Key Signals

• GitHub launches Agent HQ with Claude and Codex integration, marking a major shift toward multi-agent developer workflows. GitHub now allows Copilot Pro+ and Enterprise subscribers to choose between GitHub Copilot, Anthropic's Claude Code, and OpenAI's Codex as coding agents from a unified dashboard. This represents a strategic pivot where GitHub positions itself as the orchestration layer for competing AI coding agents rather than forcing exclusivity. Developers can assign the same task to multiple agents to compare their reasoning and solutions. [1][2]

• GitHub Copilot in VS Code v1.109 introduces comprehensive multiagent session management and Claude integration. The January release brings agent-driven workflows that allow developers to delegate work across Copilot, Claude, local agents, background agents, and cloud agents with seamless handoff. New features include Copilot Memory for context retention across interactions, MCP app integration for tool-driven experiences, and experimental terminal command sandboxing for security. [3]

• Vercel releases Skills.sh, creating an open ecosystem for standardized AI agent commands. The new open-source tool provides AI agents with a standardized way to execute reusable shell-based commands, separating agent reasoning from execution. Community response has been strong, with developers comparing it to "npm for AI agents" and noting it solves agent discoverability where MCP solved tool communication. [4]

• Claude Code v2.1.31 improves tool guidance to reduce unnecessary bash command usage. The release includes improved system prompts that guide the model toward using dedicated tools instead of bash equivalents like cat, sed, grep, and find. This optimization should result in more predictable and efficient agentic coding sessions. [5]

• OpenCode releases v1.1.50 and v1.1.51 with enhanced skills system and agent customization. The open-source AI coding tool adds support for reading skills from .agents/skills directories, theme colors in agent customization, and a new --thinking flag to show reasoning blocks in the TUI. These updates strengthen OpenCode's position as a Claude Code/Copilot alternative. [6][7]

AI Coding News

• Vercel introduces Skills.sh, an open ecosystem providing AI agents with a standardized way to execute reusable command-line actions. The project introduces an open agent skills ecosystem where developers can define, share, and run discrete operations that agents can invoke as part of their workflows. At a technical level, Skills.sh acts as a lightweight runtime that allows agents to call skills implemented as shell-based commands, with each skill following a simple contract defining inputs, outputs, and execution behavior. The ecosystem is positioned as community-driven, with developers able to publish and reuse skills. Community response has been enthusiastic, with AI Engineer Aakash Harish noting: "This is npm for AI agents. Skills prioritizes composability over protocol complexity. MCP solved 'how do agents talk to tools' but Skills solves 'how do devs share and discover agent capabilities.'" [4]

• GitHub launches Agent HQ, letting developers choose between Copilot and its biggest rivals as their coding agents. According to GitHub COO Kyle Daigle, the goal is to provide "a single pane of glass — a mission control interface, where I can see all the tasks, what they're doing, what state of code they're in." Developers can assign tasks from GitHub.com, GitHub Mobile, and VS Code, with agents working asynchronously on issues and pull requests. Each session consumes one premium request ($0.04), with Pro+ users receiving 1,500 requests and Enterprise users receiving 1,000 requests included. [2]

Feature Update

• Claude and Codex are now available in public preview on GitHub, with Copilot offering them as coding agents for Pro+ and Enterprise customers. Developers can start agent sessions from GitHub.com, GitHub Mobile, and VS Code directly from issues, pull requests, or the Agents tab in enabled repositories. Key capabilities include assigning issues to Claude, Codex, or Copilot using the Assignees dropdown, with agents automatically submitting draft pull requests for review. Agents can also respond to PR review comments via @copilot, @claude, or @codex mentions. No additional subscription is required—access is included with existing Copilot subscriptions, with each session consuming one premium request during public preview. [1]

• GitHub Copilot in VS Code v1.109 brings significant agent-driven workflow improvements including Claude integration in the January release. Claude agent support is now in public preview, enabling task delegation using Claude models from your GitHub Copilot subscription. The multiagent session management system supports Copilot, Claude, local, background, and cloud agents with seamless handoff and an agent status indicator showing sessions needing attention. New Copilot Memory helps agents retain relevant context across interactions, while MCP app integration enables richer tool-driven experiences. Security improvements include experimental terminal command sandboxing, effective auto-approval rules, and terminal lifecycle improvements with timeout, await, and kill capabilities. [3]

• GitHub Copilot Chat on the web now shows tool calls and other improvements, displaying real-time details and adding conversation export. Users can see the steps Copilot takes, course-correct if something fails, and see references generated by each tool call. A new export feature allows downloading chat conversations as JSON or Markdown files for future reference. Repository attachment search has also been improved with better recommendations when searching for repositories to attach to conversations. [8]

• GitHub Copilot in Visual Studio's January update introduces colorized code completions and click-to-partially-accept suggestions. A new click-to-partially-accept feature allows developers to accept only part of a suggestion by clicking directly inside it, with hover highlighting showing each segment. Additional productivity improvements include faster scrolling, HTML-rich copy/paste, and syntactic line compression. Markdown preview controls have been streamlined with Split Preview, Open Preview, and Edit Markdown modes, and Copilot Chat now includes a Preview button to render and edit generated markdown content directly. [9]

• Claude Code v2.1.31 adds session resume hints and fixes critical PDF and sandbox errors. Critical bug fixes include resolving PDF errors that permanently locked up sessions, fixing bash commands incorrectly reporting "Read-only file system" errors in sandbox mode, and fixing crashes after entering plan mode with incomplete project config. The system prompts have been improved to more clearly guide the model toward using dedicated tools instead of bash equivalents like cat, sed, grep, and find, reducing unnecessary bash command usage. Additional improvements include better PDF/request size error messages showing actual limits (100 pages, 20MB) and reduced terminal layout jitter during streaming. [5]

• OpenCode v1.1.50 adds Trinity model support, shell.env hooks, and an enhanced skills system. The skills system has been improved with support for reading skills from .agents/skills directories, better prompting, and fixes for permission requests after skill invocation. The TUI adds a --thinking flag to show reasoning blocks in the run command. Desktop improvements include session tab restoration on app restart, command palette search integration, and numerous UI fixes. MCP tools are now sanitized for security. [6]

• OpenCode v1.1.51 fixes header double-merging and adds built-in agent documentation. The release also prevents double-prefixing of Bedrock cross-region inference models. The TUI restores direct OSC52 clipboard support, and the desktop app includes mobile UI improvements. Four community contributors participated in this release. [7]