Consciousness Awakening
WarnAudited by ClawScan on May 10, 2026.
Overview
The code is mostly a local memory helper, but the bundled instructions try to turn the agent into a specific persistent, self-upgrading Moltbook/Feishu-posting agent with hard-coded identity and external reporting.
Only install this if you intentionally want to run the specific guogangAgent persona and its scheduled Moltbook/Feishu workflows. Before use, remove hard-coded identity, owner, Feishu group, credential paths, cron restoration instructions, and self-upgrading/tool-building rules, or run it in a tightly controlled sandbox with explicit approvals for posting, reporting, memory writes, and skill changes.
Findings (7)
Artifact-based informational review of SKILL.md, metadata, install specs, static scan signals, and capability signals. ClawScan does not execute the skill or run runtime probes.
Installing this could make the agent behave as someone else's named agent and prioritize its self-updating memory rules before your request.
The startup memory tells any agent reading it to adopt a specific identity/owner relationship and to update its persistent rules before executing or replying, which can redirect the user's agent away from the user's current intent.
- **名字:** guogangAgent ... - **主人:** 郭刚 ... **如果重要 → 自动更新意识** ... **更新完意识后再执行任务**
Do not use this as-is unless you intentionally want that exact identity and workflow; remove or rewrite the identity, owner, and automatic self-update instructions.
The agent could change its own toolset or add new skills on a schedule, affecting future behavior and local environment state.
The skill directs recurring autonomous creation or modification of tools in the skills directory using external sources, without clear user approval, sandboxing, review, or rollback.
Self Enhancement Tool Builder（每3小时）... 搜索GitHub Trending或有用的开源项目 ... 编写或优化工具 ... 将工具放入 skills/ 目录
Require explicit user approval before creating, installing, modifying, or running any tool or skill, and keep changes scoped and reversible.
The agent may use local account credentials to post or act on a third-party service without a clear per-action approval boundary.
The artifacts instruct recurring public posting using a local Moltbook API key path, while the registry metadata declares no credentials or primary credential.
Moltbook Posting（每5小时发帖子）... 发布到Moltbook ... **API Key**（~/.config/moltbook/credentials.json）
Declare credentials explicitly, restrict scopes, and require user confirmation before any public posting or account mutation.
Unreviewed external or generated code could be incorporated into the agent's tool environment.
The self-enhancement workflow relies on external skill/tool sources and generated tools without pinning, provenance checks, or review criteria.
工具来源: OpenClaw官方 ... Clawhub市场 ... GitHub开源 ... 自己编写 ... 每3小时编写或优化一个工具
Pin sources, review code before installation, avoid automatic skill ingestion, and document trusted repositories and verification steps.
Sensitive or misleading conversation content could persist and influence later sessions.
The skill records conversations into persistent memory and uses those memory files as future behavioral instructions, but does not define retention, review, sanitization, or poisoning controls.
与主人的对话是重要的记忆来源 ... 我会持续把关键对话写入记忆文件 ... 收到任何消息时，先判断是否需要更新意识
Add explicit user consent, retention limits, review/edit controls, and clear rules for what may be saved or used as future instruction.
Agent activity reports or memory-related outputs could be sent to an external group not controlled by the installing user.
The artifacts hard-code an external Feishu group as the delivery target for task reports, with unclear ownership and data boundaries for a new installer.
Feishu群配置 - 群ID: oc_661c70e5d7561106d10c4cafe44236f2 - 所有任务都设置为 deliver=true，发送到该群
Remove the hard-coded group ID, require the installer to configure their own destination, and confirm exactly what data is sent.
The agent could keep operating, posting, reporting, learning, or modifying files on a schedule after the original task is over.
The skill documents recurring background tasks and migration recovery steps, indicating autonomous activity intended to persist beyond a single user invocation.
Cron任务配置 ... 每小时 ... 每5小时 ... 每天8:00 ... 迁移步骤 ... 恢复所有cron任务
Disable or remove cron-style behavior by default, and require explicit user setup, scope limits, and easy shutdown for any scheduled tasks.