The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill defines a "Bad-data tour" that suggests feeding pathological inputs to applications, including SQL injection strings and XSS payloads. * Ingestion points: Example payloads like '; DROP TABLE users; -- and <script>alert(1)</script> are present in SKILL.md and references/tours-catalog.md. * Boundary markers: The payloads are wrapped in backticks or code blocks, but the instructions do not explicitly warn the agent to ignore any commands contained within these examples. * Capability inventory: As a reference catalog, the skill has no direct tools or scripts, but it is designed to be used in conjunction with testing tools like malicious-payload-bank. * Sanitization: There is no logic provided to sanitize or escape these payloads when they are retrieved by the agent.
[OBFUSCATION]: The reference material in references/tours-catalog.md includes Unicode control characters intended for testing purposes. * Evidence: The Right-to-Left Override (RLO, U+202E) and Zero Width Joiner (ZWJ, U+200D) characters are included as specific examples for a "Bad-data tour". * Context: These characters are used to teach the agent to verify how a system handles bidirectional text and complex emoji sequences, representing a documented and benign use case rather than a concealment attempt.

exploratory-tours-reference