arxiv arXiv cs.CL · 1d ago · research

ExecCritic uses role-specific RL to improve coding agents via test-guided repair

from English

Researchers introduce ExecCritic, a framework that combines a test-verify-revise scaffold with role-specific reinforcement learning to enhance coding agents. The system separates test construction from source-code repair, using a Test agent to generate repository-native tests and a Repair agent to revise code based on execution feedback.

The architecture uses Qwen-3.5-35B-A3B as the backbone for both roles, trained separately.
The Test agent learns to produce behaviorally valid tests that distinguish correct from incorrect patches.
The Repair agent learns direct task resolution and feedback-guided revision.
On SWE-bench Verified, tests from a post-trained Qwen Test agent raise the resolved rate to 72.6%, an 11.4-point gain over the no-test baseline.

The authors consider this important because test quality determines whether execution feedback helps, and ExecCritic achieves significant gains without requiring stronger models or Oracle feedback at evaluation time.

Importance 2/3 New harness with differentiators arXiv cs.CL Microsoft Research Alibaba (Qwen) AI agents Training methods

Benchmarks

Benchmark	Model	Score
SWE-bench Verified	Qwen Test agent (post-trained) + Repair agent (post-trained)	72.6%
SWE-bench Verified	GPT-5.6-sol (Test agent)	65.3%
SWE-bench Verified	Qwen Test agent (post-trained)	62.2%
SWE-bench Verified	base Repair agent (no-test baseline)	61.2%
SWE-bench Verified	base Test agent	57.3%

Read original

arxiv arXiv cs.AI · 84d ago · 1 view

Spotlight: Using Spot GPUs to Accelerate DiT RL Post-Training

Spotlight enables DiT RL post-training by leveraging idle spot GPUs, reducing costs by 1.4-6.4× while achieving superior image quality. It uses stale model weights in exploration and reconfigures sequence parallelism in real time, allowing efficient GPU utilization without breaking training pipelines.

arxiv arXiv cs.AI · 1d ago · 1 view

ExecCritic uses role-specific RL to improve coding agents via test-guided repair

ExecCritic introduces a framework that separates test construction from source-code repair to prevent false confidence in coding agents. The system employs a Test agent and a Repair agent, both backed by Qwen-3.5-35B-A3B, trained separately using reinforcement learning.

lab Microsoft Research Blog · 42d ago · 22 views

Microsoft Research introduces EvoLib for test-time learning with an evolving library

Microsoft Research has introduced EvoLib, a framework that enables large language models to learn from their own experience during inference without requiring ground-truth labels or external feedback. Rather than storing raw experiences as static memories, EvoLib transforms them into reusable skills and reflective insights that are continually refined, consolidated, and reweighted.

lab Microsoft Research Blog · 42d ago · 20 views

Microsoft releases Echoverse for training computer-use agents

Microsoft Research has released Echoverse, a collection of twelve deep synthetic environments designed to train computer-use agents. The initiative includes ten domain-specific worlds and two capability-focused worlds, all built to maintain coherent state and behavioral fidelity.

lab Microsoft Research Blog · 72d ago · 23 views

SkillOpt: Agent skills as trainable parameters

Microsoft Research introduces SkillOpt, a method that treats agent skill files as trainable parameters outside a frozen target model, transforming manual skill editing into a controlled optimization process. This approach improves agent reliability and consistency without updating the underlying model weights.