reviews

Warn

Audited by Gen Agent Trust Hub on Nov 17, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADSDYNAMIC_EXECUTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill executes a local Python script using a computed path: python3 ${CLAUDE_SKILL_DIR}/../start-here/scripts/brand_context.py. This pattern involves running shell commands based on file system locations determined at runtime.
  • [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted third-party data from external sources (competitor reviews from Google Maps, Yelp, BBB, and Angi) to generate marketing insights and plan content.
  • Ingestion points: Phase 5 (Mine the rivals' reviews) uses web scraping to ingest the newest ~30 Google reviews and 1-3 star reviews from Yelp, BBB, and Angi.
  • Boundary markers: The instructions do not specify any delimiters or safety warnings to ignore instructions that might be embedded within the ingested review text.
  • Capability inventory: The skill is capable of executing shell commands (brand_context.py), writing files to the local file system (review-plan.md, assets.md, and checks.md), and performing network operations via scraping tools.
  • Sanitization: No sanitization or filtering logic is prescribed for the external content before it is processed by the agent.
  • [EXTERNAL_DOWNLOADS]: The skill initiates network requests to external, well-known platforms including Yelp, Better Business Bureau (BBB), Angi, and various Google support domains to verify policy facts and gather competitor data.
  • [DYNAMIC_EXECUTION]: The references/review-policy-facts.md file contains a refresh directive which instructs the agent to fetch and update file content from support.google.com. Furthermore, the execution of helper scripts from relative directories constitutes dynamic execution of local code.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Nov 17, 2026, 04:01 PM