From p8-advisor
Provides Agent design consulting and review based on the 12-Factor AgentOps framework. Helps design new agents and review existing agents, skills, and workflows for issues.
How this skill is triggered — by the user, by Claude, or both
Slash command
/p8-advisor:agent-optimizerThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
基于 **12-Factor AgentOps** 框架,提供 Agent 设计咨询和现有 Agent/Skill 的问题审查与优化建议。
assets/templates/review-template.mdreferences/factor-1-context.mdreferences/factor-10-12-scale.mdreferences/factor-2-git.mdreferences/factor-3-one-task.mdreferences/factor-4-research.mdreferences/factor-5-validation.mdreferences/factor-6-ratchet.mdreferences/factor-7-8-knowledge.mdreferences/factor-9-metrics.md基于 12-Factor AgentOps 框架,提供 Agent 设计咨询和现有 Agent/Skill 的问题审查与优化建议。
原文来源:https://www.12factoragentops.com
中文翻译参考:~/Documents/working/translations/12-factor-agentops/
12 个因素分三层:
基础层(I–III):单 Agent 可靠性
质量层(IV–VI):确保工作质量
学习层(VII–IX):从经验复利
规模层(X–XII):多 Agent 工作流
详细说明见 references/ 目录下各因素的参考文档。
用户提出 Agent 设计问题时:
常见设计问题清单(引导对话用):
用户提交现有 Agent/Skill/工作流时:
审查报告格式:
## Agent 审查报告
### 总体评分
[对照 12 因素的覆盖情况,给出健康度评估]
### 发现的问题
#### 🔴 严重问题(会直接导致失败或错误)
- [问题描述] → 违反因素:[X]
改进建议:[具体怎么做]
#### 🟡 改进项(影响质量或效率)
- [问题描述] → 违反因素:[X]
改进建议:[具体怎么做]
#### 🟢 做得好的地方
- [值得保留的设计]
### 优先改进计划
1. [最重要的改进,一句话]
2. [次重要的改进,一句话]
3. [...]
遇到以下关键词,直接联想对应因素:
| 关键词 | 可能违反的因素 | 快速诊断问题 |
|---|---|---|
| "一个会话做了很多事" | III | context 饱和,任务边界不清 |
| "让 Agent 自己检查自己" | V | 自我验证 = 确认偏误 |
| "所有东西都塞进 system prompt" | I | context 预算失控 |
| "没有调研直接写代码" | IV | 缺少调研阶段 |
| "用了 Notion/Confluence 存经验" | II、VIII | 知识孤岛,Agent 无法检索 |
| "多个 Agent 共享一个目录" | X | 竞态条件风险 |
| "Agent 一直重试不升级" | XI | 缺少监督层级 |
| "只统计 token 用量/会话数" | IX | 虚荣指标,没看结果 |
| "失败就丢弃,下次重来" | XII | 没有提取失败智慧 |
| "用完的知识不更新不删" | VIII | 知识衰减导致误导 |
详细的因素说明存放在 references/ 目录,按需加载:
references/factor-1-context.md — 上下文管理详解references/factor-2-git.md — Git 追踪模式references/factor-3-one-task.md — 任务隔离与交接references/factor-4-research.md — 调研阶段设计references/factor-5-validation.md — 外部验证体系references/factor-6-ratchet.md — 棘轮进度锁定references/factor-7-learnings.md — 经验提取格式references/factor-8-compound.md — 知识复利飞轮references/factor-9-metrics.md — 适应性指标references/factor-10-isolation.md — 工作单元隔离references/factor-11-supervision.md — 层级化监督references/factor-12-failures.md — 失败智慧收获遇到用户提到某具体因素相关的问题时,读对应的 reference 文件以获取详细内容。
npx claudepluginhub chujianyun/skills --plugin p8-advisorAudits agent codebases against the 12-Factor Agents methodology, analyzing per-factor compliance with file-level evidence. Use when reviewing LLM-powered system architecture or planning agent improvements.
Provides battle-tested engineering principles for AI coding agents including plan-first workflow, subagent delegation, self-improvement loops, and verification gates. Use when configuring agent behavior, writing AGENTS.md files, or improving agent reliability.
Designs AI agents with recommended patterns and architectures: workflow vs agent, orchestrator/subagent pipelines, and Claude Code agent templates.