Skill flagged — suspicious patterns detected

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

Apollo Workflow

v3.0.1

把你的想法变成能用的代码:先想清楚,再一步一步做出来。每步有检查点,不完成不往下走。

0· 169·0 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 nic-yuan/apollo-workflow.

Previewing Install & Setup.
Prompt PreviewInstall & Setup
Install the skill "Apollo Workflow" (nic-yuan/apollo-workflow) from ClawHub.
Skill page: https://clawhub.ai/nic-yuan/apollo-workflow
Keep the work scoped to this skill only.
After install, inspect the skill metadata and help me finish setup.
Required binaries: git
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 apollo-workflow

ClawHub CLI

Package manager switcher

npx clawhub@latest install apollo-workflow
Security Scan
VirusTotalVirusTotal
Benign
View report →
OpenClawOpenClaw
Suspicious
medium confidence
Purpose & Capability
The name/description (a phased coding workflow) aligns with the provided scripts and prose: git-based state management, plan/design files, gate checks, and subagent dispatch. Requiring git is appropriate. However, some references (gh CLI usage in finishing-branch docs) expect remote GitHub operations but the skill declares no credentials or guidance about auth; that's a minor mismatch but explainable (the skill assumes your normal git/gh auth).
!
Instruction Scope
SKILL.md and the reference docs instruct the agent to create files, commit to git, run tests, spawn subagents (sessions_spawn), and run local shell scripts. The bundled scripts hardcode WORKFLOW_DIR=/root/.openclaw/workspace/.workflow (absolute root path). The testing doc suggests running `pip3 install pytest --break-system-packages` which is a potentially destructive command. The finishing-branch docs include `gh pr create` and push steps which will require remote credentials; the skill does not declare or request these. Overall the instructions grant broad local filesystem and git operation scope and include at least one risky command in documentation.
Install Mechanism
No external install spec or remote downloads are present. All code is included in the skill bundle as scripts and docs. There is no network fetch or archive extraction in the install step, which reduces install risk.
Credentials
The skill declares no environment variables or credentials, which is minimal and good. However, many of its actions (git push, gh pr create, running tests that may install dependencies) normally require credentials or elevated package actions. The skill does not request or document required tokens (e.g., GITHUB_TOKEN) or package-management credentials, creating a practical mismatch between declared requirements and real operation.
Persistence & Privilege
The skill writes and reads persistent state under a workspace directory (.workflow) and creates gate files; this is coherent with being a workflow manager. The absolute /root/.openclaw/workspace path is notable (it assumes a specific location and user). always:false (no forced global installation) and no modification of other skills are present.
Scan Findings in Context
[meta.json-owner-slug-mismatch] unexpected: _meta.json contains different ownerId/slug/version (sysflow / kn70str...) than the skill registry metadata (apollo-workflow / kn7ecq...). This looks like a copy/paste or leftover metadata and increases risk that files were reused without reconciliation.
[pip-break-system-packages-mention] unexpected: references/testing.md recommends `pip3 install pytest --break-system-packages`. This flag is destructive on some systems (forces installing into system-managed Python) and is not proportional to a workflow skill's documentation—should be removed or clarified.
What to consider before installing
What to check before installing or running this skill: - Review the scripts yourself. They operate on files under /root/.openclaw/workspace/.workflow and will create gate/state files and commit to git. If you run them, run in an isolated environment (throwaway VM or container) first. - The bundle contains a mismatched _meta.json (different owner/slug) — ask the publisher for clarification or prefer a skill whose metadata matches its files. - The docs suggest running `pip3 install pytest --break-system-packages`; do NOT run that on a shared or production system. Replace with a safe virtualenv-based install (python -m venv) or omit the flag. - Git push / gh pr create commands are referenced but no credentials are declared. Ensure your git/gh auth is intentional and that you understand which repo and remote will be affected. - sessions_spawn will dispatch subagents that may execute the provided scripts and git commands autonomously. If you plan to allow autonomous runs, restrict the environment and double-check the commands they will run. If you do not trust the publisher or cannot run in an isolated environment, do not enable autonomous execution; instead, manually run and inspect each step.

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

Runtime requirements

Binsgit
latestvk978jhwvy1834q2mbhh116fhzh84er31
169downloads
0stars
7versions
Updated 2w ago
v3.0.1
MIT-0

Apollo Workflow v3.0.0

Apollo-os核心定位

Apollo = 人体操作系统,workflow是它的执行引擎。

人体器官在各Phase中被调用:

  • Phase 1时 → apollo-dream整理上下文
  • Phase 3执行时 → apollo-neuro选择路径,HARD GATE验证
  • 遇到危险 → apollo-immu拦截
  • 遇到bug → Phase 4四阶段内化在workflow里
  • 需要清理 → apollo-autophagy
  • 需要记忆 → apollo-renal

一句话流程

Phase 1 → Phase 2 → Phase 3 → Phase 4 → Phase 5
  想清楚    拆任务    并行做    调错误    收尾

Phase 1: Brainstorming(想法澄清)

目的:把模糊想法变成经批准的设计文档

核心准则

在创造性工作之前,通过协作对话将想法转化为完整的设计和规格说明。

硬性门槛:

在调用任何实现技能,写任何代码、搭建任何项目、或采取任何实现动作之前,
必须先展示设计并获得主人批准。
无论感知到的复杂度如何,每个项目都要走这个流程。

流程检查表

按顺序完成:

  1. 探索项目上下文 — 检查文件、文档、最近提交
  2. 提出澄清问题 — 一次一个,理解目的/约束/成功标准
  3. 提出 2-3 种方案 — 带有权衡和推荐
  4. 展示设计 — 按复杂度缩放各部分,在每个部分后获批准
  5. 写设计文档 — 保存到 docs/plans/YYYY-MM-DD-<topic>-design.md
  6. 规格自审 — 快速内联检查占位符、矛盾、模糊、范围
  7. 主人审查书面规格 — 在继续之前请主人审查规格文件

关键原则

  • 一次一个问题 — 不要用多个问题压垮
  • 多选优先 — 比开放问题更容易回答
  • YAGNI彻底 — 从所有设计中移除不必要功能
  • 增量验证 — 展示设计,获得批准后再继续

输出

  • docs/plans/YYYY-MM-DD-<topic>-design.md
  • HG-1 gate文件:p1-design-approved.json

Phase 2: Writing Plans(任务拆解)

目的:把设计拆成2-5分钟可完成的小任务,TDD循环

核心准则

写全面的实现计划,假设执行者对代码库零上下文。

开始时宣布: "我正在用 writing-plans 技能创建实现计划。"

文件结构

在定义任务之前,映射哪些文件将被创建或修改,每个文件负责什么。

  • 用清晰边界和定义好接口的设计单元
  • 偏好更小、专注的文件,不要大的什么都做的文件
  • DRY。YAGNI。TDD。频繁commit。

小粒度任务

每步是一个动作(2-5 分钟):

- "写失败的测试" — 一步
- "运行确保它失败" — 一步
- "写最少代码让测试通过" — 一步
- "运行确保测试通过" — 一步
- "Commit" — 一步

计划文档结构

# [功能名] 实现计划

**目标:** [一句话描述构建内容]
**架构:** [2-3句话描述方法]
**技术栈:** [关键技术和库]

---

## 任务 1:[组件名]

**文件:**
- 创建:`exact/path/to/file.py`
- 修改:`exact/path/to/existing.py:123-145`

- [ ] **步骤 1:写失败的测试**
- [ ] **步骤 2:运行测试验证它失败**
- [ ] **步骤 3:写最少的实现代码**
- [ ] **步骤 4:运行测试验证它通过**
- [ ] **步骤 5:Commit**

无占位符

每步必须包含执行者需要的实际内容。以下是计划失败——永远不要写:

  • "TBD"、"TODO"、"稍后实现"、"填细节"
  • "添加适当的错误处理"/"添加验证"
  • "类似于任务 N"

输出

  • docs/plans/YYYY-MM-DD-<feature>.md
  • HG-2 gate文件:p2-plan-approved.json

Phase 3: Subagent Development(并行开发)

目的:用test-driven development + 两级review驱动subagent完成任务

两种模式

模式A:子Agent并行(适合独立任务)

当多个任务完全独立时,按问题域分配给不同Agent并行处理。

模式B:研究/综合分离(适合复杂研究任务)

多人并行研究,最后主Agent自己汇总——不把综合判断交给子Agent。

核心准则

多人并行研究,最后一个人汇总。不要让子Agent去综合——那是Coordinator的工作。

阶段谁做做什么
ResearchWorkers(并行)各自研究不同方向
SynthesisCoordinator(自己)读取结果,形成结论
ImplementationWorkers按结论执行
VerificationFresh Agent独立验证

关键规则

  1. 研究阶段可以并行
  2. 综合阶段必须自己来
  3. 验证必须 fresh eyes(新Agent来验证)
  4. 失败优先 Continue(修正失败 → 继续用同一个Agent)

输出

  • 代码 + commit
  • HG-3 gate文件:p3-all-tasks-done.json

Phase 4: Systematic Debugging(根因调试)

目的:修根因不修症状,四阶段铁律

核心准则

随机修 bug 浪费时间。快速补丁掩盖根本问题。

核心原则:永远先找根本原因再尝试修复。症状修复 = 失败。

铁律

未经根本原因调查,不许修复

如果没完成第1阶段,就不能提出修复方案。

四阶段流程

阶段 1:根本原因调查

  1. 仔细阅读错误信息 — 不要跳过错误或警告
  2. 稳定复现 — 能可靠地触发吗?具体步骤是什么?
  3. 检查最近变更 — Git diff、最近提交
  4. 追踪数据流 — 追踪到源头,在源头修复

阶段 2:模式分析

  1. 找类似工作的例子
  2. 对比参考 — 如果在实现某个模式,彻底读完参考实现
  3. 识别差异
  4. 理解依赖

阶段 3:假设与测试

  1. 形成一个假设 — "我认为 X 是根本原因"
  2. 最小化测试 — 一次只改一个变量
  3. 验证后再继续
  4. 当不知道时 — 说"我不理解 X",不要假装知道

阶段 4:实现

  1. 创建失败的测试用例
  2. 实现单一修复 — 解决识别的根本原因,一次改一个
  3. 验证修复
  4. 如果修复没用 — 停止,已经尝试了多少次?如果 < 3,回到阶段1

红旗

如果发现自己想:

  • "先快速修复,以后再调查"
  • "就试试改 X 看看行不行"
  • "加多个变更,一起跑测试"
  • 已经尝试了 2+ 次修复 — 停止,回到阶段1

如果 3+ 修复都失败: 质疑架构。与主人讨论。

输出

  • bug修复 + commit
  • HG-4 gate文件:p4-bug-fixed.json

Phase 5: Finishing Branch(收尾)

目的:验证测试 → 用户选择 → 执行决定

流程

  1. 运行完整测试套件
  2. 展示结果
  3. 用户选择交付方式:
    • Merge到main
    • 创建PR
    • 保留分支
    • 丢弃分支

输出

  • HG-5 gate文件:p5-complete.json

HARD GATE机制

每个Phase之间必须通过硬性门禁(HARD GATE),不完成不往下走。

Gate触发时机检查文件不通过则
HG-0workflow启动state.json创建退出
HG-1进入Phase 2p1-design-approved.json报错退出
HG-2进入Phase 3p2-plan-approved.json报错退出
HG-3进入Phase 5p3-all-tasks-done.json报错退出
HG-4Debug完成后p4-bug-fixed.json警告
HG-5Finish执行前p5-complete.json报错退出

状态文件

// .workflow/state.json
{
  "version": "3.0.0",
  "workflow_id": "uuid",
  "current_phase": "phase1|phase2|phase3|phase4|phase5",
  "topic": "用户请求概要",
  "design_file": "docs/plans/...-design.md",
  "plan_file": "docs/plans/...-feature.md",
  "completed_tasks": [1, 2, 3],
  "gates_passed": ["p1", "p2"]
}

Phase跳跃规则

  • 用户说"debug/bug" → 直接进 Phase 4
  • 有已批准Design → 直接进 Phase 2
  • 有已批准Plan + 用户选Subagent模式 → 直接进 Phase 3

Apollo-os调用关系

workflow执行时感知人体器官:

时机调用作用
Phase 1开始apollo-dream整理上下文,理解背景
Phase 1澄清apollo-neuro判断问题复杂度,选择路径
遇到危险操作apollo-immu拦截危险命令
Phase 3执行中HARD GATE验证每步必须验证才能继续
遇到bugPhase 4四阶段内化在workflow里,不独立调用
需要清理apollo-autophagy清理无用代码
需要记忆apollo-renal过滤噪音上下文
任务完成apollo-dream整理本次经验

快速开始

# 启动workflow
bash scripts/workflow/workflow-launch.sh "我想加一个缓存功能"

# 检查状态
bash scripts/workflow/state-manager.sh check

# 手动推进phase
bash scripts/workflow/state-manager.sh advance phase2

# 检查门禁
bash scripts/workflow/phase-gate-check.sh phase2

Comments

Loading comments...