Skill flagged — suspicious patterns detected

ClawHub Security flagged this skill as suspicious. Review the scan results before using.

SDD Executing Plans

v1.0.0

执行 spec-plan.md 实现计划,自动完成任务、验证步骤并包含重试逻辑。

0· 195·0 current·0 all-time

Install

OpenClaw Prompt Flow

Install with OpenClaw

Best for remote or guided setup. Copy the exact prompt, then paste it into OpenClaw for mahingbun-dev/sdd-executing-plans.

Previewing Install & Setup.
Prompt PreviewInstall & Setup
Install the skill "SDD Executing Plans" (mahingbun-dev/sdd-executing-plans) from ClawHub.
Skill page: https://clawhub.ai/mahingbun-dev/sdd-executing-plans
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 sdd-executing-plans

ClawHub CLI

Package manager switcher

npx clawhub@latest install sdd-executing-plans
Security Scan
VirusTotalVirusTotal
Suspicious
View report →
OpenClawOpenClaw
Suspicious
medium confidence
!
Purpose & Capability
The declared metadata lists no required config paths, binaries, or env vars, but the instructions require reading a workspace config file (.sdd-workspace) and reading/updating spec-plan.md files under {workspace}/spec/. Those are necessary for the stated purpose but were not declared. Additionally, checks/verification steps imply running build/test/verification commands which may need specific binaries (e.g., test runners, compilers) that are not listed.
!
Instruction Scope
The SKILL.md directs the agent to: read .sdd-workspace, scan and open spec-plan.md files, execute every 'execution' step and run 'check' commands, update the plan files in-place (toggle - [ ] to - [x]), perform retries, and stop on blocking failures. It allows running arbitrary verification commands referenced by the spec but does not enumerate which commands or limit scope, meaning the agent may execute arbitrary system commands or modify many files based on the plan content.
Install Mechanism
This is an instruction-only skill with no install steps or external downloads, so there is no installer risk. However, absence of an install spec increases reliance on the runtime environment matching the skill's unstated requirements.
!
Credentials
No env vars or credentials are declared, yet the skill requires access to a workspace config path and will read and write files. It may also need access to tools or credentials implicitly (for example, running tests that access network services or private registries). The lack of declared config paths and required binaries is a mismatch and reduces transparency about what the skill can access or needs.
Persistence & Privilege
The skill is not 'always' and does not request persistent platform-level privileges, but it explicitly writes to user-owned files (updating spec-plan.md in-place). That is a significant side-effect: installing/enabling the skill grants it the ability to change plans and mark steps completed. This is expected for a plan-execution tool, but it should be clearly declared and made explicit to the user.
Scan Findings in Context
[no_regex_findings] expected: The scanner found no code files to analyze (instruction-only SKILL.md). This is expected for an instruction-only skill, but lack of code does not eliminate runtime risks because the instructions require filesystem writes and command execution when run by the agent.
What to consider before installing
This skill will read and modify files in your OpenClaw workspace (it expects a .sdd-workspace pointing to {workspace} and will update spec-plan.md in-place). Before enabling it: (1) confirm where .sdd-workspace should live and that you trust the plans it will run; (2) backup your spec/ directory or run the skill on a copy, since it will change checkboxes and may create/modify files; (3) ask the author to declare required config paths and any binaries/tools the verification steps need (tests/build tools, etc.); (4) prefer interactive/manual execution or require explicit user confirmations if you are uncomfortable with fully autonomous changes; and (5) review the spec-plan.md contents to ensure the verification steps do not execute unexpected system/network commands. If those gaps are not resolved, treat the skill as risky to run autonomously.

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

latestvk9787h32zksett4v5z4r1ma8ms8329yf
195downloads
0stars
1versions
Updated 1h ago
v1.0.0
MIT-0

SDD Executing Plans — 执行计划

Overview

加载 spec-plan.md,按顺序执行所有 Task 到完成,验证每个步骤,并报告结果。

核心原则: 自主执行 — 连续运行所有 Task,不进行手动批量暂停。

启动时声明: "我正在使用 sdd-executing-plans 技能来执行此计划。"

关键概念

  • 工作区 (Workspace): 通过 .sdd-workspace 配置文件中 workspace_path 指定的根目录
  • Spec 目录: 所有 SDD 文档存储在 {workspace}/spec/

Step 0: 读取工作区配置

在任何操作之前,必须读取工作区配置:

  1. 检查当前 OpenClaw workspace 中是否存在 .sdd-workspace
  2. 如果存在,读取 workspace_path 作为工作区根目录 {workspace}
  3. 如果不存在,显示错误:"请先运行 /sdd-global-init 初始化工作区。" 并停止

验证工作区目录存在,如果不存在提示用户重新初始化。

Step 1: 模型检查

检查当前模型是否为 Opus。如果是 Opus,输出以下纯文本消息并继续(非阻塞):

⚠️ 当前模型是 Opus,执行计划任务不需要 Opus 级别的推理能力,建议切换到更快的模型以提升执行效率。输入 /model 切换模型。

Step 1: 计划选择

如果提供了路径参数

  • 直接使用它作为计划文件路径
  • 读取文件并继续 Step 2

如果没有提供路径参数

  1. 扫描 {workspace}/spec/ 中所有匹配 feature_*/spec-plan.md 的文件
  2. 按目录名排序(feature 目录包含日期,所以字母排序 = 时间顺序)
  3. 通过 AskUserQuestion 展示最新的 3 个计划:
    • 每个选项显示目录名和计划标题(第一个标题)
    • 用户也可以通过 "Other" 输入自定义路径
  4. 读取选中的 spec-plan.md

Step 2: 恢复检测与执行前确认

恢复检测

读取计划后,扫描文件中的所有复选框标记(- [ ]- [x]):

  • 如果存在任何 - [x] → 这是恢复执行

    • 完全跳过执行前确认
    • 找到仍然包含至少一个 - [ ] 步骤的第一个 Task
    • 在该 Task 中,识别第一个未勾选的步骤(- [ ])作为恢复点
    • 打印:"从 Task N: [任务名] 恢复执行,步骤: [步骤描述]"
    • 从恢复点直接继续 Step 3
  • 如果所有复选框都是 - [ ] → 这是全新执行,继续执行前确认。

执行前确认(仅全新执行)

显示汇总:

  • 功能名称(来自计划标题)
  • Task 总数(Task 部分的数量)
  • 关键文件 要创建或修改的文件

使用 AskUserQuestion 提供选项:

  • "开始执行" — 继续执行
  • "取消" — 中止

只有用户选择 "开始执行" 才继续。

Step 3: 执行循环

按顺序严格执行 Task(Task 1, Task 2, Task 3...)。不批量处理,Task 之间不暂停。

识别步骤类型

通过步骤出现的章节标题识别步骤类型:

  • **执行步骤:** → 执行步骤(实现代码更改)
  • **检查步骤:** → 检查步骤(运行验证命令)
  • **End-to-end verification:** → 验收 Task 验证(视为检查步骤)

对于每个 Task:

  1. 执行所有执行步骤(在 **执行步骤:** 下):

    • 按顺序处理每个 - [ ] 步骤
    • 步骤完成后,立即更新计划文件:将该步骤的 - [ ] 改为 - [x]
    • 跳过已标记为 - [x] 的步骤(恢复执行场景)
  2. 运行所有检查步骤(在 **检查步骤:****End-to-end verification:** 下):

    • 按顺序处理每个 - [ ] 检查
    • 检查通过后,立即更新计划文件:将该检查的 - [ ] 改为 - [x]
    • 跳过已标记为 - [x] 的检查(恢复执行场景)
  3. 处理检查失败:

    • 分析失败输出
    • 尝试修复问题(已检查的执行步骤保持 - [x] — 不要取消勾选)
    • 重新运行失败的检查命令
    • 每个检查项最多 5 轮重试
    • 如果 5 轮后仍然失败 → Task 被阻塞,进入 Step 4
  4. Task 在以下情况下完成

    • 所有执行步骤都是 - [x]
    • 所有检查步骤都是 - [x]
    • 继续下一个 Task

验收 Task(如果存在)的处理方式与其他 Task 完全相同 — 执行其步骤并运行其验证。

Step 4: 阻塞 → 立即停止

当被阻塞(任何验证项重试耗尽 5 轮)时:

  1. 立即停止整个执行 — 不要跳到下一个 Task
  2. 直接进入 Step 5(执行摘要)

Step 5: 执行摘要

完成或阻塞后打印简洁摘要:

## 执行摘要
- 任务: X 完成 / Y 总计
- 新建文件: [列表]
- 修改文件: [列表]
- 💡 **建议下一步:** 运行 `/sdd-plan-human-verify` 生成人工验收清单

如果阻塞,同时包含:

## 阻塞信息
- 任务: [Task 编号和名称]
- 验证项: [失败的检查项]
- 最后错误: [简要描述]

规则

  • 开始时检测模型 — 检测到 Opus 时,输出纯文本建议切换到更快的模型;非阻塞,始终继续
  • 无外部技能依赖 — 此技能完全独立
  • 无批量审核暂停 — 连续运行所有 Task
  • 无 git 操作 — 不进行 branch、commit、merge、push
  • 严格按照计划步骤执行
  • 永不跳过验证
  • 阻塞时停止,不猜测
  • 只有两种状态的复选框:- [ ](待处理)和 - [x](已完成)
  • 不要生成验收清单 — 那由 sdd-plan-human-verify 技能处理

Comments

Loading comments...