Henry Chen

maxxsandboxing

[Gabriele Farei @jayfarei]

Sharing an OS tool I have been building for a few weeks: https://t.co/84Vdwo3Fqv - git for agent traces.

With OT, you have a local client that helps you santise, review and securely share traces from your coding sessions, privately or publicly, as a @huggingface dataset.

0.3 adds experimental session blame which I am very excited about - attributing commit/PR changes back to the traces that produced them (all local) unlocks a new layer in observability and issue tracking.

A few short demos of the local TUI client and the upcoming opentraces hub platform.

Try it out! keen on your feedback🙏

[Junyang Lin @JustinLin610]

https://t.co/mgVhhZVTKg

[Henry Chen @________chh]

npx skills add chenhunghan/code-mode-skill
https://t.co/zltXsbzkr9

[Henry Chen @________chh]

Local Code Mode for MCP

Most MCP servers just wrap CRUD JSON APIs into tools — I did it too. It works, until you realize a tool call dumps 100KB+ into context.

[Henry Chen @________chh]

Works best with well-known APIs (SCIM, Kubernetes, GitHub, Stripe, Slack, AWS) because LLMs already know the schemas — they write the extraction script in one shot.

[Henry Chen @________chh]

Inspired by Cloudflare's Code Mode, but using a local sandboxed runtime instead of a remote one — no external dependencies, isolated from filesystem and network by default.

[Henry Chen @________chh]

That's what code mode does. The LLM writes a small script, the server runs it in a sandbox against the raw data, and only the script's compact output enters context.

[Henry Chen @________chh]

MCP isn't dead — but we need to design MCP tools with the context window in mind.

[Mert Koseoglu @mksglu]

OpenAI published Symphony — a spec for turning project work into autonomous implementation runs. Manage work, not agents.

I took that idea and rebuilt it with Claude Code @AnthropicAI

hatice is my implementation of the Symphony spec, powered by Claude Code Agent SDK. It watches your Linear board. When you move an issue to "In Progress" — a Claude agent spins up, clones your repo, writes tests, implements the feature, and marks the issue done.

You move cards. Agents ship code.

In this video: I dragged 5 issues to "In Progress" at once. hatice spawned 5 parallel Claude agents. They built an entire presentation website from scratch — live, with hot-reload. I didn't write a single line.

One WORKFLOW.md file. That's the entire setup.

https://t.co/FNmN1dvQYW

cc: @bcherny @trq212 @muratcan

[Crypto Mafia @CryptoMafia420]

Giving away 1M $NAP 🐊

Drop your $SOL wallets, follow & RT 🎁

~ 24 hours⏳

[Henry Chen @________chh]

@simonw @flutterrausch @halomancer1 I made a cli to generate commit message using ‘git def’, the LLM backend is 7B @MistralAI instruct v0.2. It does hallucinate but most of time produces quality commit messages (better then mine)

https://t.co/jioMagDwW7

[Eric Hartford @QuixiAI]

I released dolphin-2.5-mixtral-8x7b

https://t.co/PhF65JYdlJ

deleted samantha, wizardlm
added dolphin-coder, @Magicoder_AI magicoder, @migtissera Synthia, @teknium openhermes, @ldjconfirmed Pure-Dove
still has @jon_durbin airoboros

Uncensored you're responsible for any content you generate

[Migel Tissera @migtissera]

Releasing two updated Tess models (Tess-7B & Tess-34B)-v1.4 today.

V1.4 was trained to have improved coding capabilities, Tree-of-Thought, and another 10 new system prompts that I've designed to elicit working memory, reasoning and logic, philosophical thinking and Chain-of-Thought reasoning.

Tess-7B (v1.4): https://t.co/zvhiTANGY5

Tess-34B (v1.4): https://t.co/4IXsSPHmot

Enjoy!

[Binyuan Hui @huybery]

🔥 Thanks to all the enthusiasm, let's add some fuel to the fire! We've further open-sourced 🎤 Qwen-Audio, including Base and Chat, as well as the demo!

🤗 Base: https://t.co/c01Pfpjx9E
🤗 Chat: https://t.co/wGJYTdj7AM
🤗 Demo: https://t.co/hmuxEdbuYg

[Eric Hartford @QuixiAI]

I just released dolphin-2.1-mistral-7b - a model inspired by Microsoft's Orca paper, designed to be uncensored.
https://t.co/cZf1mXonu2
Enhanced with @jon_durbin's excellent airoboros.
Currently topping the 7b leaderboard.
Uses ChatML prompt template.
Sponsored by @a16z

[Haotian Liu @imhaotian]

We've just released all training data/scripts & evaluation scripts for LLaVA-1.5!

✅Training: https://t.co/MNRG99qMjV
✅Evaluation: https://t.co/jvremNHZNo

You can also try LLaVA v1.5 now on 🤗Space: https://t.co/XYa58lWLTi.

[Anil Chandra Naidu Matcha @matchaman11]

OpenAI embeddings have a new open-source competitor 🔥🔥

Introducing gte-tiny

Results comparable to OpenAI as can be seen in screenshot below

Checkpoint size of just 20MB

Link below 👇

[Aaron Iker ✨ @aaroniker_me]

New Browser Viewport Units in CSS ✨
Easily get the correct viewport size with svh, lvh and dvh (instead of vh) especially if the browser UI is resizing for example 👇
📄 Docs (incl. other units) https://t.co/WyZPanBrpF
✅ Browser Support https://t.co/74s67NmGkh
🔗 Article by @shadeed9 https://t.co/UexFIDCBo7
🔗 Summary by @chriscoyier https://t.co/UIbxsEUzjc

Nowadays it's widely supported, but you can still fallback to vh for example which was introduced in 2015 ☝️ so just set: min-height: 100vh; min-height: 100dvh;

[Henry Chen @________chh]

@haider1 I’m running codellama https://t.co/CYRANMkFOT