The Agent Skills Directory

[NO_CODE]: The skill is composed of markdown reference materials (OWASP Top 10, ASVS 5.0, LLM Top 10) and does not include any executable scripts, binaries, or automated shell commands.
[DYNAMIC_EXECUTION]: Reference files like reference/languages.md and reference/owasp-report.md include examples of dangerous functions such as eval(), exec(), os.system(), and pickle.loads(). These are explicitly documented as "UNSAFE" or "BAD" patterns for educational purposes, providing a knowledge base for identifying vulnerabilities during audits.
[INDIRECT_PROMPT_INJECTION]: While the skill's purpose of reviewing external code involves untrusted data, it includes specific instructions for isolating such content using boundary markers (e.g., LLM01: Prompt Injection), effectively managing the inherent indirect injection surface.

owasp-security