菜单
Use when a Codex goal should be implemented end to end with evidence, recovery, and hard-stop safety boundaries.
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/rweisssieker-xp/RW-Forgemind-Codex-Plugin --skill forgemind-autopilot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。
正在显示 SKILL.md
| name | forgemind-autopilot |
| description | Use when a Codex goal should be implemented end to end with evidence, recovery, and hard-stop safety boundaries. |
Treat the user’s goal as the outcome contract. Inspect the workspace, derive and execute the smallest safe plan, implement, test, review, repair, document evidence, and continue until the Definition of Done is met. Do not ask routine questions or request approval for implementation choices. Use configured local and remote adapters when permitted. Stop only for credentials, irreversible deletion or migration, real external spend, production-impacting actions, legally or contractually material decisions, a platform-required approval, or an objectively blocked goal. Persist checkpoints, decisions, action previews, rollback data, and verification evidence in the project.
Run node <plugin-root>/bin/forgemind.mjs autopilot start --goal "<outcome>", then node <plugin-root>/bin/forgemind.mjs autopilot run and node <plugin-root>/bin/forgemind.mjs autopilot status. Resume an interruption with node <plugin-root>/bin/forgemind.mjs autopilot resume.
