop. And w…

> you start off checking every diff like a hawk, expecting it to break things, but honestly, soon you see it's not necessary most of the time. I see it's necessary ALL the time. Th…

The calculation misses subagents tokens, that can be a significant differences. Better to parse session jsonl files ~/.claude/projects/ / .jsonl and ~/.claude/projects/ / /subagent…

This mostly matches my own estimates for pr-review command that I use. But it's pretty sophisticated: 6 specialized agents, best practices skills, CVE database, bunch of scripts. T…

Actually, they extremely bad at that. All training data contains cod + tests, even if tests where created first. So far, all models that I tried failed to implement tests for inter…

Considering how little data needed to poison llm https://www.anthropic.com/research/small-samples-poison , this is a way to replace SEO by llm product placement: 1. create several …

I do in similar way, connect claude code to litellm router that dispatches model requests to different providers: bedrock, openai, gemini, openrouter and ollama for opensource mode…

For me, AI is the best for code research and review Since some team members started using AI without care, I did create bunch of agents/skills/commands and custom scripts for claud…

I did create and actively use a similar tool, but with different purpose: configure AI tools for each team member to use the same code style and architecture guides across projects…

I do use Obsidian on pair with Claude code and git. I organize notes by tags, folders, and links from tree of "map of content" notes. Those documented as rules for AI. All notes ca…

This was implemented far ago, at least by huggingface "smolagents". https://huggingface.co/docs/smolagents/index . I did use them, with evaluations. For the most cases, modern mode…

This is exact the impression that I got. Every question or task given to LLM returns pretty reasonable, but flawed result. For the coding, those are hard to spot but dangerous mist…

This is not about understanding the message, but switching user mental activity. I go myself in the similar situations many times. One example: I tried to pay my bills in online ba…

A lot of discussions around vibe AI coding flaws: awful architecture, performance problems, security holes, lack of maintainability, bugs, and low code quality. All correct, but no…