The Agent Skills Directory

[SAFE]: The skill implements a 'two-tier action model' that explicitly gates high-risk operations such as spending budget, shifting budget, publishing social posts, or sending emails behind mandatory human approval checkpoints.
[SAFE]: Extensive guardrails are provided in references/loop-guardrails.md, including hard spend caps, volume limits, and strict compliance rules (GDPR, CAN-SPAM, FTC) to ensure automated workflows remain within safe and legal boundaries.
[SAFE]: The skill uses robust state management and idempotency patterns (cursors, handled sets, cooldowns) stored in .agents/loops/ to prevent double-acting or repetitive messaging, which effectively mitigates the risk of loops becoming a nuisance or performing redundant operations.
[SAFE]: For content-sourcing loops such as 'newsjacking' or 'social-listening', the skill defines a mandatory veto list to prevent the agent from acting on sensitive or inappropriate topics (tragedies, crises, unverified sources), significantly reducing the risk of brand reputation damage.
[SAFE]: Platform-specific scheduling features like CronCreate and ScheduleWakeup are used as intended for the primary purpose of managing recurring workflows, with no evidence of unauthorized persistence or malicious command execution.
[SAFE]: The skill instructions emphasize the use of internal identifiers or hashes instead of logging raw Personal Identifiable Information (PII) in loop state or run logs, demonstrating adherence to data privacy best practices.

marketing-loops