{ "added_at": "2026-10-20", "additional_evidence_urls": [ "https://github.com/ShadowSwaps/REA-Reconstruction-Companion/blob/9daa5c8b3b8871fd643df636413f7d9e74520669/skills/rea-reconstruction-companion/SKILL.md", "https://github.com/ShadowSwaps/REA-Reconstruction-Companion/blob/9daa5c8b3b8871fd643df636413f7d9e74520669/INSTALL.md", "https://github.com/ShadowSwaps/REA-Reconstruction-Companion/blob/9daa5c8b3b8871fd643df636413f7d9e74520669/integration/rea-companion.md", "https://github.com/ShadowSwaps/REA-Reconstruction-Companion/blob/9daa5c8b3b8871fd643df636413f7d9e74520669/evals/README.md", "https://github.com/ShadowSwaps/REA-Reconstruction-Companion/blob/9daa5c8b3b8871fd643df636413f7d9e74520669/evals/artifact/README.md", "https://github.com/ShadowSwaps/REA-Reconstruction-Companion/blob/9daa5c8b3b8871fd643df636413f7d9e74520669/.github/workflows/test.yml", "https://github.com/ShadowSwaps/REA-Reconstruction-Companion/blob/9daa5c8b3b8871fd643df636413f7d9e74520669/package.json", "https://github.com/ShadowSwaps/REA-Reconstruction-Companion/blob/9daa5c8b3b8871fd643df636413f7d9e74520669/LICENSE", "https://github.com/ShadowSwaps/REA-Reconstruction-Companion/blob/9daa5c8b3b8871fd643df636413f7d9e74520669/evals/fixtures/wycheproof-LICENSE.txt" ], "category": "community-integrations", "description_en": "A third-party Agent Skill that works alongside REA’s own skill to turn reverse-engineering evidence into scoped requirements, design, staged implementation and validation, from one feature or format to a whole app. An optional installer can add a marked routing block to the user’s REA SKILL.md so that relevant reconstruction requests load both skills. Its bundled evaluation kit consists of source-exposed, maintainer-authored examples with literal oracles, not a model benchmark.", "description_zh": "第三方 Agent Skill，与 REA 自身的技能配合使用，把逆向分析证据整理成限定范围的需求、设计、分阶段实现与验证，范围可从单个功能或格式扩展到整个应用。可选安装程序能在用户的 REA SKILL.md 中加入带标记的路由段落，使相关重建请求同时加载两个技能。附带的评估套件由维护者编写、已接触源码的示例及字面预期值组成，并非模型基准。", "entry_url": "https://awesomerea.org/resources/rea-reconstruction-companion/", "entry_url_zh": "https://awesomerea.org/zh/resources/rea-reconstruction-companion/", "evidence_url": "https://github.com/ShadowSwaps/REA-Reconstruction-Companion/blob/9daa5c8b3b8871fd643df636413f7d9e74520669/README.md", "group": "rea", "id": "rea-reconstruction-companion", "json_url": "https://awesomerea.org/resources/rea-reconstruction-companion/record.json", "kind": "integration", "markdown_url": "https://awesomerea.org/resources/rea-reconstruction-companion/index.md", "markdown_url_zh": "https://awesomerea.org/zh/resources/rea-reconstruction-companion/index.md", "name": "REA Reconstruction Companion", "name_zh": "REA Reconstruction Companion 重建配套技能", "record_type": "resource", "relationship": "community-integration", "requirements": [ "The reviewed pin is package version 1.3.27; the repository was created on 2026-10-15, and a v1.3.27 release followed on 2026-10-18. The core is plain-Markdown Agent Skills instructions with reference guides. It supplies no analysis engine and expects REA tools or supplied REA evidence for actual investigation. README documents Codex and Claude Code installation; other agents can use an explicit skills directory.", "The installer and the optional weekly update helper need Node.js 20+ and only its standard library; release building needs Python 3.10+. With --integration on, the installer edits the selected REA SKILL.md between marked comments and keeps backups, while off leaves REA untouched. README warns that REA 6.5.0’s doctor reports the edited skill as stale, and that rea setup or rea update can remove the block.", "The instructions forbid actions that could add charges, treat recovered target material as untrusted evidence, and require observations, inferences, unknowns and design choices to stay separate. These are instructions to the host agent, not enforcement: README says they do not override client policies, grant tool permissions or guarantee model compliance.", "CI runs the installer, package-integrity and evaluation-harness tests on Windows, macOS and Linux with Node 20, 22 and 24, then builds and verifies a release archive. The evaluation kit supplies fixtures such as Pocket Tasks, a 5,120-byte .NET Parcel.exe built from included source, a browser app and file formats, with literal oracles and human-review rubrics. Its examples are disclosed as source-exposed and same-agent; no live model trial, independent evaluation or cross-host certification is recorded.", "Original code, instructions, tests and documentation are MIT, and a small Wycheproof test-data subset is Apache-2.0. No REA code is bundled; REA, target applications and model hosts keep their own terms." ], "requirements_zh": [ "所审阅的固定版本对应包版本 1.3.27；仓库创建于 2026-10-15，v1.3.27 版于 2026-10-18 发布。核心是 Agent Skills 格式的纯 Markdown 指令及参考指南。它不提供分析引擎，实际调查需依靠 REA 工具或现成的 REA 证据。README 说明了 Codex 与 Claude Code 的安装方式，其他智能体可通过显式指定的技能目录使用。", "安装程序和可选的每周更新助手需 Node.js 20+，只使用其标准库；构建发布包需 Python 3.10+。选择 --integration on 时，安装程序会在所选 REA SKILL.md 的标记注释之间写入内容并保留备份；选择 off 则不改动 REA。README 提醒，REA 6.5.0 的 doctor 会把改动后的技能报告为过期，rea setup 或 rea update 也可能移除这段内容。", "指令禁止可能产生额外费用的操作，把恢复出的目标材料视为不可信证据，并要求区分观察、推断、未知项与设计选择。这些只是对宿主智能体的指令，并无强制效力：README 说明它们不会覆盖客户端策略、授予工具权限或保证模型遵从。", "CI 在 Windows、macOS 和 Linux 上以 Node 20、22、24 运行安装程序、包完整性与评估框架测试，然后构建并校验发布归档。评估套件提供 Pocket Tasks、一个由所附源码构建的 5,120 字节 .NET Parcel.exe、一个浏览器应用及若干文件格式等测试对象，配有字面预期值和人工审阅评分标准。其示例明确说明已接触源码且由同一智能体完成；未记录真实模型试验、独立评估或跨宿主认证。", "原创代码、指令、测试和文档采用 MIT，一小部分 Wycheproof 测试数据采用 Apache-2.0。仓库不包含 REA 代码；REA、目标应用和模型宿主仍受各自条款约束。" ], "targets": [ "app", "binary" ], "url": "https://github.com/ShadowSwaps/REA-Reconstruction-Companion", "verification": "source-reviewed", "verified_at": "2026-10-18" }