{
  "schema_version": "1.1",
  "id": "s13:https://arxiv.org/abs/2609.22068v1",
  "slug": "2609-22068v1-0mwdzky",
  "url": "https://feed7.dev/p/2609-22068v1-0mwdzky",
  "title": "CodeMidas: Scaling Agentic Coding RL Environments from Code Itself",
  "why_included": "CodeMidas derives verified coding-agent RL tasks directly from working source, widening training data beyond issues and commits while improving repair, construction, and terminal benchmarks.",
  "summary": "CodeMidas turns existing implementations into executable RL environments using source code as the only task-specific input. Its agents derive specifications, build execution-grounded tests, and filter tasks through checks and repeated solution attempts, producing **5,545 tasks from 3,185 repositories**.",
  "practical_implication": "For coding-agent training, treat working code as a source of behavioral tasks even when issue and commit histories are sparse. MiMo-V2.5 trained with GRPO improved across all five tested benchmarks, including **+17% on ProgramBench** and **+11.7% on DeepSWE**.",
  "agent_context": "CodeMidas turns existing implementations into executable RL environments using source code as the only task-specific input. Its agents derive specifications, build execution-grounded tests, and filter tasks through checks and repeated solution attempts, producing **5,545 tasks from 3,185 repositories**.\n\nFor coding-agent training, treat working code as a source of behavioral tasks even when issue and commit histories are sparse. MiMo-V2.5 trained with GRPO improved across all five tested benchmarks, including **+17% on ProgramBench** and **+11.7% on DeepSWE**.\n\nThe abstract reports that more high-quality tasks helped and that trained agents explored and self-verified more, but it does not establish how well the pipeline handles underspecified, unsafe, or hard-to-execute repositories.",
  "source": {
    "name": "arXiv",
    "url": "https://arxiv.org/abs/2609.22068v1",
    "published_at": "2026-09-18T17:55:17.000Z"
  },
  "source_class": "blog_post",
  "content_type": "Paper",
  "layer": "benchmark",
  "domains": [
    "coding"
  ],
  "topics": [
    "coding-agents",
    "agent-evals",
    "harness-engineering"
  ],
  "verification": {
    "status": "needs_review",
    "label": "Needs Review",
    "method": "unverified",
    "verified_at": null
  },
  "uncertainty": [
    "The abstract reports that more high-quality tasks helped and that trained agents explored and self-verified more, but it does not establish how well the pipeline handles underspecified, unsafe, or hard-to-execute repositories."
  ],
  "connected_context": {
    "meaning": "CodeMidas broadens executable task generation beyond repositories with usable issue or PR histories by deriving behavior from implementations themselves, and supplies training gains rather than only a data-pipeline proposal. It complements history-derived environments, while leaving task validity, unsafe repositories, execution feasibility, and supervision quality as unresolved filters rather than solved properties.",
    "corpus_size": 831,
    "generated_at": "2026-09-21T09:04:18.040Z",
    "connections": [
      {
        "title": "Change2Task: From Repository Changes to Executable Coding Agent Tasks and Environments",
        "source_name": "arXiv",
        "source_url": "https://arxiv.org/abs/2607.28591v1",
        "feed7_url": "https://feed7.dev/p/2607-28591v1-169rlqh",
        "reason": "Change2Task derives executable tasks from merged changes, whereas CodeMidas uses implementations alone, making the approaches complementary for repositories with rich versus sparse histories."
      },
      {
        "title": "MindForge: Teaching Small Language Models Whole-Life-Cycle Software Engineering via Source-Free Program Synthesis",
        "source_name": "arXiv",
        "source_url": "https://arxiv.org/abs/2607.27146v1",
        "feed7_url": "https://feed7.dev/p/2607-27146v1-0gcx6ia",
        "reason": "Both report ProgramBench gains from constructed coding environments, but MindForge reconstructs programs from documentation and behavior while CodeMidas derives tasks from existing source."
      },
      {
        "title": "SWE-Prime: Fewer Trajectories, Better Performance",
        "source_name": "arXiv",
        "source_url": "https://arxiv.org/abs/2608.27449v1",
        "feed7_url": "https://feed7.dev/p/2608-27449v1-0lkylqd",
        "reason": "SWE-Prime shows that successful trajectories still require quality filtering, an implementation consequence for CodeMidas after its task filters and repeated solution attempts produce training traces."
      },
      {
        "title": "State of Data — Sean Cai, Independent / State of Data",
        "source_name": "AI Engineer",
        "source_url": "https://www.youtube.com/watch?v=ZyIoTOAbRfs",
        "feed7_url": "https://feed7.dev/p/state-of-data-sean-cai-independent-state-of-data-0v9fy69",
        "reason": "The real-trace thesis provides a useful contrast: CodeMidas demonstrates gains from manufactured source-derived tasks, but does not establish that they capture the behavior or scaffold dependence of live work."
      }
    ]
  },
  "lifecycle": "Current",
  "published_at": "2026-09-18T17:55:17.000Z",
  "modified_at": "2026-09-18T17:55:17.000Z",
  "supersedes": [],
  "expires_at": null,
  "formats": {
    "html": "https://feed7.dev/p/2609-22068v1-0mwdzky",
    "json": "https://feed7.dev/p/2609-22068v1-0mwdzky.json",
    "markdown": "https://feed7.dev/p/2609-22068v1-0mwdzky.md"
  }
}