Evidence for AI clean decision
- The package is a DSH skills and MCP marketplace plugin with JavaScript entry points and no declared lifecycle scripts.
- On plugin startup, the background warmup fetches catalogue data for the first panel open.
- The added source code requests catalogue data from named upstream APIs such as skillsmp.com and ai.trusttools.cn.
- src/verify.js implements a safety scan of staged skill bundles; its shell-related patterns are detection rules, not commands it runs.
Evidence against
- The plugin creates configured directories, mounts skill and MCP providers, and may connect configured MCP servers as part of its stated function.
- The package uses network requests and curl/Python subprocess transports for catalogue and download operations; these paths appear tied to user-facing marketplace features.
Behavioral surface
SourceChild process · Dynamic require · Environment variables · Filesystem · Network
Supply chainHigh entropy strings · URL strings
ManifestNo manifest risk signals triggered.
Scanned 20 files · 333 KB of source
16 external domains referenced
- ai.trusttools.cn
- api.github.com
- api.skillhub.cn
- browse.sh
- cdn.jsdelivr.net
- clawhub.ai
- codeload.github.com
- data.jsdelivr.com
- github.com
- lobehub.com
- market.lobehub.com
- modelscope.cn
- raw.githubusercontent.com
- registry.modelcontextprotocol.io
- skillsmp.com
- www.skills.sh