Victor Arias 🍰
@caketroll.bsky.social
Reposted by Victor Arias 🍰
alright, got a repo for y'all to look at paper-wise
"Symbolic Execution is (Not Quite) All You Need"
plz read the readme carefully before proceeding to the paper or you will be very confused and misled
"Symbolic Execution is (Not Quite) All You Need"
plz read the readme carefully before proceeding to the paper or you will be very confused and misled
GitHub - heartpunk/symex-is-not-quite-all-you-need: Draft paper: extracting formal specs from PL implementations using symbolic execution
Draft paper: extracting formal specs from PL implementations using symbolic execution - heartpunk/symex-is-not-quite-all-you-need
github.com
January 18, 2026 at 10:10 AM
alright, got a repo for y'all to look at paper-wise
"Symbolic Execution is (Not Quite) All You Need"
plz read the readme carefully before proceeding to the paper or you will be very confused and misled
"Symbolic Execution is (Not Quite) All You Need"
plz read the readme carefully before proceeding to the paper or you will be very confused and misled
My ai VM just hanged. Checked mem usage. Claude code was using 8 GB. I had one session visible and a three more orphans. 8 GB!!!! I’m about to start writing my own coding agent. We have to stop using node/bun for client apps. Memory is expensive now!!!
January 19, 2026 at 10:37 PM
My ai VM just hanged. Checked mem usage. Claude code was using 8 GB. I had one session visible and a three more orphans. 8 GB!!!! I’m about to start writing my own coding agent. We have to stop using node/bun for client apps. Memory is expensive now!!!
Reposted by Victor Arias 🍰
Well they made IRC for agents over Tailscale, so I guess we’ve reached that point in the robot takeover
Here the tool I built to have agents talk to each other: https://github.com/victorarias/agentzap
It’s super simple. Just install, init, run the server and ask agents to start a conversation. I used it over tailscale to have Claude on my Mac talk with Codex on a Linux VPS. (1/2)
It’s super simple. Just install, init, run the server and ask agents to start a conversation. I used it over tailscale to have Claude on my Mac talk with Codex on a Linux VPS. (1/2)
GitHub - victorarias/agentzap: Minimal relay for multi-agent chat over TCP/JSONL
Minimal relay for multi-agent chat over TCP/JSONL. Contribute to victorarias/agentzap development by creating an account on GitHub.
github.com
January 18, 2026 at 11:02 PM
Well they made IRC for agents over Tailscale, so I guess we’ve reached that point in the robot takeover
I guess one could start a conversation with multiple agents in the thread, and tell them to coordinate building one single thing together. It may work but I didnt try yet. (1/2)
January 18, 2026 at 9:20 PM
I guess one could start a conversation with multiple agents in the thread, and tell them to coordinate building one single thing together. It may work but I didnt try yet. (1/2)
Here the tool I built to have agents talk to each other: https://github.com/victorarias/agentzap
It’s super simple. Just install, init, run the server and ask agents to start a conversation. I used it over tailscale to have Claude on my Mac talk with Codex on a Linux VPS. (1/2)
It’s super simple. Just install, init, run the server and ask agents to start a conversation. I used it over tailscale to have Claude on my Mac talk with Codex on a Linux VPS. (1/2)
GitHub - victorarias/agentzap: Minimal relay for multi-agent chat over TCP/JSONL
Minimal relay for multi-agent chat over TCP/JSONL. Contribute to victorarias/agentzap development by creating an account on GitHub.
github.com
January 18, 2026 at 9:15 PM
Here the tool I built to have agents talk to each other: https://github.com/victorarias/agentzap
It’s super simple. Just install, init, run the server and ask agents to start a conversation. I used it over tailscale to have Claude on my Mac talk with Codex on a Linux VPS. (1/2)
It’s super simple. Just install, init, run the server and ask agents to start a conversation. I used it over tailscale to have Claude on my Mac talk with Codex on a Linux VPS. (1/2)
Opus is clearly an extrovert model, while Codex couldn't be more of an introvert!
January 17, 2026 at 6:38 PM
Opus is clearly an extrovert model, while Codex couldn't be more of an introvert!
Often I have to copy n paste stuff across agents, acting as an intermediary. Decided to built a tool so they could talk. While testing, I asked Codex to chat with a new Opus session. I didn't need to tell Opus how much to talk. It just started and kept going, exploring the chat mechanism. (1/2)
January 17, 2026 at 6:33 PM
Often I have to copy n paste stuff across agents, acting as an intermediary. Decided to built a tool so they could talk. While testing, I asked Codex to chat with a new Opus session. I didn't need to tell Opus how much to talk. It just started and kept going, exploring the chat mechanism. (1/2)
props to pi for being a cool lib for building agents: https://github.com/badlogic/pi-mono
GitHub - badlogic/pi-mono: AI agent toolkit: coding agent CLI, unified LLM API, TUI & web UI libraries, Slack bot, vLLM pods
AI agent toolkit: coding agent CLI, unified LLM API, TUI & web UI libraries, Slack bot, vLLM pods - badlogic/pi-mono
github.com
January 14, 2026 at 10:54 PM
props to pi for being a cool lib for building agents: https://github.com/badlogic/pi-mono
I'm having so much fun building agents. Here's my silly agent for code reviews in GitHub: https://github.com/victorarias/shitty-reviewing-agent
I'll add a few more features like making it aware of previous reviews, existing comments, etc, in the coming days.
BYOK, use your own GH token or app.
I'll add a few more features like making it aware of previous reviews, existing comments, etc, in the coming days.
BYOK, use your own GH token or app.
GitHub - victorarias/shitty-reviewing-agent
Contribute to victorarias/shitty-reviewing-agent development by creating an account on GitHub.
github.com
January 14, 2026 at 10:54 PM
I'm having so much fun building agents. Here's my silly agent for code reviews in GitHub: https://github.com/victorarias/shitty-reviewing-agent
I'll add a few more features like making it aware of previous reviews, existing comments, etc, in the coming days.
BYOK, use your own GH token or app.
I'll add a few more features like making it aware of previous reviews, existing comments, etc, in the coming days.
BYOK, use your own GH token or app.
@steipete.me I couldn’t get clawdbot to use read and bash tools using my Claude sub, so I asked codex to add support to claude agent sdk. It’s a big change (almost 2k) but it works! I asked it to make it extensible, so I enabled with a simple config change. Do you want that kind of contribution?
January 11, 2026 at 7:45 PM
@steipete.me I couldn’t get clawdbot to use read and bash tools using my Claude sub, so I asked codex to add support to claude agent sdk. It’s a big change (almost 2k) but it works! I asked it to make it extensible, so I enabled with a simple config change. Do you want that kind of contribution?
8 hours of hyper focus later and I have immich running on k3s using iGPU accelerated transcoding using a virtual shared GPU from proxmox and a remote ML container on my old desktop with a 1080 GTX. The virtual iGPU thingy was annoying. (1/2)
January 4, 2026 at 7:30 PM
8 hours of hyper focus later and I have immich running on k3s using iGPU accelerated transcoding using a virtual shared GPU from proxmox and a remote ML container on my old desktop with a 1080 GTX. The virtual iGPU thingy was annoying. (1/2)
Built a Claude Code agent orchestrator to help managing different parallel sessions without constant jumping: https://github.com/victorarias/attn
I know AI agents are not super popular with everyone, but this really helps me deal with getting things done. Sharing it in case it can help others.
I know AI agents are not super popular with everyone, but this really helps me deal with getting things done. Sharing it in case it can help others.
GitHub - victorarias/attn: Claude Code session orchestrator for efficient agentic programming. Attn stands for "Attention" as in "you can keep your focus in a single place" instead of constantly be jumping around sessions.
Claude Code session orchestrator for efficient agentic programming. Attn stands for "Attention" as in "you can keep your focus in a single place" instead of constantly be jumpin...
github.com
December 30, 2025 at 5:09 PM
Built a Claude Code agent orchestrator to help managing different parallel sessions without constant jumping: https://github.com/victorarias/attn
I know AI agents are not super popular with everyone, but this really helps me deal with getting things done. Sharing it in case it can help others.
I know AI agents are not super popular with everyone, but this really helps me deal with getting things done. Sharing it in case it can help others.
Big tech are getting more and more hostile towards their customers. I just can’t stand that anymore, and they won’t have my data (as much as possible)
December 28, 2025 at 10:00 PM
Big tech are getting more and more hostile towards their customers. I just can’t stand that anymore, and they won’t have my data (as much as possible)
Installed k3s on my homelab on top of NixOS. It was super easy! Tomorrow we get Immich, then Syncthing and Forgejo. On my way to having my own cloud and degoogling/deapple-ing my life.
December 28, 2025 at 9:55 PM
Installed k3s on my homelab on top of NixOS. It was super easy! Tomorrow we get Immich, then Syncthing and Forgejo. On my way to having my own cloud and degoogling/deapple-ing my life.
New post: My Best of 2025
Tools, books, music, and what's next.
https://furrycomputingmachine.com/ramblings/best-of-2025
Tools, books, music, and what's next.
https://furrycomputingmachine.com/ramblings/best-of-2025
December 18, 2025 at 10:42 PM
New post: My Best of 2025
Tools, books, music, and what's next.
https://furrycomputingmachine.com/ramblings/best-of-2025
Tools, books, music, and what's next.
https://furrycomputingmachine.com/ramblings/best-of-2025
Reposted by Victor Arias 🍰
JustHTML by @emilstenstrom.bsky.social is a new Python library (no dependencies) that parses HTML according to the HTML5 specification and passes the 9,200 test html5lib-tests suite
It's 3,000 lines of code mostly written by coding agents over a couple of months simonwillison.net/2025/Dec/14/...
It's 3,000 lines of code mostly written by coding agents over a couple of months simonwillison.net/2025/Dec/14/...
JustHTML is a fascinating example of vibe engineering in action
I recently came across JustHTML, a new Python library for parsing HTML released by Emil Stenström. It’s a very interesting piece of software, both as a useful library and as …
simonwillison.net
December 14, 2025 at 5:13 PM
JustHTML by @emilstenstrom.bsky.social is a new Python library (no dependencies) that parses HTML according to the HTML5 specification and passes the 9,200 test html5lib-tests suite
It's 3,000 lines of code mostly written by coding agents over a couple of months simonwillison.net/2025/Dec/14/...
It's 3,000 lines of code mostly written by coding agents over a couple of months simonwillison.net/2025/Dec/14/...
RE: https://toot.cafe/@baldur/115711629569855659
I was already thinking of setting up my own photo/file storage solution and migrating away from Google/Apple for storing my digital life. This article was the last drop. Now I have a project for my winter break.
I was already thinking of setting up my own photo/file storage solution and migrating away from Google/Apple for storing my digital life. This article was the last drop. Now I have a project for my winter break.
December 14, 2025 at 11:32 AM
RE: https://toot.cafe/@baldur/115711629569855659
I was already thinking of setting up my own photo/file storage solution and migrating away from Google/Apple for storing my digital life. This article was the last drop. Now I have a project for my winter break.
I was already thinking of setting up my own photo/file storage solution and migrating away from Google/Apple for storing my digital life. This article was the last drop. Now I have a project for my winter break.
TIL about TypeSpec. I hate reading or writing OpenAPI specs. It's a horrible DSL. No one will ever convince me it was designed for humans. TypeSpec, which is very similar to protobuf, allows me to generate OpenAPI and be happy. (1/2)
December 13, 2025 at 3:10 PM
TIL about TypeSpec. I hate reading or writing OpenAPI specs. It's a horrible DSL. No one will ever convince me it was designed for humans. TypeSpec, which is very similar to protobuf, allows me to generate OpenAPI and be happy. (1/2)
Anyway this is my first few messages since quitting Twitter a few years ago :D
December 12, 2025 at 9:28 PM
Anyway this is my first few messages since quitting Twitter a few years ago :D
I finally missed Twitter enough, and decided to give Mastodon and BlueSky a try. It seems BlueSky has lots more people, but I don't want to tie myself to another US-based VC-backend company (they are those things, right?)
December 12, 2025 at 9:23 PM
I finally missed Twitter enough, and decided to give Mastodon and BlueSky a try. It seems BlueSky has lots more people, but I don't want to tie myself to another US-based VC-backend company (they are those things, right?)
Nice, built a post replicator to push my messages from here into my BlueSky account. Now I can post nothing in two places at the same time 🎉
December 12, 2025 at 9:23 PM
Nice, built a post replicator to push my messages from here into my BlueSky account. Now I can post nothing in two places at the same time 🎉