The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process data from external, untrusted sources such as news feeds, social media platforms (Reddit, X, HN), and customer feedback systems.
Ingestion points: External data is ingested in multiple loops defined in references/loop-catalog.md, specifically 'The newsjacking loop', 'The social-listening loop', 'The community-engagement loop', and 'The voice-of-customer loop'.
Boundary markers: The skill implements strong logical boundaries by classifying actions into two tiers (Tier 1 safe vs Tier 2 gated). All Tier 2 actions, such as publicly visible posts or actions involving spending, require a mandatory human checkpoint.
Capability inventory: The skill orchestrates channel-specific tools for advertising, social media, and email, and uses platform-native scheduling primitives like CronCreate and ScheduleWakeup. It writes state information and logs to the .agents/loops/ directory.
Sanitization: All generated outputs intended for external use are staged as drafts for human review and approval, preventing the automated execution of potentially malicious instructions embedded in the source data.
[COMMAND_EXECUTION]: The skill instructs the agent to use platform-specific command-line interfaces (CLIs) for marketing tools (e.g., ad platform CLIs) and platform-native scheduling commands such as CronCreate, ScheduleWakeup, and /loop to manage the cadence and data retrieval for the marketing loops. These are legitimate, scoped operations intended for the skill's primary function.

marketing-loops