amazon-sorftime-research-mcp-skill
Pass
Audited by Gen Agent Trust Hub on Jul 7, 2026
Risk Level: SAFEPROMPT_INJECTIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONCREDENTIALS_UNSAFE
Full Analysis
- [PROMPT_INJECTION]: The skill processes untrusted external data (Amazon product reviews and keyword lists) through LLM analysis workflows. This creates an indirect prompt injection surface where malicious instructions embedded in reviews could potentially manipulate agent behavior.
- Ingestion points: SKILL.md (via data acquisition methods for listings, reviews, and keywords in the Python implementation logic).
- Boundary markers: None identified in the provided templates for wrapping external content.
- Capability inventory: File system writes and directory creation (os.makedirs, open), and network communication (requests.post).
- Sanitization: No explicit sanitization or filtering of fetched content before LLM processing is shown.
- [EXTERNAL_DOWNLOADS]: The skill logic communicates with external MCP service endpoints: mcp.sorftime.com, mcp.sif.com, and mcp.xiyou.com. These are the primary data providers for the skill's stated purpose.
- [COMMAND_EXECUTION]: The provided Python implementation patterns utilize os.makedirs and file write operations to create directories and save generated reports locally.
- [CREDENTIALS_UNSAFE]: The setup instructions for .mcp.json recommend including API keys directly in the URL query parameters (?key=${SORFTIME_API_KEY}). This is an insecure pattern that can expose sensitive credentials in HTTP access logs, proxy logs, or shell history.
Audit Metadata