Key Signals

• OpenAI Codex v0.112.0 introduces @plugin mentions, bringing first-class plugin context into chat sessions. Users can now reference installed plugins directly with an @ mention and automatically pull in the plugin's associated MCP servers, apps, and skills as context for the current conversation. This reduces the friction of manually configuring plugin context and signals Codex's move toward a richer, more composable plugin ecosystem. The release also merges executable permission profiles into per-turn sandbox policies for zsh-fork skill execution, tightening security while keeping privilege handling additive. [1]

• Snowflake Cortex Code CLI expands beyond Snowflake with dbt and Apache Airflow support, plus a new open subscription model. This is the first time Snowflake's AI coding agent has been applied to workloads outside the Snowflake ecosystem, using Agent Skills to auto-generate dbt semantic models, propagate downstream lineage changes, and automate Airflow DAG setup. The new self-service monthly subscription removes the Snowflake-customer requirement entirely, positioning Cortex Code as a general-purpose data pipeline agent rather than a vendor-locked tool. [2]

• Vibe coding continues to shape framework design, with TanStack Start positioned as a type-safe, minimal-boilerplate target for AI-assisted full-stack development. TanStack Start's file-based routing, built-in server functions, and streaming SSR minimize the architectural decisions that typically trip up AI-generated code, allowing developers to move from prompt to working prototype with fewer configuration steps than Next.js or Remix. This trend of frameworks explicitly optimizing for AI-assisted workflows reflects a maturing vibe-coding ecosystem where the toolchain is adapting to the agent, not just the human. [3]

AI Coding News

• TanStack Start is emerging as a vibe-coding-friendly full-stack framework with type safety, file-based routing, and streaming SSR baked in. Unlike Next.js or Remix, TanStack Start emphasizes minimal boilerplate and fully typed routes and server functions, which reduces runtime errors during rapid AI-assisted prototyping. Adding a new page is as simple as creating a file in the routes folder — the framework auto-generates all routing boilerplate. The combination of TypeScript-first design and lightweight architecture makes it particularly well-suited for workflows where developers prompt an AI to scaffold features and iterate in real time. [3]

Feature Update

• OpenAI Codex v0.112.0 adds @plugin mentions, updated model selection, and sandbox security hardening. The headline feature lets users type @plugin-name in chat to auto-include that plugin's MCP, app, and skill context in the conversation, streamlining multi-tool workflows. A new model-selection surface update ensures the latest model catalog changes are surfaced in the TUI picker. On the security side, executable permission profiles are now merged into per-turn sandbox policies for zsh-fork skill execution, and the Linux bubblewrap sandbox always unshares the user namespace for consistent isolation even under root. Additional fixes include JS REPL state persistence after failed cells, graceful SIGTERM handling for websocket shutdown, and hardened emitImage to accept only data: URLs. [1]

• OpenCode v1.2.22 fixes MCP toggling, prevents TTY corruption, and improves desktop session management. The core now canonicalizes the working directory after cd commands in the TUI, and a community-contributed fix restores broken /mcp toggling. A new OPENCODE_SKIP_MIGRATIONS flag lets users bypass database migrations, and beta/stable channels can now share the same database. TUI stability improvements prevent premature exit and TTY corruption from double cleanup. On the desktop side, the release addresses session cache management, sidebar styling, terminal buffer trimming, and review panel transitions — with three community contributors participating. [4]

• Snowflake Cortex Code CLI adds dbt and Apache Airflow Agent Skills for AI-powered data pipeline automation. This marks the first time the coding agent has operated on workloads outside Snowflake. The dbt Agent Skills can auto-generate semantic models from well-structured tables in minutes and propagate column-level changes across downstream lineage automatically. For Airflow, the agent automates DAG setup with tags for ingestion, cleaning, extraction, aggregation, and loading. Snowflake also introduced a self-service monthly subscription that opens the agent to non-Snowflake customers, signaling ambitions to become a general-purpose data engineering agent. [2]