Study Finds Instruction Files for AI Agents Show Mixed Results in Pull Request Success
Researchers analyzed 15,549 AI-generated pull requests across 148 projects to understand how instruction files affect agent performance, finding that while some projects improved their merge rates by 20% or more, others saw declines. Instruction files are guidelines developers create to help AI agents like GitHub Copilot navigate projects and follow best practices. The findings suggest that instruction quality and structure matter significantly, with successful projects using longer, well-organized instruction files.
A new study from arXiv examined the relationship between instruction files—developer-created guides for AI agents—and the success of AI-generated pull requests in software projects. Using data from 15,549 agentic pull requests across 148 projects in the AIDev dataset, researchers compared project performance before and after instruction files were created, measuring outcomes like merge rates, code complexity, and effort required to merge changes. The results were nuanced: while 27.7% of projects saw merge rate increases of at least 20%, 26.35% experienced decreases, and similar mixed patterns appeared for code churn and merge effort metrics. Projects that successfully improved their merge rates tended to have substantially longer instruction files with well-defined structure and multiple sections. The researchers conclude that instruction file development should be treated as a formal software engineering practice, proposing the concept of "Instructions-as-Code" to guide future work in this emerging area.
What's missing
The study does not appear to address what specific characteristics or content within instruction files correlate with success, beyond length and structural organization. Additionally, the research does not discuss potential confounding variables (such as project maturity, team experience, or agent model versions) that might influence the relationship between instruction files and PR outcomes.
What different sources said
- arXiv cs.AICenter
Toward Instructions-as-Code: Understanding the Impact of Instruction Files on Agentic Pull Requests
Related
Topology-Aware Thermodynamics Improves DNA Probe Specificity Design
Researchers developed a new framework for designing DNA probes that accounts for the spatial organization of matched sequences, not just overall thermodynamic stability. Traditional methods rely on scalar measures like melting temperature and free energy, which miss how mismatches are distributed along the probe. The approach could improve diagnostic accuracy in applications like HPV detection and gene expression profiling.
Study Identifies Optimal Thermal Dose for Combining Focused Ultrasound with Immunotherapy in Tumors
Researchers used multimodal PET imaging to identify an optimal thermal dose range for focused ultrasound ablation that destroys tumor tissue while preserving conditions for immunotherapy delivery. The study found that excessive heating collapses blood vessels needed for antibody access, while insufficient heating fails to adequately reduce tumor burden. The findings could guide clinical design of combination treatments pairing thermal ablation with immunotherapies.
Plant MSH1 Protein Functions as Mismatch-Directed Nuclease for Organelle Genome Maintenance
Researchers have identified the precise mechanism by which the AtMSH1 protein in Arabidopsis plants recognizes and cleaves DNA mismatches and lesions, preventing mutations in organellar genomes. The protein combines a DNA mismatch recognition module with a nuclease domain that makes staggered cuts at specific positions relative to DNA damage. This discovery explains how plants maintain unusually low mutation rates in their mitochondrial and chloroplast DNA compared to other eukaryotes.