The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill provides templates for generating adversarial payloads, including SQL injection, cross-site scripting (XSS), and template injection. This identifies a surface where the agent generates potentially malicious content; however, this behavior is explicitly for the purpose of identifying vulnerabilities in the system under test.
Ingestion points: Instructions for generating dangerous strings in SKILL.md.
Boundary markers: The skill suggests testing through agreed seams but does not specify delimiters for the generated payloads.
Capability inventory: Instructions involve string construction and execution of regex timing tests in child processes.
Sanitization: The skill recommends using grep and od to verify the integrity of generated files and detect unintended characters.
[OBFUSCATION]: The instructions include techniques for generating homoglyphs (e.g., Cyrillic characters mimicking Latin ones), invisible characters (U+2028, U+2029), and bidirectional overrides (U+202E) to test a system's resilience against obfuscation-based attacks.
[DYNAMIC_EXECUTION]: The skill suggests using String.fromCodePoint to programmatically generate special characters and recommends executing regex performance tests within a child process with a hard timeout to detect catastrophic backtracking.
[COMMAND_EXECUTION]: The skill mentions using standard utilities like grep and od to audit source files for non-ASCII characters or invisible control points.

test-synthesizer