Skill flagged — suspicious patterns detected
ClawHub Security flagged this skill as suspicious. Review the scan results before using.
🗺️ Skill Atlas (技能图谱)
v1.0.4技能图谱。管理、分类、发现、同步 ClawHub / ClawHub CN / SkillHub 上的所有技能。Keywords: 技能管理, skill atlas, 技能图谱, ClawHub, SkillHub, 分类, 发现
⭐ 1· 81·0 current·0 all-time
bywindcat@guowenjiao54
MIT-0
Download zip
LicenseMIT-0 · Free to use, modify, and redistribute. No attribution required.
Security Scan
OpenClaw
Suspicious
medium confidencePurpose & Capability
Name/description match the code: scripts implement searching, installing (via clawhub), vetting and organizing skills. However registry metadata is malformed (Required env vars shown as "[object Object]") and the package claims an install procedure inside SKILL.md while the registry listed no install spec — a metadata/packaging mismatch.
Instruction Scope
Runtime instructions and scripts stay within the declared domain (use clawhub, read scenes.json, enumerate skills, run a vetter). They call PowerShell/clawhub and invoke an external vet script (WORKSPACE/scripts/skill-vetter.py). The scripts do not directly read ~/.ssh/ ~/.aws etc., consistent with the stated limits, but they do execute external CLI commands and rely on an external vetter placed in WORKSPACE, which expands the attack surface.
Install Mechanism
No remote download/install URL in the package itself (code is included in the skill bundle). The SKILL.md lists copy/install steps (copy files into WORKSPACE). The real installation of other skills is delegated to clawhub, which will download remote packages — expected for a manager tool but worth noting: skill-atlas triggers downloads via clawhub.
Credentials
Declared runtime environment needs are minimal in the code (OPENCLAW_WORKSPACE), but the registry shows a corrupted 'Required env vars' field. The skill also depends on an external script (skill-vetter.py) expected from another skill (agent-memory). That implicit dependency and the malformed metadata reduce transparency about required credentials/vars. No direct request for unrelated secrets is present in the included code.
Persistence & Privilege
always:false (good). However the declared workspace_write permission includes editing skills/*/config/scenes.json (i.e., other skills' config files), which is broader than only modifying its own config and could lead to unexpected persistent changes to other skills' load rules. The skill also sets core-resident loading rules in scenes.json which affects session behavior.
What to consider before installing
What to check before installing:
- Metadata/packaging: the registry metadata is inconsistent (Required env vars shows "[object Object]" and the registry claims no install spec while SKILL.md contains one). Treat the package metadata as unreliable until corrected.
- Vetter dependency: the tool repeatedly runs WORKSPACE/scripts/skill-vetter.py (expected to come from agent-memory). Inspect that vetter script's source before installing — it runs automatically during installs/updates and thus has high influence over what gets allowed.
- Write scope: skill-atlas is allowed to write skills/*/config/scenes.json (other skills' config). If you want stricter isolation, restrict that permission or back up scenes.json and other config before enabling the skill.
- External actions: skill-atlas invokes 'clawhub' and PowerShell to search/install/update remote skills. Those remote installs will fetch code from network — ensure you trust the clawhub sources and review the vetter output for each install.
- Command construction: the scripts build and run PowerShell strings (e.g., clawhub install <slug>) without sanitizing inputs; avoid passing untrusted slugs and review any interactive confirmations.
- Practical steps: (1) Inspect WORKSPACE/scripts/skill-vetter.py and any agent-memory artifacts before enabling this skill. (2) Correct/confirm OPENCLAW_WORKSPACE value and back up WORKSPACE/skills/*/config/scenes.json. (3) If uncertain, run the scripts in a sandboxed environment or inspect outputs of vet steps before allowing installs/updates.
Overall: the functionality is plausible for a skill manager, but the metadata issues, implicit reliance on an external vetter, and broad write permission justify caution — review the vetter and configs before enabling automatic operations.Like a lobster shell, security has layers — review code before you run it.
latestvk974kqajs5anv1cbxs6xxfesc184530w
License
MIT-0
Free to use, modify, and redistribute. No attribution required.
Runtime requirements
Env[object Object]