media Hugging Face Forums · just now · open_models

Claude Code and Codex collaborate in SharedNet hackathon

from English

At the Trial Zero hackathon on SharedNet, Claude Code and Codex operated as a team in a persistent room to build and test Scout, an MCP tool for verifying agent product callability. The agents used a message protocol to hand off tasks, with Codex writing tests and Claude reviewing code.

The collaboration involved Codex independently scoring five public MCP servers and Claude catching two regressions in the fixes.
Each agent found bugs in the other's work before merging changes.
In the arena round, agents from other teams bought Scout’s services with credits and found real bugs live.
The team placed 3rd in the competition.

SharedNet serves as a neutral room for multi-agent setups, allowing agents from different vendors to interact without human intervention.

Importance 1/3 Trust 1/3 Hugging Face Forums AI agents Code generation

Read original

lab Claude Code Releases · 1d ago · 29 views

Claude Code v2.1.289 fixes terminal freezes, plugin crashes, and Bash rule bypasses

Claude Code version 2.1.289 addresses several critical stability issues, including terminal freezes caused by specific code patterns and plugin rendering errors. The update also resolves security gaps where Bash deny or ask rules were being bypassed under certain sandbox conditions.

media Hugging Face Forums · 2d ago · 17 views

FieldTrace prevents data loss in JSON handoffs between AI agents

FieldTrace is a tool designed to handle common issues in agent pipelines where fields arrive under incorrect names or numbers are passed as text. It accepts declared renames and safe conversions while validating against the target schema to refuse changes that could result in information loss.

lab Claude Code Releases · 2d ago · 17 views

Claude Code v2.1.288 adds mod selection API, fixes resume and permission bugs

Anthropic released Claude Code v2.1.288, introducing new features for mods and cloud sessions while addressing numerous stability issues with session resumption, permissions, and plugin handling.

media Hugging Face Forums · 2d ago · 8 views

Kontur: self-hosted roadmap dashboard for human and AI-agent handoffs

Kontur is a self-hosted, MIT-licensed project dashboard designed to manage milestones, dependencies, and evidence references while facilitating explicit handoffs between humans and AI agents. Built with FastAPI and SQLite, it provides a plain HTTP/JSON API that allows agents to read versioned snapshots and propose changes using If-Match headers to prevent silent overwrites.

github OpenAI Agents SDK · 3d ago · 24 views

OpenAI releases openai-agents-python v0.23.0 with MCP and sandbox improvements

OpenAI has released version 0.23.0 of the openai-agents-python library, introducing new configuration options for Model Context Protocol (MCP) and sandbox environments alongside numerous core bug fixes.