The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill directs the agent to perform searches and process data from external websites, which introduces a surface for indirect prompt injection.
Ingestion points: The "Research Step" section in SKILL.md instructs the agent to analyze competitor landing pages and pull customer review language from sites such as Trustpilot.fr and Avis Vérifiés.
Boundary markers: The instructions do not define specific delimiters or include warnings to the agent to ignore potentially malicious instructions embedded within the retrieved external content.
Capability inventory: The skill is designed for text generation and analysis; it does not possess capabilities for command execution, filesystem modification, or unauthorized data exfiltration that would allow an injection to escalate into a system compromise.
Sanitization: No methods for sanitizing, escaping, or filtering the content retrieved from search results are provided in the instructions.

french-copywriting