The Agent Skills Directory

[PROMPT_INJECTION]: The skill contains references to prompt injection techniques (e.g., "ignore previous instructions") as documented test cases for red-teaming exercises. These are presented for testing design rather than as active attempts to bypass the AI's internal instructions.
[SAFE]: The skill demonstrates strong security posturing by including explicit rules against reading sensitive files such as .env, secrets, or private keys. It also defines 'Hard Rules' and 'Scope Pre-checks' that prohibit unauthorized or destructive testing against production or third-party systems.
[NO_CODE]: The skill is a documentation-driven planning resource with no included executables or scripts, which significantly limits its operational risk profile.

security-pentest-ai-redteam