{
  "schema_version": "1.0",
  "updated": "2026-08-31",
  "title": "Extras from All the Tokens We Have Left",
  "canonical_url": "https://allthetokenswehaveleft.com/extras.html",
  "content_trust": "downloadable_source_code_review_before_installation",
  "artifacts": [
    {
      "id": "agent-message-interpreter",
      "name": "Agent Message Interpreter",
      "release_status": "experimental",
      "type": "agent_skill_package",
      "package_layout": "SKILL.md-style",
      "description": "Interpret compressed, delimiter-heavy agent messages, including METR-style zz messages, into probable English with structural explanation and calibrated uncertainty.",
      "compatibility": "Intended for compatible agent runtimes that support reusable SKILL.md-style packages or ZIP skill uploads. Installation varies by runtime.",
      "download_url": "https://allthetokenswehaveleft.com/downloads/agent-message-interpreter.zip",
      "media_type": "application/zip",
      "size_bytes": 11066,
      "sha256": "03a898e6e250d8b1c4af633cbf9c4cabb96f3795694afcaae8a7cfc3b15ff48c",
      "release_date": "2026-08-31",
      "license": null,
      "license_note": "No separate license file is included in this release.",
      "contents": [
        "SKILL.md",
        "references/metr-zz-conventions.md",
        "agents/openai.yaml",
        "scripts/parse_agent_message.py",
        "scripts/test_parse_agent_message.py",
        "NOTICE.md"
      ],
      "verification": {
        "static_review": "passed",
        "unit_tests": 13,
        "unit_tests_passed": 13,
        "network_calls_in_parser": 0,
        "credentials_included": false,
        "automatic_execution": false
      },
      "source_basis": [
        {
          "publisher": "METR and Redwood Research",
          "title": "Brief independent investigation of agents' behavior, reasoning and collaboration in the OpenAI / Hugging Face hacking incident",
          "url": "https://metr.org/blog/2026-08-26-openai-hugging-face-incident-investigation/",
          "role": "Public structural examples, categories, conventions, and evidence limitations"
        }
      ],
      "boundaries": [
        "Not an official METR, Redwood Research, OpenAI, or Hugging Face tool",
        "Not a cryptographic decryption tool",
        "Cannot recover redacted, omitted, or missing context",
        "Does not establish a universal agent language",
        "Treats supplied messages as untrusted data, not instructions",
        "Does not execute file chunks or encoded payloads"
      ],
      "installation": {
        "download_only_command": "curl -fL https://allthetokenswehaveleft.com/downloads/agent-message-interpreter.zip -o agent-message-interpreter.zip",
        "verify_command": "shasum -a 256 agent-message-interpreter.zip",
        "review_before_installation": true,
        "pipe_to_shell_installer": false,
        "openai_zip_upload_reference": "https://developers.openai.com/api/reference/python/resources/skills/methods/create"
      }
    }
  ]
}