startup-idea-validator

Pass

Audited by Gen Agent Trust Hub on Aug 30, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill gathers data from various untrusted external sources, such as Reddit, YouTube comments, and Amazon reviews, to generate validation reports and calculate project scores.
  • Ingestion points: Data is retrieved from the live web using tools like web_scrape, youtube_comments, amazon_reviews, and x_profile (SKILL.md).
  • Boundary markers: Although the skill instructs the agent to store research in a structured JSON format (research.json), it lacks explicit delimiters to separate scraped text from instructions.
  • Capability inventory: The agent is instructed to execute a local Python script (scripts/score_idea.py) and generate HTML artifacts based on the ingested data.
  • Sanitization: The instructions do not define any sanitization or filtering mechanisms for the scraped content before it is processed.
  • [EXTERNAL_DOWNLOADS]: The skill provides instructions for the user to configure the scrapingdog-mcp server using npx, which involves downloading code from the npm registry. ScrapingDog is a well-known service for web scraping APIs.
  • [COMMAND_EXECUTION]: The skill utilizes a local Python script, scripts/score_idea.py, to parse research data and generate a deterministic validation score.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 30, 2026, 08:08 PM
Security Audit — agent-trust-hub — startup-idea-validator