The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process data from untrusted external sources, including search trends, news stories, social media mentions, and competitor website updates, as documented in references/loop-catalog.md. This creates a vulnerability surface for indirect prompt injection, where an attacker could embed malicious instructions in external content to influence the agent's behavior during a loop iteration.
Ingestion points: Multiple loops in references/loop-catalog.md ingest external data, such as the 'newsjacking loop' (news/HN/Reddit/X scanning), 'social-listening loop' (thread/mention scoring), and 'competitor-watch loop' (competitor page snapshots).
Boundary markers: The skill explicitly implements a 'Two-tier action model' in references/loop-guardrails.md, which mandates human approval for all high-risk 'Tier 2' actions (spending money, sending messages, or publishing public content). This acts as a primary defensive layer against automated exploitation.
Capability inventory: The skill orchestrates various capabilities across multiple companion marketing skills (e.g., ads, emails, seo-audit), providing a broad reach for potential downstream effects of an injection.
Sanitization: A 'veto list' is employed in the newsjacking loop to skip sensitive topics, but the skill does not define specific methods for sanitizing or filtering natural language instructions from ingested data.

marketing-loops