New Benchmark Evaluates Coding Agent Instruction Following Accuracy

Zining Huang, Haoran Que, Hong Zeng, Ge Zhang, Zuo Wang, Jin Chen, Haodong Wang, Zhongfei Hou, Changxin Pu, Shen Yan, Wenhao Huang· August 13, 2026 View original

Key takeaways

  • Current AI coding agent benchmarks may overstate instruction compliance by not distinguishing true adherence from default behavior.
  • Harness-IF introduces Against-Prior Accuracy (AP-Acc) to specifically measure compliance when instructions oppose an agent's natural tendencies.
  • Models perform significantly worse on "against-prior" rules, indicating a gap in reliable instruction following.
  • The placement of instructions (e.g., system prompts vs. user instructions) can influence an agent's adherence.

Who benefits

Software DevelopmentAI/ML EngineeringCybersecurityAutomotiveAerospace

Summary

Researchers introduce Harness-IF, a new benchmark to accurately measure how well coding agents follow instructions, distinguishing genuine compliance from coincidental behavior. It evaluates rule adherence across various instruction surfaces and uses a novel metric, Against-Prior Accuracy, to assess performance when rules contradict default actions.

A new research paper introduces Harness-IF, a novel benchmark designed to rigorously evaluate the instruction-following capabilities of AI coding agents. Unlike previous methods that often conflate an agent's inherent tendencies with its actual compliance to instructions, Harness-IF focuses on isolating true adherence. It achieves this by testing agents against 60 realistic multi-turn coding scenarios, incorporating 256 specific rules placed across five different configurable instruction surfaces that an agent might encounter. The benchmark introduces a crucial metric called Against-Prior Accuracy (AP-Acc). This metric specifically scores rules that are designed to oppose an agent's unprompted default behaviors. By re-running tasks with and without specific rules, researchers can determine if an agent is genuinely following an instruction or merely performing an action it would have taken anyway. Initial tests on 12 frontier models revealed that overall accuracy ranged from 72.1-85.9%, but AP-Acc scores were significantly lower (66.1-78.6%), indicating that models overstate compliance by an average of 5.81 points. This suggests that current aggregate scores might be misleading, and a deeper understanding of how agents process and prioritize instructions from different sources (system prompts, project files, user instructions, tool/skill descriptions) is needed.

Why it matters

Professionals developing or deploying AI coding agents need accurate metrics to understand their true reliability and ensure they follow specific directives, especially when safety or precise execution is critical. This research highlights limitations in current evaluation methods and offers a more robust approach.

How to implement this in your domain

  1. 1Review current AI agent evaluation protocols to incorporate "against-prior" testing methodologies.
  2. 2Prioritize explicit instruction placement in agent prompts, considering the varying impact of different instruction surfaces.
  3. 3Develop internal benchmarks that specifically test for compliance with rules that contradict an agent's default behavior.
  4. 4Train and fine-tune coding agents with datasets that emphasize nuanced instruction following and conflict resolution.

Original post by Zining Huang, Haoran Que, Hong Zeng, Ge Zhang, Zuo Wang, Jin Chen, Haodong Wang, Zhongfei Hou, Changxin Pu, Shen Yan, Wenhao Huang

"arXiv:2608.11727v1 Announce Type: new Abstract: When a coding agent obeys a rule, it may simply have been going to do that anyway. Existing instruction-following benchmarks cannot tell the difference: they concentrate rules in the user turn, while coding-agent benchmarks emphasiz…"

View on X

Originally posted by Zining Huang, Haoran Que, Hong Zeng, Ge Zhang, Zuo Wang, Jin Chen, Haodong Wang, Zhongfei Hou, Changxin Pu, Shen Yan, Wenhao Huang on X · view source

Want to go deeper?

Turn these trends into skills with Learnijoy's hands-on AI & tech courses.

Explore courses

More in AI Research