The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted lead data and interpolates it into natural language templates for voice call scripts and WhatsApp messages, which could be exploited if input data contains malicious instructions.
Ingestion points: Lead data (names, preferences, property types) is ingested from external JSON files in scripts/voice-campaign.js and scripts/client-match.js via the --input or --leads flags.
Boundary markers: The templates do not use boundary markers or delimiters to isolate lead-supplied strings from the rest of the generated prompt/script.
Capability inventory: The skill generates configurations for automated voice agents (Retell.ai) and writes output to the local file system using fs.writeFileSync.
Sanitization: The replaceVariables function in scripts/voice-campaign.js performs direct string substitution using regular expressions without any sanitization, filtering, or escaping of the lead data.

dubai-realestate