arXiv preprint arXiv:2609.20804, 2026.
An Empirical Study of Harness Design for Coding Agents
An empirical study of how planning, tool interfaces, and context management shape the accuracy–cost trade-offs of long-horizon coding agents.
Research output
For a complete and current record, please see my Google Scholar profile.
arXiv preprint arXiv:2609.20804, 2026.
An empirical study of how planning, tool interfaces, and context management shape the accuracy–cost trade-offs of long-horizon coding agents.
Research preprint, 2026.
A live benchmark and runtime auditing framework for evaluating what agents access and disclose while completing everyday tasks.
NeurIPS 2026 · Evaluations & Datasets Track · Accepted.
The same image, different contexts: evaluating when a multimodal agent should share it.
Findings of EMNLP 2025.
Evaluating and mitigating inappropriate information sharing in realistic agent workflows.
ECAI 2025 · Oral presentation.