Descriptions:
This All About AI tutorial shows how to extend an autonomous Claude Code agent — running on a dedicated Mac Mini — with a new skill that invokes OpenCode from the command line to run parallel LLM benchmark tests. The creator walks through the full skill-building process: fetching OpenCode’s CLI documentation, prompting Claude Code to discover the correct `opencode –run -m provider/model “prompt”` syntax, verifying it live with multiple models, then saving the working workflow as a reusable skill file.
The practical use case is automated creative benchmarking across models via OpenRouter. Claude Code spawns simultaneous OpenCode instances targeting GLM5, Minimax 2.5, Gemini 3 Pro, and Opus 4.6, each generating a full-screen animated retro arcade game as a self-contained HTML file. Outputs are saved to a structured folder with consistent naming (e.g., `game_at_glm5.html`), and a previously built Remotion skill then assembles all four outputs into a grid-style comparison video ready to post to X — entirely without human intervention.
Beyond the specific benchmark use case, the video demonstrates a repeatable pattern for agentic skill development: document the target tool, let Claude Code figure out the integration through live testing, iterate until the output matches requirements, and persist the workflow for future autonomous reuse. The setup requires only OpenRouter API credentials and runs locally with no custom server infrastructure.
📺 Source: All About AI · Published February 15, 2026
🏷️ Format: Hands On Build
