Auto Revolution
This is the ClawHub publishing variant.
Use this skill for
- - creating structured task JSON files
- activating queued tasks whose dependencies are complete
- scanning proposed instruction text for risky shell patterns
- applying review JSON back into a task file
- logging workflow events in JSONL
- generating a review prompt for manual or supervised review
Do not use this package for
- - autonomous cron loops
- unsupervised background execution
- direct execution of arbitrary shell instruction strings
- remote agent spawning from bundled scripts
Files to use
- -
task-schema.json for task structure reference - INLINECODE1 for example role configuration
- INLINECODE2 to create a task interactively
- INLINECODE3 to move ready tasks from
queued to INLINECODE5 - INLINECODE6 to scan instruction text or a task review block
- INLINECODE7 to generate a review prompt for supervised review
- INLINECODE8 to write a review result back into a task JSON file
- INLINECODE9 to append JSONL audit events
- INLINECODE10 ,
scripts/force-unlock.sh, scripts/unblock-task.sh for task lock maintenance
Minimal workflow
- 1. Create or edit a task JSON.
- If the task depends on other tasks, run
scripts/activate-queued-tasks.js when dependencies may be complete. - Generate a review prompt with
scripts/trigger-review.js. - Obtain a supervised review result as JSON.
- Apply that result with
scripts/apply-review.js. - If you need to inspect instruction text, run
scripts/security-scan.js before any manual execution.
Safety rules
- - Keep a human in the loop for any real execution.
- Treat generated instructions as untrusted until reviewed.
- Prefer structured task updates over free form shell execution.
- Do not add scripts that call
execSync on model generated instruction strings in the publishing package.
Notes
This package is intentionally minimal for publication. Internal development documents, autonomous heartbeat runners, and direct execution engines belong in the private local version, not the published package.
自动革命
这是ClawHub发布变体。
使用此技能用于
- - 创建结构化的任务JSON文件
- 激活依赖项已完成的排队任务
- 扫描建议的指令文本以查找危险的shell模式
- 将审查JSON应用回任务文件
- 以JSONL格式记录工作流事件
- 生成用于人工或监督审查的审查提示
请勿将此包用于
- - 自主cron循环
- 无监督的后台执行
- 直接执行任意shell指令字符串
- 从捆绑脚本生成远程代理
使用的文件
- - task-schema.json 用于任务结构参考
- config/models.json 用于示例角色配置
- scripts/create-task.js 用于交互式创建任务
- scripts/activate-queued-tasks.js 用于将就绪任务从queued移至pending
- scripts/security-scan.js 用于扫描指令文本或任务审查块
- scripts/trigger-review.js 用于生成监督审查的审查提示
- scripts/apply-review.js 用于将审查结果写回任务JSON文件
- scripts/log-event.js 用于追加JSONL审计事件
- scripts/atomic-lock.sh、scripts/force-unlock.sh、scripts/unblock-task.sh 用于任务锁维护
最小工作流
- 1. 创建或编辑任务JSON。
- 如果任务依赖其他任务，在依赖项可能完成时运行scripts/activate-queued-tasks.js。
- 使用scripts/trigger-review.js生成审查提示。
- 获取监督审查结果（JSON格式）。
- 使用scripts/apply-review.js应用该结果。
- 如需检查指令文本，在任何手动执行前运行scripts/security-scan.js。
安全规则
- - 任何实际执行都需有人工参与。
- 在审查前将生成的指令视为不可信。
- 优先使用结构化任务更新而非自由格式的shell执行。
- 不要在发布包中添加对模型生成的指令字符串调用execSync的脚本。
备注
此包特意保持最小化以便发布。内部开发文档、自主心跳运行器和直接执行引擎应属于私有本地版本，而非发布包。
