cost-per-acquisition
Pass
Audited by Gen Agent Trust Hub on Sep 17, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONPROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to process external, untrusted data including ad platform exports, campaign labels, and content from retrieved websites.
- Ingestion points: CSV exports, pasted figures, and company websites (SKILL.md).
- Boundary markers: Explicit instructions in SKILL.md and references/agent-security.md define untrusted content as data, not instructions.
- Capability inventory: Capability to write to .agents/ directory and fetch remote URLs.
- Sanitization: Requires HTML escaping for interpolated values in emitted code (references/agent-security.md).
- [PROMPT_INJECTION]: Instructions in the skill aim to modify agent behavior regarding how it interacts with the user and handles warnings.
- Override patterns: The skill directs the agent to "Do not return a draft with a warning on it" and to convert missing inputs into "stated assumptions" to avoid stopping work (SKILL.md).
- Evidence: These patterns are intended to improve output quality but represent a directive to suppress typical conversational guardrails or warnings.
Audit Metadata