creative-brief-selector

Pass

Audited by Gen Agent Trust Hub on Aug 10, 2026

Risk Level: SAFE
Full Analysis
  • [SAFE]: The skill consists entirely of Markdown-based instructional guidelines, templates, and a bank of curated reference data. No scripts, binaries, or automated command executions are present within the skill files.
  • [PROMPT_INJECTION]: The skill contains a standard surface for indirect prompt injection through its reference discovery process. The agent is instructed to visit and analyze external websites to update the reference bank.
  • Ingestion points: External websites are visited during the 'Pull references' phase in references/01-process.md and references/reference-bank/README.md.
  • Boundary markers: The skill employs a structured brief template in references/02-brief-template.md which effectively isolates external information into specific fields like 'Live reference sites'.
  • Capability inventory: The agent uses file reading/writing for bank maintenance and web browsing for reference discovery.
  • Sanitization: External site content is curated and summarized by the agent into 'one-line why' statements, which provides a layer of interpretation between the raw site data and the final output.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 10, 2026, 09:41 AM
Security Audit — agent-trust-hub — creative-brief-selector