iconfont-downloader
Pass
Audited by Gen Agent Trust Hub on Sep 28, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONDATA_EXFILTRATIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill interacts with the well-known service iconfont.cn (Alibaba) to fetch icon assets and metadata. It also encourages the installation of well-known browser automation libraries like Playwright and Puppeteer from official registries.
- [COMMAND_EXECUTION]: The README.md and IMPLEMENTATION_GUIDE.md provide standard shell commands for package installation (
pnpm add) and skill setup (pnpm skill:install), which are typical for developer-oriented tools. - [DATA_EXFILTRATION]: The skill prompts the user for credentials to access the Iconfont platform. These credentials are used in memory to establish an authenticated session with the official iconfont.cn login endpoint and are not stored persistently, according to the implementation details in scripts/index.js.
- [DYNAMIC_EXECUTION]: The script uses dynamic
import()calls in scripts/index.js to detect the presence of Playwright or Puppeteer on the host system. It also utilizespage.evaluate()to execute DOM manipulation scripts within the controlled browser context to extract icon information. - [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted search results from the Iconfont web platform, which is a potential surface for indirect prompt injection.
- Ingestion points: External icon names and author metadata from iconfont.cn (scripts/index.js).
- Boundary markers: The skill uses structured Markdown tables to present choices and requires the user to select icons by numeric index, preventing the agent from acting autonomously on external text.
- Capability inventory: The skill has file-writing capabilities (saving SVGs) and network access to Iconfont APIs.
- Sanitization: Filenames are processed through a
sanitizeFileNamefunction to prevent path traversal attacks.
Audit Metadata