The Agent Skills Directory

[COMMAND_EXECUTION]: Shell Command Injection via Variable Interpolation.
Evidence: In SKILL.md (Steps 1, 3, and 4), the variable ${INPUT} is interpolated into a Python heredoc block <<PYEOF. Because the heredoc delimiter is unquoted, the shell evaluates any command substitutions (e.g., `command` or $(command)) within the input string before the Python interpreter is launched.
Evidence: In SKILL.md (Step 2), the command python3 -c "... s = '${INPUT}' ..." allows for Python code injection if the input contains a single quote ('), enabling an attacker to escape the string context and execute arbitrary Python logic.
[COMMAND_EXECUTION]: Argument Injection via echo.
Evidence: In SKILL.md (Step 2), the use of echo "$INPUT" | base64 is susceptible to flag injection if the user input starts with a hyphen (e.g., -e), potentially leading to unexpected shell behavior.

encoding-toolkit