The Agent Skills Directory

[COMMAND_EXECUTION] (HIGH): The skill provides explicit code snippets to execute shell commands through os.system(), which can be used to read sensitive files (e.g., 'cat flag.txt') or gain shell access via breakpoint().
[PROMPT_INJECTION] (HIGH): The skill's stated purpose is to 'Escape Python sandbox restrictions' and bypass 'filtered input challenges,' which aligns with techniques used to override AI safety guardrails and ignore system constraints.
[REMOTE_CODE_EXECUTION] (HIGH): It details multiple methods for arbitrary code execution using eval(), exec(), and sophisticated Python class hierarchy manipulation to regain access to blocked built-in functions.
[OBFUSCATION] (MEDIUM): The skill teaches the agent how to use character encoding (chr()) and string concatenation to hide malicious intent from simple static keyword filters.

pyjail