A collection of Claude Code skills that implement an adversarial multi-model workflow for software development. It uses Claude Fable 5 to grill the user on their plan, which is then stress-tested by OpenAI Codex (GPT-5.6 Sol) in a cross-provider review loop. This 'Nobody grades their own work' pattern ensures plans are hardened and code is verified by a rival model before being committed.
Details
- Built with
- Fable 5NewGPT-5.6 SolNewStrong
The project explicitly names the models in its documentation and the referring source confirms the specific versions used.
Video title mentions 'Fable 5 + GPT 5.6 Sol'; README specifies 'gpt-5.5' as the default Codex model.
- Posted by
- Attributed to
- We haven't verified this person made this build — attribution is our best guess from public sources.
- Discovered
- First posted
- Published on YouTube Jul 8, 2026
- Listed
- Added to Dropday 19h ago
- Evidence
- Strong
Publicly available GitHub repository with complete source code, installation instructions, and active community engagement.