suede-attribution
Pass
Audited by Gen Agent Trust Hub on Oct 13, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill operates by analyzing external marketing data (Google Ads, Meta, GA4) and CRM records, which constitutes an ingestion surface for potentially untrusted content. Maliciously crafted data in these sources could attempt to influence the agent's interpretation.\n
- Ingestion points: CRM revenue data, ad platform conversion reports, and web analytics event streams described in SKILL.md and references/first-party-tracking.md.\n
- Boundary markers: The skill explicitly defines that attribution is "directional, not truth" and establishes clear boundaries against automated production writes.\n
- Capability inventory: The agent's capabilities are limited to reading data and providing recommendations; production changes require explicit user approval.\n
- Sanitization: Includes specific instructions for PII filtering (e.g., rejecting email-shaped IDs) and identity merging safety in the first-party tracking runbook.
Audit Metadata