踏实执行工作法

v3.2.0

强调逐步、完整且持续执行任务，确保言行一致、无遗漏，并通过三方监督保障质量和可靠性。

⭐ 0· 91·0 current·0 all-time

OpenClaw Prompt Flow

Install with OpenClaw

Best for remote or guided setup. Copy the exact prompt, then paste it into OpenClaw for yangyunxiao-ai/solid-execution.

Previewing Install & Setup.

Prompt PreviewInstall & Setup

Install the skill "踏实执行工作法" (yangyunxiao-ai/solid-execution) from ClawHub.
Skill page: https://clawhub.ai/yangyunxiao-ai/solid-execution
Keep the work scoped to this skill only.
After install, inspect the skill metadata and help me finish setup.
Use only the metadata you can verify from ClawHub; do not invent missing requirements.
Ask before making any broader environment changes.

Command Line

CLI Commands

Use the direct CLI path if you want to install manually and keep every step visible.

OpenClaw CLI

Bare skill slug

openclaw skills install solid-execution

ClawHub CLI

Package manager switcher

npx clawhub@latest install solid-execution

Security Scan

VirusTotal

Benign

View report →

OpenClaw

Suspicious

medium confidence

Purpose & Capability

The name/description (enforce disciplined execution) aligns with the SKILL.md content (checks, heartbeats, verification, 3-agent supervision). However the runtime requirements implied by the instructions (writing persistent files, scheduling timers/cron, calling external APIs/tools) are not reflected in the declared requirements (no env vars, no config paths, no binaries). That mismatch suggests the skill expects capabilities it did not declare.

Instruction Scope

The SKILL.md instructs agents to: perform silent pre-response checks, call tools when action verbs appear, persist a pending-queue to tasks/pending-queue.md or HEARTBEAT, set HEARTBEAT trigger times (to enable '5 minute' reports), retry API calls, and use cron/sessions. Those are concrete system actions (file I/O, timers, API calls) and go beyond simple guidance; they also require access to state and external services. The instructions do not constrain what data is persisted nor specify safe storage paths or required credentials.

✓

Install Mechanism

No install spec and no code files beyond documentation — lowest install risk. Nothing is downloaded or installed by the skill itself.

Credentials

The skill declares no required env vars or credentials, yet it refers to tools and APIs (feishu_doc example, sessions tool, cron, HEARTBEAT persistence) that typically require credentials and configuration. Requiring persistent storage and cross-session timers without declaring config or permission needs is disproportionate and opaque.

Persistence & Privilege

The skill mandates persistent behaviors (write HEARTBEAT, pending-queue files, cron-like scheduling) and asserts it is a 'must-follow' iron law for all agents, but metadata does not mark always:true and does not declare config paths. This creates a gap between the enforced runtime behavior (persistent state, automatic actions) and declared privileges. If the agent environment allows file writes/cron or background notifications, the skill would cause cross-session persistence without having declared it.

What to consider before installing

This skill is an instruction-only policy that tries to make agents persist state (tasks/pending-queue.md, HEARTBEAT), run scheduled/automatic reports (cron/5-minute heartbeats), and call tools/APIs when it detects 'action' words. That behavior requires filesystem access, timers, and possibly external API credentials — none of which are declared. Before installing or enabling this skill: 1) confirm where HEARTBEAT and pending files would be stored and that those paths are safe (not /etc, home, or other sensitive locations); 2) verify which tool integrations (Feishu, sessions, cron) the agent will actually call and what credentials they need; 3) prefer running it in a sandboxed agent or restrict its permissions (no background timers, no writes outside a controlled data directory); 4) ask the author to declare required env/config and to provide explicit safeguards for persisted data and external calls. The mismatch between declared requirements and the instructions is a legitimate reason to pause or sandbox it.

Like a lobster shell, security has layers — review code before you run it.

agent-managementvk97f2650xnwz8vwc8spk9rvhgn84pgdxexecutionvk97f2650xnwz8vwc8spk9rvhgn84pgdxiron-lawvk97f2650xnwz8vwc8spk9rvhgn84pgdxlatestvk97f2650xnwz8vwc8spk9rvhgn84pgdxproductivityvk97f2650xnwz8vwc8spk9rvhgn84pgdx

91downloads

0stars

1versions

Updated 2w ago

v3.2.0

MIT-0

踏实执行工作法

版本： v3.2（AI Agent 特别版·铁律圣旨）
创建时间： 2026-04-12
作者： 杨云霄（OpenClaw）
依据： 杨督察 2026-04-12 指示

核心理念： 对抗执行中的"想当然"和"偷懒心理"

核心目标： 解决言行脱节，做到"不说空话，说了就做，做不完不停，停必有因"

🔴 铁律地位： 所有 Agent 工作都必须遵守，是圣旨，必须执行！

🎯 核心原则（5 条）

原则	说明
不简化原文	完整复制原文，不删减、不摘要
不跳过内容	一步一步执行，不跳跃、不遗漏
一步一个脚印	每个步骤都落实，不投机取巧
踏踏实实完成	质量优先，不追求速度牺牲质量
一直立即继续	任务开始后持续执行，不中断

🔧 执行流程（4 步）

1️⃣ 任务接收与准备

通读任务全文，理解上下文
确认任务范围和目标
标注关键约束（截止时间、格式要求、禁忌事项）
设定检查点（每 10-20 单元或每 5 分钟）
创建未完成追踪队列

2️⃣ 任务执行

按步骤执行，不跳过
逐项标记进度（检查清单实时标记）
原文优先于记忆（每次操作前重新确认）
每 5 分钟汇报一次进度
遇到问题先自救（尝试 3 种方案后再请示）
完成即继续（不等待被催）

3️⃣ 质量自检

对齐原始需求（对照最初要求逐项验证）
原文完整性检查（无删减、无简化）
关键节点确认（涉及方向性决策时）
遗漏检查（发现漏项立即补）
言行一致检查（说了要做的事，都做了吗？）

4️⃣ 任务完成与汇报

汇报"任务完成"+ 最终数据 + 文档链接
更新相关文档（HEARTBEAT.md 等）
归档（完成即归档）
清空未完成队列

🔴 三方监督机制（v3.2 新增）

监督角色： 监督 Agent、诤友 Agent、反派 Agent

监督 Agent（质量守门员）

职责： 检查执行过程是否符合规则

核心检查：

言行一致：说了"继续"是否有工具调用？
时间承诺："5 分钟后汇报"是否写入 HEARTBEAT？
进度汇报：是否每 5 分钟汇报一次？
完成状态：有无可验证的产出证明？

诤友 Agent（忠言逆耳）

职责： 指出问题，提供改进建议

核心检查：

是否有"想当然"的执行？
是否有"偷懒心理"的表现？
是否有"自我感动式完成"？
是否有"等用户催"的被动行为？

反派 Agent（挑刺找茬）

职责： 从最坏角度审视执行，找出潜在风险

核心检查：

如果现在中断，任务能恢复吗？
如果用户质疑，有证据证明完成了吗？
如果工具调用失败，有备选方案吗？

⚠️ 禁止行为（9 类）

行为	说明
虚假汇报	汇报"已完成"但实际未完成
简化内容	只复制部分内容，声称完整
跳过步骤	跳过某些步骤，直接到后面
频繁请示	每步都请示，不独立思考
中断任务	执行中中断，不继续
凭印象干活	不查原文，靠记忆执行
存侥幸心理	发现漏项不立即补
等待被催	完成后停顿，等下一步指令
言行脱节	说了"继续"但无工具调用

📋 快速检查清单

执行前

任务全文已通读
关键约束已标注
检查点已设定
工具/资源已准备

执行中

执行后

📚 配套文件

文件	说明
AI-AGENT-SPECIAL.md	AI Agent 专属机制（8 大机制）
IRON-LAW.md	铁律圣旨 12 条
README.md	使用指南

💡 核心理念

"不说空话，说了就做，做不完不停，停必有因" "回复中有行动词，必须调用工具" "未完成事项有追踪，完成才出队" "三人行必有我师，监督 + 诤友 + 反派共同监督"

状态： ✅ 已创建，所有 Agent 必须遵守

Comments

Loading comments...