berkayturanci/ai-jury — reverse-engineered prompt

Build me a Python 3.11 command line tool called ai jury that can take a git diff, GitHub PR, or issue text and have several coding agents review the same thing together. I want it to use the agents people already have installed, like Claude Code, Codex, and Antigravity, plus an optional free local model through Ollama or another OpenAI compatible local server. Each reviewer should first look independently, then react to the others, then a chair should verify the claims and produce one final verdict, either approve, comment, or request changes, along with a readable report.

Please make it local first and secure by default, with reviewers running read only and sandboxed where possible. I want a simple setup command like jury init that detects available agents and writes a config file. It should support live output, full transcripts, panel vote or chair synthesis, suggested patches, large diff chunking, incremental review, and CI friendly exit codes. If PR posting is supported, use GitHub CLI. Look up current CLI docs online if you need to.