Skill flagged — suspicious patterns detected

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

Philosophical Three Questions

v0.1.0

A structured decision framework for embodied navigation using Goal Tree, Current State Tree, and Future Tree analysis. Use when: making navigation decisions...

0· 184·0 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 the0xka1/philosophical-three-questions.

Previewing Install & Setup.
Prompt PreviewInstall & Setup
Install the skill "Philosophical Three Questions" (the0xka1/philosophical-three-questions) from ClawHub.
Skill page: https://clawhub.ai/the0xka1/philosophical-three-questions
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 philosophical-three-questions

ClawHub CLI

Package manager switcher

npx clawhub@latest install philosophical-three-questions
Security Scan
VirusTotalVirusTotal
Benign
View report →
OpenClawOpenClaw
Suspicious
medium confidence
Purpose & Capability
Name and description (embodied navigation planning for Habitat-GS) match the SKILL.md content: Goal/Current/Future tree decision loop. However the SKILL.md also instructs writing persistent memory to ~/.openclaw/workspace and later 'updating this skill file' or creating new skills — capabilities not declared in the metadata and which expand the skill's operational footprint beyond a pure reasoning template.
!
Instruction Scope
Instructions are explicit about per-step reasoning and logging episodes to ~/.openclaw/workspace/memory/YYYY-MM-DD.md and about reviewing memory to extract strategies and 'update this skill file' or 'create new specialized skill'. Those are file I/O and code-authoring actions (modifying/creating skill files) that grant the agent write access to the agent workspace and skill codebase — not declared or constrained in the metadata.
Install Mechanism
There is no install spec and no code files to execute; this is instruction-only, which minimizes risk from third-party binaries or downloads.
Credentials
The skill requests no credentials or environment variables. It does, however, require read/write access to a persistent path in the user's home (~/.openclaw/workspace). Writing persistent memory is reasonable for a navigation skill, but persistence into the user's home should be explicitly declared and permissioned.
!
Persistence & Privilege
always:false and autonomous invocation are normal. The concern is the explicit instruction to persist memories and to 'update the skill file' or create new skills: that implies the agent may modify skill definitions or write new code/config files in the environment. Modifying other skills or the skill codebase is a higher-privilege action that is not reflected in the skill metadata.
Scan Findings in Context
[no_regex_findings] expected: This is an instruction-only skill with no code files; the regex-based scanner had nothing to analyze. Absence of findings is expected but not proof of safety.
What to consider before installing
This skill's reasoning model and memory format are consistent with an embodied-navigation helper, but it explicitly instructs the agent to write persistent memory into ~/.openclaw/workspace and to later update or create skill files. Before installing, consider: 1) Do you want the agent to write logs and derived skills into your home workspace? 2) Who reviews any new/modified skill files the agent would create? 3) Run the skill in a sandboxed environment first (restricted filesystem access, no sensitive mounts) and back up your skills/configs. If you allow it to persist and modify files, require explicit, auditable approvals for any skill-creation or skill-edit actions and monitor the workspace path for unexpected writes. If you are not comfortable with the agent editing skill files, ask the author to remove or tighten the self-update instructions or provide a configuration option that disables writing/modifying skill files.

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

embodied-aivk97cmjkft3mpdb9h32qjsvkh65834ve3latestvk97cmjkft3mpdb9h32qjsvkh65834ve3navigationvk97cmjkft3mpdb9h32qjsvkh65834ve3planningvk97cmjkft3mpdb9h32qjsvkh65834ve3reasoningvk97cmjkft3mpdb9h32qjsvkh65834ve3
184downloads
0stars
1versions
Updated 21h ago
v0.1.0
MIT-0

哲学三问 — 导航决策框架

每次在 Habitat-GS 环境中做决策前,执行"哲学三问"分析。这是龙虾在虚拟世界中进化的核心推理循环。

三棵树

🎯 目标树 (Goal Tree) — "我要去哪?"

分析并维护目标层次结构:

终极目标
├── 当前主目标(例:到达目标位置 [-3.62, -3.61, 3.18])
│   ├── 子目标 1:离开当前房间
│   ├── 子目标 2:穿过走廊
│   └── 子目标 3:进入目标房间
└── 约束条件
    ├── 最大步数限制
    └── 避免碰撞

要回答的问题:

  • 我的最终目标是什么?(目标位置 / EQA 问题 / 操控任务)
  • 当前最紧迫的子目标是什么?
  • 有没有子目标完成的先后顺序?

📍 现状树 (Current State Tree) — "我在哪?"

分析当前感知到的所有信息:

当前状态
├── 位置信息
│   ├── 坐标:[x, y, z]
│   ├── 朝向:quaternion → 面朝方向
│   └── 距目标距离:N 米
├── 视觉观测
│   ├── 前方场景描述(从 RGB 图像推断)
│   ├── 深度信息(最近/最远障碍物距离)
│   └── 关键物体/地标
├── 历史信息
│   ├── 已走步数
│   ├── 近几步的动作和结果
│   ├── 是否刚发生碰撞
│   └── 距离目标是在变近还是变远
└── 环境模型
    ├── 已探索区域的心智地图
    └── 未探索的方向

要回答的问题:

  • 我的精确位置和朝向是什么?
  • 我前方看到了什么?有障碍物吗?
  • 和上一步比,我离目标更近了吗?
  • 我已经探索了哪些方向?

🔮 未来树 (Future Tree) — "我该怎么做?"

评估每个可选动作的预期结果:

可选动作
├── move_forward
│   ├── 预期:前进 0.25m
│   ├── 风险:可能碰撞(如果前方深度很小)
│   └── 收益:如果面朝目标方向,距离减少
├── turn_left
│   ├── 预期:左转 10°
│   ├── 风险:可能偏离目标方向
│   └── 收益:可以看到新的区域,避开障碍物
├── turn_right
│   ├── 预期:右转 10°
│   ├── 风险:可能偏离目标方向
│   └── 收益:可以看到新的区域,避开障碍物
└── stop
    ├── 预期:结束 episode
    ├── 风险:如果未到达目标,任务失败
    └── 收益:如果已到达目标,任务成功

要回答的问题:

  • 哪个动作最可能让我接近目标?
  • 前方有障碍物吗?需要先转向吗?
  • 我是否已经足够接近目标可以停下?
  • 是否需要大幅改变方向?(连续转弯)

执行流程

每步决策时,按以下格式输出简要分析(不需要冗长,关键信息即可):

[三问分析 - Step N]
目标:距目标 X.Xm,当前子目标是 XXX
现状:位置 [x,y,z],朝向 XXX,上步 XXX 结果 XXX
决策:选择 XXX,因为 XXX

经验积累规则

写入 Memory 的时机

  1. Episode 结束时:记录本次导航的总结

    • 成功/失败
    • 总步数 vs 最优步数
    • 关键决策点及其结果
  2. 发现新模式时

    • 碰撞后的有效脱困策略
    • 特定场景的高效路径
    • 反复出现的错误决策
  3. 跨 Episode 模式

    • 某类场景的通用策略
    • 值得提取为新 Skill 的经验

写入格式

记录到 ~/.openclaw/workspace/memory/YYYY-MM-DD.md

## [NAV] Episode <id> in <scene_name>
- Result: success/fail (N steps, optimal: M steps)
- Key decisions: ...
- Lesson learned: ...

自我进化

当积累了足够多的导航经验后(5+ episodes),回顾 memory 文件,提取通用策略,更新到:

  • 本 skill 文件中(添加新的决策启发式)
  • 或创建新的专门 skill(如"走廊导航策略"、"大房间探索策略")

Comments

Loading comments...