mlsanityData to 14 Oct 2026

Envs-FORGE: Frontier-Optimized Reward-Grounded Environment Synthesis for Agent RL

arXiv 2608.14312 ↗First posted 14 Aug 2026cs.CLcited by 4

On the map: Deep reinforcement learning methods → RL training of language and agent models

Abstract (arXiv)

Reinforcement learning (RL) for terminal agents needs executable training environments with reliable rewards and useful difficulty. Fixed recipes such as few-shot, Self-Instruct, and Evol-Instruct apply the same prompting policy to every seed, even when the current policy would benefit from a harder, easier, or simply different task. We present Envs-FORGE, a prompting policy that converts verifier rewards into per-seed environment-synthesis actions. Envs-FORGE estimates seed pass rates, scores six projection--direction actions around a target learning frontier, and solves a per-seed mixed-integer linear program (MILP) to choose the action that conditions generation. The selected action drives synchronized rewriting of the instruction, fixtures, oracle solution, tests, and Docker environment; only gold-verified bundles enter RL training. The indexed MILP form also supports optional soft skill coverage for portfolio planning. On Qwen 3.5 35B, Envs-FORGE improves Pass@1 over Base by 9.2 percentage points on tb-core (40.0% to 49.2%) and 6.4 points on tb-2.0 (23.0% to 29.4%), exceeding the strongest fixed-recipe baseline by 2.4 and 2.1 points. It reaches 77.1% on SWE-bench Verified versus 73.4% for Base, and improves tb-core by 6.8--9.2 points across the evaluated 4B--35B models. All synthesis methods export 100 verified environments and use 2.27M--2.88M synthesis tokens, placing the comparison at the same downstream training-set size and the same operational scale. The source code is available at https://github.com/DataArcTech/DataArc-SynData-Toolkit/.

Key facts. Cited by 4 papers in the map (data through 14 Oct 2026). Placed in RL training of language and agent models, one of 15 directions in Deep reinforcement learning methods. Builds on 10 of its 13 references that are in the map. First posted on arXiv 14 Aug 2026.

Builds on

10 of 13 references.

References · 10

  1. 2022Self-Instruct: Aligning Language Models with Self-Generated Instructionscited by 2,724
  2. 2023WizardLM: Empowering large pre-trained language models to follow complex instructionscited by 1,125
  3. 2026Terminal-Bench: Benchmarking Agents on Hard, Realistic Tasks in Command Line Interfacescited by 585
  4. 2024AgentInstruct: Toward Generative Teaching with Agentic Flowscited by 89
  5. 2026Endless Terminals: Scaling RL Environments for Terminal Agentscited by 54
  6. 2026Large-Scale Terminal Agentic Trajectory Generation from Dockerized Environmentscited by 33
  7. 2026TermiGen: High-Fidelity Environment and Robust Trajectory Synthesis for Terminal Agentscited by 29
  8. 2026CLI-Gym: Scalable CLI Task Generation via Agentic Environment Inversioncited by 23
  9. 2026Toward Scalable Terminal Task Synthesis via Skill Graphscited by 22
  10. 2024Targeted synthetic data generation for tabular data via hardness characterizationcited by 4

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2026: 4.

4 papers.

  1. 2026FrogNano: Training a 4B Coding Agent via Online Task Synthesiscited by 4
  2. 2026V-Gym: Enhancing Agentic Visual Reasoning via Skill-Data Co-Evolutioncited by 0
  3. 2026AutoEnvScaling: Automating the Data Flywheel with Terminal Agentscited by 0
  4. 2026WorkRover: Scaling Long-Horizon Cowork Agent Training Via Recursive Environment Evolutioncited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2026EnvCraft: Synthesizing Executable Environments in Agentic RL for Claw-like Agent
  2. 2026RLAnything: Forge Environment, Policy, and Reward Model in Completely Dynamic RL System
  3. 2025AutoForge: Automated Environment Synthesis for Agentic Reinforcement Learning
  4. 2026HiRE: Hindsight Reward Editing for Policy Finetuning ↗
  5. 2026Controllable and Verifiable Tool-Use Data Synthesis for Agentic Reinforcement Learning
  6. 2026Terminal-World: Scaling Terminal-Agent Environments via Agent Skills
  7. 2026T1: Terminal Agent Reinforcement Learning for Long-Horizon Tasks
  8. 2026EnvRL: Learn from Environment Dynamics in Agentic Reinforcement Learning