Minimal AI Agent hip-agent Outperforms Codex CLI on Coding Benchmark With Just Two Tools

A minimalist AI agent called hip-agent, built with just two tools and a self-readable core loop, outperforms OpenAI's Codex CLI on a 113-task coding benchmark, resolving 64.6% of DeepSWE challenges versus Codex CLI's 63.7%, proving that radical transparency in agent design can rival proprietary systems.

• hip-agent is a minimal AI agent harness where the core loop fits directly in the prompt, giving models only two tools and letting them read the source code to understand exactly how the harness works.
• The system uses the OS as its runtime, with configuration handled via environment variables, actions executed as shell commands, and subagents spawned as child processes, while leveraging existing protocols for plugins, hooks, and session management.
• Benchmarked against Codex CLI on 113 DeepSWE tasks, hip-agent resolves 64.6% compared to Codex CLI's 63.7%, demonstrating that a transparent, model-readable reference harness can match proprietary tools while enabling better model evaluation and adaptability.