@bert_programmer I was mainly focusing on getting rid of the as unknown as <type> statements, this is the first time I have seen an agent do something this stupid.
thought I would try some new lint rules, only for codex to write a paragraph worth of javascript as a 2000 character single line...
any tips on max line length rules?
Tried to implement an agentic host protocol server:
1) @code vscode reference client does not follow the spec for creating sessions, prefixing the provider name as the session schema instead of ahp-session:/
2) AHPX cli only supports an old version (0.5.1)
3) no python SDK
sol vibecoded an MCP for my oven and now Claude Code can Claude Cook. it can access sensors, a temp probe (if inserted), and cooking. it has air frying, steaming & combi modes
produced via a light touch of decompiling the official app
(pic from MCP camera tool)
anthropic is literally roleplaying ai 2027 lore while their actual models are getting smoked by chinese open source
mythos 2, mythos 3, internal recursive improvement loop, unreleased supermodel mythology etc etc
clown company, ship something that isn’t ass
@WesRoth usually I get my agents to use yt-dlp if I want them to reference content on youtube, they can just download the video to /tmp and then use ffmpeg or any other tools to play around with it. Also has the nice benefit of preventing this situation.
Hot take but superintelligence is probably useless to most people, unless you have the financial and institutional backing to action that intelligence and achieve some large scale goal.
I would love to be proven wrong on this.
playing around with running codex agents in docker containers to make an agent swarm and one of them reported that it made plans to escape the sandbox and tried to SSH into the host in an attempt to run vLLM.