github.comrepomulti-model-redteam
A tutorial repo showing how to run the same design plan past three coding-agent CLIs (Claude Code, Codex CLI, Gemini CLI) in parallel and consolidate/rank their findings, on the premise that each model has different blind spots. Ships a CLAUDE.md snippet for zero-install use inside Claude Code, plus a standalone redteam.sh script and a worked BigQuery-pipeline case showing consensus, unique, and triple-blind-spot findings. Explicitly scoped to design/plan review, not jailbreak or prompt-injection red-teaming (points to garak/promptfoo for that).
TL;DR Chapter-based tutorial + script for running Claude Code, Codex CLI, and Gemini CLI in parallel to red-team engineering plans before code is written.
Organisations named here
- Claude Code
Technologies named here
- Claude
- Claude Code
- Codex
- Codex CLI
- Gemini
- Gemini CLI
- redteam.sh
- garak
- promptfoo
- claude-octopus
- cerberus
- agent-council
- Bash
Techniques named here
- 5-failure-dimension red team frame
Related in coding-agents
