The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to process data from untrusted external sources like Reddit, X (Twitter), news sites, and community forums, which could contain instructions designed to manipulate the agent's behavior.
Ingestion points: The Newsjacking, Social-listening, and Community-engagement loops in references/loop-catalog.md are designed to scan external platforms for trending topics and mentions.
Boundary markers: The skill provides a comprehensive safety framework in references/loop-guardrails.md which mandates that all "Tier 2" actions (sending, spending, publishing) be held in a staging queue for human approval.
Capability inventory: The loops orchestrate other skills such as social, public-relations, and emails to draft content for distribution.
Sanitization: references/loop-catalog.md defines a specific veto list to exclude sensitive or controversial topics, and the loop-guardrails document requires human review before any drafted content is actually sent or published.
[COMMAND_EXECUTION]: The skill provides guidance on how to execute shell-level commands for scheduling tasks.
Evidence: SKILL.md includes instructions for using cron or platform-specific scheduling commands (like CronCreate) to run the marketing workflows on a cadence.
[PERSISTENCE]: The skill's primary function is to establish long-running, recurring automation that survives between user sessions.
Evidence: The core design focuses on "always-on" loops that use cron schedules or wake-up triggers to maintain a persistent presence and operational cadence.

marketing-loops