agentleFS
edisonzerolamedisonzerolam/openclaw-skills/agent-planner/SKILL.md

agent-planner

Skill0 starsChanged 4 months agoMITView on GitHub

Agent 规划修正技能。用于 auditor 的 P-Sub-P 阶段,提供 P1/P2/P3 级规划。触发词:「规划」「方案设计」「Agent 架构」「任务分解」。不适用:纯信息查询、单步执行指令。

6 other projects use a version of this.

Adopted by

Most recently taken up by

  • Other versions6 projects use a version of this.

    Word for word · 0

    None found.

    Edited · 0

    None found.

    Wrote their own version · 6

    1. ruvnet/claude-flow★ 80ksame name
    2. ruvnet/ruflo★ 79ksame name
    3. Zekiog/ruflo★ 2same name
    4. fidgetcoding/fidgetflo★ 1same name
    5. jpoindexter/theft-kit★ 1same name
    6. Soulcynics404/AgentForge★ 1same name

    How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.

  • In this repository5 other files it gives its agents.

The file

Files
SKILL.md

需求路由

def route_demand(demand: str) -> str:
    d = demand.lower()
    if any(kw in d for kw in ["派生子代理", "spawn", "执行审查"]): return "local"
    if any(kw in d for kw in ["规划", "方案设计", "帮我生成"]): return "llm"
    if any(kw in d for kw in ["坑点", "风险"]): return "hybrid"
    if any(kw in d for kw in ["检索", "知识库"]): return "hybrid"
    return "llm"
模式说明
local子代理派发、知识缓存、索引更新
llm规划方案生成、Agent 架构设计
hybrid本地优先 → 置信度<70% 时 fallback 到 LLM

执行流程

F0 — 调查(强制入口)

任何规划任务必须先完成调查,产出调查摘要,再进入 F1。

调查维度:事实 / 范围 / 深度 / 依赖 / 风险 / 资源

调查摘要格式:

## 调查摘要
- 事实:{现有条件概述}
- 范围:{规划边界}
- 复杂度:{简单/中等/复杂 + 依据}
- 主要依赖:{依赖项}
- 风险点:{风险}
- 资源约束:{时间/Token/预算}

F0.5 — 事实验证(强制)

调查摘要产出后、进入 F1 前,必须对关键数字声明执行现场验证。

验证维度:文件行数 / 文件大小 / 路径存在性 / API 签名 / 版本号 / 文件数量

from fact_check import verify_source_claims
verifications = verify_source_claims(source_claims, base_dir=".")

结果判定:

结果条件动作
PASS所有验证通过,偏差<10%放行 → F1
WARN偏差 10%-30%标记待核实, 用户决策
FAIL偏差>30% 或文件不存在阻塞, 修正后重试

验证结果写入调查摘要的 verification_results 字段。

F1 — 前置验证

  • rules.md + identity-config.md 存在 → 否:阻塞
  • 核心 skill 已安装 → 否:降级到通用能力
  • 变更目标可确认 → 否:追问确认

F2 — 用户确认

  • 规划方案输出后等待用户确认
  • 修改型变更(≥P1)需显式确认

F3 — 成本预估

指标预估方式
Token 消耗简单≤5K / 中等≤20K / 复杂≤50K
模型调用次数Task 数 × 1.2(含重试)
迭代上限5 版,超限强制用户决策

F4-F12 — 详细规划流程

详细流程(F4 架构方案 → F12 残余 Cron 生成)见 references/workflow.md。

步骤概要
F4 架构方案Agent 组成 / rules.md / identity-config.md / 通信 / 降级
F5 技能选型KANO 模型:M(必备)/O(期望)/A(魅力),MVP 只含 M
F6 工作流编排Epic→Task→Step 三级分解,含并行标记
F7 坑点预警读取坑点库 + 外部经验,≥3 个匹配坑点
F8 环境配置依赖清单 + 验证命令 + 安装方式
F9 Workspace 卫生4-Zone 模型检查
F10 适用场景明确适用/不适用边界
F11 子代理调用3 级执行路径(L1-L3)
F12 交付追踪T1-T5 版本追踪 + 残余 Cron 生成

阶段A — 审计(强制)

规划方案完成后、用户确认前必须执行审计。见 references/audit-protocol.md。

维度通过标准
完整性质量维度全部≥3/5
准确性无虚假/夸大描述,数字经 F0.5 验证
一致性与调查摘要/rules.md 一致
可执行性Task 依赖/步骤清晰
风险控制坑点≥3 个且降级策略完整

知识库检索

规划涉及特定领域时检索知识库。详见 _knowledge/_enhancement/knowledge-base-integration.md。

关键词目录优先级
医疗器械/投标medical-device, biddingP0
股票/A股/港股stockP1
基金/ETFfundP1
理财/资产配置financeP1

质量自检

维度评分
完整性X/5
可执行性X/5
风险控制X/5
Token 效率X/5
模型调用效率X/5

综合≥3.5 → 通过;<3.0 → 需改进

版本迭代

  • 迭代上限 5 版,超限强制用户决策
  • 版本号格式:v{主}.{次}
  • 每版记录变更摘要到 _knowledge/_enhancement/plan-tracker/evolution-log.md

分级加载索引

层级文件加载时机
L1SKILL.md(核心+F0-F3)首次加载
L2references/workflow.md, references/audit-protocol.mdSKILL.md 后自动
L2_knowledge/_enhancement/knowledge-engineering-pattern.mdSKILL.md 后自动
L3_knowledge/_enhancement/knowledge-base-integration.md知识库检索关键词命中
L3_knowledge/_enhancement/plan-tracker/*.mdF12 版本追踪触发
L3_knowledge/_enhancement/pitfall-library.mdF7 坑点预警触发
L4_knowledge/_enhancement/plan-template.md规划文档输出
L4_knowledge/_enhancement/spawn-patterns.mdF11 spawn 调用
L4_knowledge/_enhancement/workspace-zones.mdF9 workspace 检查

自学习 & 事实核查

事实核查:每次执行完成、返回结果前,必须调用 _shared/fact-checker/fact_check.py。 详见 references/self-learning.md。

参考文件

文件用途
references/workflow.mdF4-F12 详细流程
references/audit-protocol.md阶段A 审计 + 质量自检
references/self-learning.md自学习 + 事实核查 + 优化模块
_knowledge/_enhancement/plan-template.md规划方案模板
_knowledge/_enhancement/pitfall-library.md坑点库 v1.1(12 坑)
_knowledge/_enhancement/spawn-patterns.mdF11/E6 子代理调用模式
_knowledge/_enhancement/workspace-zones.md4-Zone Workspace 分区
_knowledge/_enhancement/knowledge-base-integration.md知识库检索
_knowledge/_enhancement/plan-tracker/版本追踪 + 残余 Cron
_knowledge/_enhancement/expert-distillation-flow.md蒸馏/专家注入
_knowledge/_refined/LEARNINGS.md经验精化记录
_knowledge/_index/关键词/主题索引
_knowledge/core-principles/核心原则
references/plan-tracker-deprecated.md旧版追踪(已弃用)

故障与降级

场景处理
fact_check.py 不可达回退到 exec 命令验证
子代理全部超时强制降级主会话执行
知识缓存 I/O 失败跳过缓存,直接读取文件
F0.5 验证 FAIL阻塞,通知用户修正后重试
审计迭代超 5 版强制用户决策

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.