analytics-tracking
Pass
Audited by Gen Agent Trust Hub on Feb 15, 2026
Risk Level: LOWPROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- PROMPT_INJECTION (LOW): Indirect Prompt Injection surface via local file ingestion.
- Ingestion points: The skill instructions in
SKILL.mdcommand the agent to read.claude/product-marketing-context.mdif it exists. - Boundary markers: Absent. The agent is told to "use that context" and only ask for information not already covered.
- Capability inventory: Based on the provided files, the agent generates implementation plans, naming conventions, and JavaScript code snippets.
- Sanitization: Absent. If an attacker can modify the
.claude/product-marketing-context.mdfile, they can inject instructions to influence the agent's output or recommendations. - EXTERNAL_DOWNLOADS (LOW): The skill documentation includes references to external script sources.
- Evidence:
references/gtm-implementation.mdcontains a snippet for the Facebook Pixel pointing tohttps://connect.facebook.net/en_US/fbevents.js. - Context: While these are standard industry practices for marketing analytics, they represent a dependency on external code which is presented to the user for implementation.
Audit Metadata