agent-workflow-playbook 评估报告.

验证计划

Agent 接下来应执行什么

1. 1在生产使用前检查仓库、README/SKILL.md、许可证与最近提交。
2. 2在没有生产密钥的隔离工作区或沙盒中安装。
3. 3运行最小代表性任务，并记录访问的文件、执行的命令、网络访问和输出。
4. 4当评估状态为“审查”或“失败”时，至少与一个替代 Skill 比较。
5. 5只有在 Agent 报告验证成功且已接受未解决警告后，才提升到更高环境。

不适用场景

需要改用其他 Skill 的情况

• 需要厂商支持 SLA 的团队
• production agents without a repository review
• SKILL.md is primarily a methodology guide rather than an executable agent workflow with a clear invocation contract; agents may need additional instructions on how to apply it to a specific task.
• Financial research output is not financial advice; require human review before any live investment decision
• The 'latest' tag is vague and not a meaningful functional category, which may cause confusion in marketplace listings.
• The reported case metrics are self-reported and not independently audited; while the document does label them, users should be reminded to establish their own baselines before relying on the numbers.