The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes untrusted external data from sources such as news trends, social media mentions, and customer reviews to generate content and trigger marketing actions, which creates an indirect prompt injection surface. Evidence: (1) Ingestion points: The loop catalog in references/loop-catalog.md describes scanning news, social listening, and processing customer feedback. (2) Boundary markers: references/loop-guardrails.md defines a 'two-tier action model' that requires human approval for all outbound actions (Tier 2). (3) Capability inventory: The loops orchestrate external skills for emails, social media, ads, and public relations. (4) Sanitization: references/loop-guardrails.md and references/loop-state.md specify hashing for PII and require human review of all drafted content before publication.
[DATA_EXFILTRATION]: The skill demonstrates privacy awareness by explicitly instructing the agent in references/loop-state.md and references/loop-guardrails.md to avoid logging raw PII in state or run logs, recommending internal IDs or hashes instead.
[COMMAND_EXECUTION]: The skill documentation mentions the use of scheduling primitives like CronCreate and ScheduleWakeup for persistent workflow execution. These are aligned with the primary intended functionality of the skill and are accompanied by documented stop conditions and requirements for human checkpoints.

marketing-loops