Kostya

I built an HTTP debugging proxy that runs in your terminal. Introducing httpmon - a TUI HTTP monitor with: - Vim-style filtering - JavaScript interceptor scripts - gRPC-Web request debugging - HAR export - MCP server for AI model access to captured traffic Website, docs, and a product video - all built with AI assistance. The funny thing is that I enjoyed building the marketing part more than the code, it's amazing what can be achieved with remotion to create videos. I linked it to ElevenLabs to record the voice overs. Apparently there are tons of tools that can also record a TUI session in the terminal to create awesome looking demos.

Hey @cursor_ai Did you even try using this? This feature is broken. When creating automations it's impossible to select a cloud agent without selecting a repo, however then the tasks are not being picked up by "any repo" workers. A simple human smoke test would show you this is not working. I want all my cloud agents to run on my infrastructure. For automations if I don't select any repos its impossible to select what worker pool should run on, and it never gets picked up. Did you actually try using it through Slack? Are users to tag Cursor in slack and specify those arguments such as repo or a worker pool for each invocation? Is this your answer to Devin? Super disappointed with this feature.

Every retail site should be agents friendly. There is absolutely no reason I should waste time manually browsing an e-commerce website looking for products that I need. I want to point my agent and ask it to find the best VFM skis for me, check reviews for each one and give me a top 10 options and not waste my time doing it manually. If you block agents on your ecommerce website you are ngmi.

Code review has gotten a lot harder. Yes, there are a million code review skills and products out there, and most of them do a decent job - of producing walls of text. But as a reviewer, I still have to read all of it before I post any of it on the pull request. You could bury your head in the sand and hope for the best: maybe the agents are good enough at reviewing code on their own. The problem with not reading the comments your agent generates is that you can be almost certain the author won't read them either. They'll just hand them straight to their own agent to "fix." So as a reviewer, I end up spending more time on the actual review, not less. 1. I can't trust that the author truly understands the code they shipped, or the implications it carries. Does it handle every edge case? Twenty AI-generated tests don't mean the thing is actually tested. 2. I can't trust that anyone will read my comments. The responsibility bounces right back to me: read the author's code, read the AI's comments, and make sure both actually make sense. Some days it feels like it would be faster to just generate the code myself from scratch than to do all of the above. And AI cuts both ways here - it also lets you go much deeper down the rabbit hole. What's the blast radius of this change? How is this pattern implemented in other services? What are the second-order effects the author never considered? You can spend hours digging into this. So how are you handling this as a responsible code reviewer? Do you have a system that works? Or should we just let go and YOLO the slop?

After a month with GPT 5.5, switched back to Claude Max subscription to try Fable for a week. It's super slow and basically unusable, same for Opus 4.8, either everything is overloaded and I had a bad timing or GPT 5.5 feels much faster. Now I have to endure 1 month of Claude subscription before going back.

httpmon.dev Star, fork, enjoy :)

Hey guys I have Claude Code running on a VPS with tmux and I use my phone to control it (with Termius). Now I would like to have a voice mode so I can have a conversation with it and get executive updates instead of typing. Lets say when going for a walk. Did anyone build this?