Tianyi Cui

• EmbodiedLab: Benchmarking Embodied Agents That Act to KnowVerdict: Useful

  Oct 26, 2026 · DeepSeek engineer associated with DeepSeek Harness, one of the evaluated harnesses.

• HarnessMM: Recursive Self-Improvement for Harness Token EfficiencyVerdict: Weak

  Oct 23, 2026 · DeepSeek engineer on DeepSeek Harness, whose runtime the paper modifies.

• Trial by Trace: Benchmarking AI Agent Harnesses for Reliability and Failure AttributionVerdict: Important

  Oct 23, 2026 · DeepSeek engineer on DeepSeek Harness, one of the benchmarked harnesses.

1. github.com/tianyicui
2. github.com/deepseek-ai/deepseek-harness

Verified Oct 29, 2026