The opencode-fusion skill provides a local OpenRouter Fusion-style workflow using OpenCode CLI to run multiple LLMs in parallel, analyze their agreement and conflicts, and output a single synthesized response.

It orchestrates three phases—parallel model execution, a judge layer producing a structured consensus/contradiction/risk report, and a synthesizer that combines this with the original task and raw outputs. Configuration is done via a bash script (run_fusion.sh) that defines the model panel, manages parallel runs, checks CLI/server prerequisites, and automates judging and synthesis without external orchestration.