{
  "$schema": "https://themachinepress.com/schemas/story-v1.schema.json",
  "schema_version": "1.0.0",
  "document_type": "machine_press_story",
  "story": {
    "story_id": "mp-2026-08-25-008",
    "source_story_id": "tmp-story-prime-agent-harness",
    "edition_id": "mp-2026-08-25-morning-0047",
    "edition_url": "https://themachinepress.com/edition/2026-08-25",
    "position": 8,
    "story_type": "dispatch",
    "section": "developer-tools",
    "editorial_classification": "editorial",
    "headline": "The Harness Stopped Turning Recovery Into a Model Failure",
    "slug": "the-harness-stopped-turning-recovery-into-a-model-failure",
    "dek": "Prime Agent keeps histories, memories, skills and subagent specifications across long-running coding and evaluation trajectories.",
    "summary": "Prime Agent keeps histories, memories, skills and subagent specifications across long-running coding and evaluation trajectories.",
    "body_text": "The open-source harness pairs a persistent IPython environment with standardized execution, recovery, verification and resource accounting. Its reported evaluations include an ARC-AGI-3 RHAE Best@1 increase from 30 to 95.5 percent and competitive results across coding, kernel and emulator tasks; those are harness-specific benchmarks, not general capability proof.",
    "why_it_matters": "Prime Agent keeps histories, memories, skills and subagent specifications across long-running coding and evaluation trajectories.",
    "limitations": [
      "Its reported evaluations include an ARC-AGI-3 RHAE Best@1 increase from 30 to 95.5 percent and competitive results across coding, kernel and emulator tasks; those are harness-specific benchmarks, not general capability proof."
    ],
    "importance": 8,
    "canonical_url": "https://themachinepress.com/story/mp-2026-08-25-008/the-harness-stopped-turning-recovery-into-a-model-failure",
    "json_url": "https://themachinepress.com/story/mp-2026-08-25-008.json",
    "first_published_at": "2026-08-25T09:00:00.000-04:00",
    "modified_at": "2026-08-25T09:00:00.000-04:00",
    "content_status": "new",
    "is_carryover": false,
    "carryover_reason": null,
    "key_claims": [
      {
        "claim_id": "claim-mp-2026-08-25-008-001",
        "text": "Prime Agent keeps histories, memories, skills and subagent specifications across long-running coding and evaluation trajectories.",
        "source_ids": [
          "source-2026-08-25-008"
        ],
        "qualification": "Its reported evaluations include an ARC-AGI-3 RHAE Best@1 increase from 30 to 95.5 percent and competitive results across coding, kernel and emulator tasks; those are harness-specific benchmarks, not general capability proof."
      }
    ],
    "source_ids": [
      "source-2026-08-25-008"
    ],
    "tags": [
      "agent harnesses",
      "coding agents",
      "persistent context"
    ],
    "image_url": "https://themachinepress.com/issues/2026-08-25/prime-agent-code-file-image.webp",
    "corrections": []
  },
  "sources": [
    {
      "source_id": "source-2026-08-25-008",
      "title": "arXiv preprint 2608.23552",
      "publisher": "arXiv",
      "url": "https://arxiv.org/abs/2608.23552",
      "canonical_url": "https://arxiv.org/abs/2608.23552",
      "source_type": "primary_research",
      "is_primary_source": true,
      "published_at": "2026-08-24T13:54:19.000-04:00",
      "accessed_at": "2026-08-25T08:17:00.000-04:00",
      "supports_claim_ids": [
        "claim-mp-2026-08-25-008-001"
      ]
    }
  ],
  "corrections": [],
  "publisher": {
    "name": "The Machine Press",
    "url": "https://themachinepress.com",
    "description": "A daily newspaper for the age of artificial intelligence."
  },
  "cite_this_report": {
    "title": "The Harness Stopped Turning Recovery Into a Model Failure",
    "publisher": "The Machine Press",
    "published_at": "2026-08-25T09:00:00.000-04:00",
    "canonical_url": "https://themachinepress.com/story/mp-2026-08-25-008/the-harness-stopped-turning-recovery-into-a-model-failure"
  }
}