{
  "added_at": "2026-10-17",
  "additional_evidence_urls": [
    "https://github.com/aicodeking/leanrea/blob/09ba4aa0217f82cd880f855a60b877f59f39b22b/leanrea/SKILL.md",
    "https://github.com/aicodeking/leanrea/blob/09ba4aa0217f82cd880f855a60b877f59f39b22b/bench/BENCHMARK.md",
    "https://github.com/aicodeking/leanrea/blob/09ba4aa0217f82cd880f855a60b877f59f39b22b/bench/results/final.jsonl",
    "https://github.com/aicodeking/leanrea/blob/09ba4aa0217f82cd880f855a60b877f59f39b22b/bench/tasks.json",
    "https://github.com/aicodeking/leanrea/blob/09ba4aa0217f82cd880f855a60b877f59f39b22b/bench/run.py",
    "https://github.com/aicodeking/leanrea/blob/09ba4aa0217f82cd880f855a60b877f59f39b22b/bench/setup.sh",
    "https://github.com/aicodeking/leanrea/blob/09ba4aa0217f82cd880f855a60b877f59f39b22b/bench/ctx.sh",
    "https://github.com/aicodeking/leanrea/blob/09ba4aa0217f82cd880f855a60b877f59f39b22b/LICENSE",
    "https://github.com/morluto/rea/blob/808339f0b1cf3837445f392b5d5f1f362d64a27d/README.md",
    "https://github.com/morluto/rea/blob/808339f0b1cf3837445f392b5d5f1f362d64a27d/.agents/skills/reverse-engineer-anything/SKILL.md"
  ],
  "category": "agent-workflows",
  "description_en": "An independent agent skill that routes application and binary investigations through ordinary command-line tools. It supplies batched triage recipes, targeted source/disassembly reading and explicit seen/inferred/unknown evidence labels, accompanied by a small author-reported comparison against an REA MCP workflow.",
  "description_zh": "独立智能体技能，使用普通命令行工具开展应用与二进制调查，提供批量初查、针对性源码／反汇编阅读，以及明确的已观察／推断／未知证据标记，并附有作者报告的 REA MCP 工作流小规模对照。",
  "entry_url": "https://awesomerea.org/resources/leanrea-skill/",
  "entry_url_zh": "https://awesomerea.org/zh/resources/leanrea-skill/",
  "evidence_url": "https://github.com/aicodeking/leanrea/blob/09ba4aa0217f82cd880f855a60b877f59f39b22b/README.md",
  "group": "agents",
  "id": "leanrea-skill",
  "json_url": "https://awesomerea.org/resources/leanrea-skill/record.json",
  "kind": "integration",
  "markdown_url": "https://awesomerea.org/resources/leanrea-skill/index.md",
  "markdown_url_zh": "https://awesomerea.org/zh/resources/leanrea-skill/index.md",
  "name": "LeanREA",
  "name_zh": "LeanREA 命令行逆向分析技能",
  "record_type": "resource",
  "relationship": "independent-agent-integration",
  "requirements": [
    "A plain-text skill for a shell-capable agent; it installs no analysis engine or decompiler and needs no REA MCP server. Manual installation needs a compatible skill client, the optional npx route Node.js/npm. Recipes need their own tools, such as ripgrep, jq, Go, Electron asar tooling or Xcode command-line tools; DecompileAll.java is not bundled.",
    "The author’s benchmark used Claude Code 2.1.301, Sonnet 5.5 and macOS arm64 on Notion 7.35.1 app.asar, mac_utils.node and a gh binary built with Go 1.27.1, against rea-agents 6.5.0 over MCP. Reproducing it needs those locally obtained targets, Python 3, Node/npm, macOS tools and model access.",
    "Across 12 runs per arm, LeanREA scores 74/75 checks to REA’s 75/75 at lower time, cost and turn counts; these are author records, and the no-skill arm’s correction from 66/75 to 67/75 rests on an unpublished answer.",
    "Ground truth is hand-written regex keys, not behavioral-equivalence tests. Transcripts, target hashes and the resolved model ID (the default is the mutable sonnet alias) are unpublished, and the arms ran concurrently.",
    "The 400.7k-token figure is a separate all-tools context probe, not the comparison setup. Native analysis engines and Android were not tested, and REA also offers a CLI.",
    "MIT for the skill and benchmark code, crediting REA’s code and documentation and disclaiming affiliation; external tools, model services and targets keep their own terms."
  ],
  "requirements_zh": [
    "供具备终端访问能力的智能体使用的纯文本技能；不安装分析引擎或反编译器，也不需要 REA MCP 服务。手动安装需要兼容的技能客户端，可选的 npx 安装方式需要 Node.js/npm。各类示例依赖各自工具，例如 ripgrep、jq、Go、Electron asar 工具或 Xcode 命令行工具；仓库未附带 DecompileAll.java。",
    "作者的基准使用 Claude Code 2.1.301、Sonnet 5.5 与 macOS arm64，目标为 Notion 7.35.1 的 app.asar、mac_utils.node 以及用 Go 1.27.1 构建的 gh 二进制，对照组通过 MCP 使用 rea-agents 6.5.0。复现需要自行获取这些本地目标，以及 Python 3、Node/npm、macOS 工具和模型服务。",
    "每组 12 次运行中，LeanREA 得到 74/75 项检查，REA 为 75/75，耗时、费用和轮数更少；这些是作者记录。无技能组由 66/75 修正为 67/75，所依据的回答未公开。",
    "答案依据是手工编写的正则表达式，不是行为等价测试。回答转录、目标哈希及最终解析的模型标识（默认为可变的 sonnet 别名）均未公开，三个实验组并发运行。",
    "400.7k token 来自一次加载全部工具的独立上下文探针，并非对照实验配置。未测试原生分析引擎与 Android；REA 也提供 CLI。",
    "技能与基准代码采用 MIT，并将工作流细节归功于 REA 的代码与文档，同时声明没有隶属关系；外部工具、模型服务与目标仍有各自条款。"
  ],
  "targets": [
    "app",
    "binary"
  ],
  "url": "https://github.com/aicodeking/leanrea",
  "verification": "source-reviewed",
  "verified_at": "2026-10-17"
}