AI

Harness Engineering for Software Engineering via Modular Executable Dev-Primitives

Read full article at source ↗

Based on a feed excerpt; full article content has not been acquired. Source evidence retrieved 07/10/2026, 02:18:25 UTC.

Briefings

This article presents Dev-Primitives, a modular framework that converts repository components into active agents for software engineering. The proposed HERMES harness engineering system achieves a 12.4% average improvement over baseline harnesses on four benchmarks. Experiments show HERMES with Qwen3-8B Dev-Primitives stays within 4.5% of the homogeneous GPT-5.6 Sol configuration while reducing inference costs by 26.2% on Terminal-Bench 4.0.

Key points

• Dev-Primitives transforms repository artifacts into active agents by pairing them with resident LLMs that provide an agent-native interface.
• HERMES framework instantiates Dev-Primitives at repository scale using a dependency-aware dynamic activation mechanism and a bug diagnosis mechanism.

Context

Large language models with terminal access can automate software engineering tasks but struggle with long-horizon workflows due to brittle state reconstruction and context explosion. This article introduces Dev-Primitives and HERMES to solve these challenges by enabling natural-language reasoning and localized self-modification of repository components.

Limitations

Experiments were conducted on four software engineering benchmarks. The 12.4% improvement is an average across these benchmarks, and the 4.5% margin against GPT-5.6 Sol represents the best-case scenario with strong activation and diagnosis models. Inference cost reduction of 26.2% applies specifically to Terminal-Bench 4.0.

Generated by Bittrees-hosted qwen35-2b sha256:aaf42c8b7c3c on 24/10/2026, 20:21:46 UTC. Based on publisher-provided feed evidence; may not cover the full original article. Check the source for details.

View archived briefing on IPFS
bafybeicqujsrsbbg7ne5j5og7tvruqgowreoy67jws36qneiydp3ano4fm