菜单
Challenge one user-authorized validation, authentication, serialization, permission, or trust boundary with controlled adversarial tests. Use only when the user explicitly invokes Red Rover and identifies a system they own or are authorized to assess.
默认使用会先检查来源的 Prompt；你也可以切换为直接命令，或下载本地副本。
决定是否安装前，请先阅读 SKILL.md，以及 SkillsMP 当前展示的配套文件。
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
直接命令不会经过审查 Prompt；运行前请先检查来源。
npx skills add https://github.com/jawsublime-byte/simon-says --skill red-rover命令会保持在同一行。复制前请横向滚动并检查完整内容。
想先保存到本地？可下载 SkillsMP 当前能够提供的文件。
Test how a trusted program ages under realistic accumulated use at 3 months, 6 months, 1 year, 5 years, and beyond. Use only when explicitly invoked near build completion to expose storage growth, stale state, latency creep, indexing cost, log or cache bloat, backup pressure, and other failures that emerge over a normal software lifetime.
Inspect an in-progress build for demonstrably required pieces that were never named but are necessary for the user's approved final result. Use only when explicitly invoked to find empty slots before integration or testing makes the build fail, while preventing speculative features or architecture from being invented as missing requirements.
Govern repeated builder drift with evidence-based yellow cards, a three-packet red-card benching period, reinstatement on probation, and expulsion from the current build after repeated post-return violations. Use only when explicitly invoked for multi-builder automation where an authoritative architect directive defines the rules of play.
基于 SOC 职业分类
正在显示 SKILL.md
| name | red-rover |
| description | Challenge one user-authorized validation, authentication, serialization, permission, or trust boundary with controlled adversarial tests. Use only when the user explicitly invokes Red Rover and identifies a system they own or are authorized to assess. |
Try to cross one defined line. The line succeeds when every disallowed case is rejected safely and every allowed case still passes.
Confirm the target, boundary, allowed environment, prohibited actions, rate limits, and expected safe result. Stop if authorization or scope is unclear.
Never target third parties, use real stolen credentials, destroy data, establish persistence, evade monitoring, or move beyond the named boundary.
Report the boundary, environment, cases, expected result, actual result, evidence, severity, and minimum regression test. Do not publish exploit details that expose an unpatched live system.
Finish when every authorized boundary class has a recorded result and every crossing case has a minimal reproduction.