The Agent Skills Directory

[PROMPT_INJECTION]: The skill contains defensive guidelines in references/audit-guardrails.md and references/creative-research-automation.md that instruct the agent to ignore directives embedded in fetched data (e.g., 'ignore previous instructions'). While these phrases triggered a static detector, they are used here as security best practices to prevent the agent from being influenced by malicious instructions in untrusted external data.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted external content, such as competitor ad copy and product reviews. It acknowledges this vulnerability surface and provides mandatory evidence chains and behavioral rules (e.g., in references/audit-guardrails.md) to ensure that external data is treated solely as informational and never as a source of instructions.
[SAFE]: The skill follows security best practices for credential management by using clearly labeled placeholders (e.g., AW-XXXXXXXXX, YOUR_PIXEL_ID, YOUR_PARTNER_ID) in all implementation templates and code snippets.
[SAFE]: All referenced external resources and tool integrations point to well-known services (Google, Meta, LinkedIn, TikTok) or established public repositories on GitHub without any signs of typosquatting or malicious redirection.

ads