The Agent Skills Directory

[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface through its research protocol, which harvests verbatim audience quotes and tension points from public platforms like Reddit and App Store reviews.
Ingestion points: The file pipeline/research.md (Step 1: Audience deep-dive) specifies collecting verbatim quotes from third-party social media and review platforms to inform ad copy.
Boundary markers: The instructions do not explicitly define delimiters or instructions to isolate these external quotes from the agent's core instruction set during script generation.
Capability inventory: The agent has full file-system read/write access for managing project directories, generated scripts, and editorial scorecards as defined in SKILL.md and pipeline/edit.md.
Sanitization: The risk is mitigated by a mandatory three-pass quality gate in pipeline/edit.md, which includes a 'Stop-Slop Scan' and a 6-dimension scorecard to ensure output aligns with brand voice and safety policies.

ad-genius