positioning-angles

Pass

Audited by Gen Agent Trust Hub on Nov 17, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from external sources including competitor websites, pricing pages, and review platforms such as Reddit, G2, Capterra, and Yelp. This data is used to generate marketing angles and offers, which creates an attack surface for instructions embedded in those external sources.
  • Ingestion points: Phase 2 of SKILL.md identifies 3-5 competitors and fetches their site content and reviews using WebSearch, WebFetch, and external APIs.
  • Boundary markers: The skill provides no explicit technical prompt delimiters for external content, though it requires the agent to build a 'fact sheet' and verify every line against sourced facts (the 'claim gate').
  • Capability inventory: The skill can write files to the ./brand/ and ./campaigns/ directories, perform network searches, and execute local scripts.
  • Sanitization: The instructions explicitly require the agent to filter bold claims against verified facts, but do not provide specific escaping for the raw data ingested.
  • [DYNAMIC_EXECUTION]: The skill executes a local Python script (brand_context.py) located in a relative path (../start-here/scripts/) to gather project context. This is a controlled execution of a local utility script.
  • [EXTERNAL_DOWNLOADS]: The skill is configured to use well-known external services (Exa and Firecrawl) to perform web scraping and competitor research. These services are used for data retrieval related to the skill's primary purpose.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 17, 2026, 04:01 PM