The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data from sources like news feeds, competitor websites, and community forums.
Ingestion points: Multiple loops in references/loop-catalog.md (e.g., newsjacking, social-listening, competitor-watch) scan the open web, social media, and forums.
Boundary markers: The skill explicitly implements a 'two-tier action model' in references/loop-guardrails.md, mandating that all 'Tier 2' actions (sending, spending, publishing) be held in a staging queue for human approval.
Capability inventory: The skill orchestrates channel-specific tools for email, social media, advertising, and public relations.
Sanitization: Security is maintained through human review checkpoints, explicit veto lists for sensitive topics, and compliance with CAN-SPAM/GDPR/FTC regulations. These controls effectively mitigate the risk of malicious instructions in external data being executed without oversight.
[DATA_EXPOSURE]: The skill manages internal state and configuration files within the .agents/ directory.
Evidence: references/loop-state.md defines state persistence in .agents/loops/ using JSON and log files. This is a standard practice for maintaining idempotency and does not involve unauthorized data exposure or exfiltration of sensitive credentials.

marketing-loops