The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill framework is designed to process untrusted data from various external sources, which could be used to attempt to influence the agent's behavior.
Ingestion points: references/loop-catalog.md identifies external data sources including search console data, social media threads (Reddit/X), news feeds, and customer reviews.
Boundary markers: references/loop-guardrails.md implements a two-tier action model that mandates human approval for any 'Gated' action such as spending budget or publishing content.
Capability inventory: The loops described in references/loop-catalog.md are capable of drafting email sequences, social media posts, and advertising adjustments.
Sanitization: Risk is mitigated by requiring manual verification of all staged outputs before they are executed in live environments.
[SAFE]: The skill provides instructions for scheduling tasks using mechanisms like CronCreate and system cron, which are standard for the skill's purpose of establishing repeatable workflows.

marketing-loops