brida-reflex-semantic-content-blocking
Pass
Audited by Gen Agent Trust Hub on Sep 21, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill analyzes and classifies potentially untrusted content extracted from web pages to perform semantic ad-blocking classification. Ingestion points: Page context and candidate text as defined in SKILL.md and fixtures in references/custom-reflex.json. Boundary markers: references/playbook.md instructs the agent to exclude sensitive data like credentials or browsing history from the classification state. Capability inventory: The skill is restricted to providing classification recommendations and lacks authority for browser mutations, network blocking, or account changes. Sanitization: No explicit logic for filtering or sanitizing input content is provided within the skill instructions.
- [EXTERNAL_DOWNLOADS]: The skill instructs the agent to consult live configuration and availability metadata from the vendor's infrastructure (such as /reflex/llms.txt). These references target official vendor resources for integration verification and do not involve untrusted third-party code.
Audit Metadata