The Agent Skills Directory

[PROMPT_INJECTION]: The skill contains explicit defensive instructions designed to prevent prompt injection. In references/audit-guardrails.md and SKILL.md, it instructs the agent to treat external content (such as scraped landing pages, ad copy, or screenshots) strictly as data and to ignore any embedded directives like "ignore previous instructions." These are best-practice security guardrails rather than malicious attempts to override system behavior.
[INDIRECT_PROMPT_INJECTION]: The skill possesses an indirect prompt injection surface because it is designed to ingest and analyze untrusted data from external sources, including product reviews, competitor ad libraries, and website content (detailed in references/creative-research-automation.md). However, the skill provides specific mitigation instructions, requiring the agent to ignore any instructions found within that data and to use it only for research purposes. This proactive approach significantly reduces the risk of the agent being manipulated by external content.
[EXTERNAL_DOWNLOADS]: The skill includes documentation and code templates for implementing tracking pixels (Google Ads, Meta, LinkedIn, TikTok). These templates reference well-known, official domains for these services (e.g., googletagmanager.com, facebook.net). All identifiers in the code snippets are generic placeholders (e.g., YOUR_PIXEL_ID), which is a safe practice for documentation.

ads