marketing-loops

Pass

Audited by Gen Agent Trust Hub on Jul 27, 2026

Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection because it processes untrusted data from multiple external sources.
  • Ingestion points: The newsjacking, social-listening, community-engagement, voice-of-customer, and brand-mention loops ingest content from news sites (HN/Reddit/X), social media platforms, community forums (Slack/Discord), and customer feedback tools (surveys, tickets, reviews).
  • Boundary markers: The instructions lack explicit requirements for using delimiters or 'ignore' instructions when interpolating external content into prompts.
  • Capability inventory: The loops orchestrate skills like emails, social, public-relations, and ads, which are capable of network operations (sending/publishing) and file modifications.
  • Sanitization: While the loop-guardrails.md mandates PII hashing, it does not specify sanitization or escaping techniques to neutralize malicious instructions embedded in the external data.
  • [COMMAND_EXECUTION]: The skill facilitates the creation and scheduling of recurring tasks (loops), which involves dynamic logic generation.
  • The references/loop-template.md and SKILL.md provide a framework for the agent to author new 'loops' (repeatable workflows). These are then scheduled for execution using platform-specific primitives (e.g., CronCreate, ScheduleWakeup in Claude Code).
  • Generating and scheduling logic from templates constitutes a form of dynamic execution, although it is constrained to the intended marketing domain.
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 27, 2026, 12:30 AM
Security Audit — agent-trust-hub — marketing-loops