The Agent Skills Directory

[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection because it lacks security boundaries when processing untrusted user data alongside sensitive brand information.
Ingestion points: The skill accepts user-provided input for 'Key message or topic' and 'Competitor video references' as defined in SKILL.md.
Boundary markers: Absent. There are no instructions to use delimiters (e.g., XML tags or triple quotes) or to ignore embedded instructions within user-provided strings.
Capability inventory: The skill performs read operations on local files including brand profiles, compliance rules, and agency SOPs located in ~/.claude-marketing/ and skills/context-engine/ as detailed in the process section of SKILL.md.
Sanitization: Absent. The skill does not validate or sanitize external input before interpolating it into the prompt for script generation.

video-script