fried-catfish
NewAI Framework with multi-agent tournament, deterministic Bradley-Terry scorer, human-gated writes — portable across Claude Code and Codex from a single MCP core.
Summary
Fried Catfish is an AI framework that orchestrates multi-agent tournaments using a deterministic Bradley-Terry scorer, enabling robust evaluation and ranking of AI agents.
- It features human-gated writes for safe, controlled execution and is portable across Claude Code and Codex from a single MCP core, making it ideal for testing and deploying multi-agent systems.
Install & Usage
mkdir -p .claude/agentsAdd the configuration to .claude/agents/fried-catfish.md
@fried-catfishUse Cases
Usage Examples
/fried-catfish tournament --agents agent1,agent2,agent3 --tasks task1,task2 --scorer bradley-terry
Run a multi-agent tournament with human-gated writes to compare code generation agents.
/fried-catfish rank --results results.json --output leaderboard.json
Security Audits
Frequently Asked Questions
What is fried-catfish?
Fried Catfish is an AI framework that orchestrates multi-agent tournaments using a deterministic Bradley-Terry scorer, enabling robust evaluation and ranking of AI agents. It features human-gated writes for safe, controlled execution and is portable across Claude Code and Codex from a single MCP core, making it ideal for testing and deploying multi-agent systems.
How to install fried-catfish?
To install fried-catfish: create the agents directory (mkdir -p .claude/agents), then add the config to .claude/agents/fried-catfish.md. Finally, @fried-catfish in Claude Code.
What is fried-catfish best for?
fried-catfish is a agent categorized under General. It is designed for: mcp, agent. Created by JamalMansuri.
What can I use fried-catfish for?
fried-catfish is useful for: Run a tournament to compare the performance of multiple AI agents on a set of benchmark tasks.; Use the Bradley-Terry scorer to generate a ranked leaderboard of agents based on pairwise comparisons.; Implement human-gated writes to review and approve agent-generated code or actions before execution.; Port an existing multi-agent workflow seamlessly between Claude Code and Codex environments.; Evaluate new agent architectures by pitting them against baseline agents in a controlled tournament.; Automate the selection of the best-performing agent for a specific task using deterministic scoring..