The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes content from untrusted external sources (news sites, social media feeds, competitor pages), which presents a surface for indirect prompt injection where malicious instructions could be embedded in the processed data.
Ingestion points: The newsjacking-loop, social-listening-loop, and competitor-watch-loop in references/loop-catalog.md describe scanning sources like HN, Reddit, X, and competitor websites.
Boundary markers: The framework in references/loop-guardrails.md establishes a mandatory two-tier action model where all 'Tier 2' actions—such as spending money, sending emails, or publishing—require a human checkpoint by default.
Capability inventory: The skill orchestrates other agents (ads, emails, social, etc.) and utilizes scheduling primitives like CronCreate to manage workflows.
Sanitization: The newsjacking-loop contains a specific veto list to automatically discard stories involving crises, tragedies, or unverified sources, and the framework consistently defaults to staging drafts for review rather than autonomous publishing.

marketing-loops