brand-voice

Pass

Audited by Gen Agent Trust Hub on Nov 17, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill uses Firecrawl and Exa APIs as optional-keys to perform web scraping for brand research. It also utilizes WebFetch for basic page retrieval. These are standard tools for its stated purpose of brand voice extraction and are documented appropriately.
  • [COMMAND_EXECUTION]: Executes a local python script brand_context.py to load existing brand files (MARKETING.md, etc.). This is a scoped execution within the agent's provided environment to provide necessary context for the skill.
  • [INDIRECT_PROMPT_INJECTION]: As a tool that ingests external URLs and text (Auto-Scrape and Extract modes), it is theoretically susceptible to indirect prompt injection. However, the skill implements strict boundary controls by using a dedicated subagent architecture for evidence gathering and requiring a 'Voice Test Loop' for human verification before saving any profile. It also specifically loads ai-tells.md to flag and avoid common AI-generated patterns.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 17, 2026, 04:23 PM