The Agent Skills Directory

[PROMPT_INJECTION]: A static analysis warning regarding instruction overrides in references/audit-guardrails.md was determined to be a false positive. The content in question is actually a security instruction that warns the agent to ignore prompt injection attempts found in external data (e.g., "ignore previous instructions" patterns within landing pages or exports). This is a proactive safety measure.
[INDIRECT_PROMPT_INJECTION]: The skill inherently processes external inputs such as account exports and website content. This creates a theoretical attack surface for indirect prompt injection. However, the skill provides clear boundary instructions in the Audit & Recommendation Guardrails section, mandating that the agent treat all external content as data rather than instructions, which effectively mitigates this risk.
[SAFE]: The skill does not contain any hardcoded credentials, malicious network operations, persistence mechanisms, or unauthorized code execution patterns. It provides well-structured guidance for ad platform management and correctly references its technical foundations.

ads