competitive-intel

Pass

Audited by Gen Agent Trust Hub on Nov 17, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from external websites, including competitor homepages, ad libraries, and review platforms, to synthesize market gaps.
  • Ingestion points: External content is fetched during Phase 2 and Phase L2 using tools such as Firecrawl and WebFetch from arbitrary competitor domains.
  • Boundary markers: There are no explicit instructions to use secure delimiters or "ignore embedded instructions" warnings when passing evidence packs to the synthesis phase, although verbatim quotes are requested.
  • Capability inventory: The skill possesses significant capabilities, including network access (WebSearch, WebFetch, Firecrawl), parallel sub-agent execution via the Task tool, and file-writing permissions to the ./brand/ and ./campaigns/ directories.
  • Sanitization: The instructions lack specific sanitization or escaping requirements for content fetched from the web before it is processed by the synthesis logic.
  • [COMMAND_EXECUTION]: The skill instructs the agent to execute a local Python script (brand_context.py) located in a relative directory to load the brand context at runtime.
  • [EXTERNAL_DOWNLOADS]: The skill utilizes several well-known third-party services for search and data extraction, specifically Firecrawl (firecrawl.dev), Exa (exa.ai), and ScrapeCreators (scrapecreators.com), which require user-provided API keys.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 17, 2026, 04:01 PM