The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill's primary function involves ingesting and reacting to data from various untrusted external environments.
Ingestion points: As documented in references/loop-catalog.md, the workflows read data from public news feeds (Reddit, Hacker News, X), social media mentions, customer feedback (NPS surveys, support tickets), and search engine reports.
Boundary markers: The skill does not explicitly instruct the agent to use specific delimiters or boundary markers to isolate untrusted data when processing it.
Capability inventory: Per references/loop-catalog.md and references/loop-guardrails.md, the skill facilitates high-impact actions including ad budget reallocation, outbound email/SMS messaging, and social media publishing.
Sanitization: The risk is mitigated by the 'two-tier action model' defined in references/loop-guardrails.md, which classifies sensitive actions (spending, sending, publishing) as 'Gated' Tier 2 tasks requiring explicit human approval.
[PERSISTENCE]: The skill is designed specifically to maintain agent activity across sessions through scheduled marketing tasks.
Evidence: SKILL.md instructs the agent to use platform-native scheduling tools such as CronCreate and ScheduleWakeup to ensure that marketing loops run continuously on a set cadence (daily, weekly, or monthly).
[COMMAND_EXECUTION]: The skill enables the use of environment-specific commands to automate task execution and scheduling.
Evidence: SKILL.md guides the agent to use tools like /loop and CronCreate to operationalize workflows within the platform's shell environment.

marketing-loops