The Agent Skills Directory

[COMMAND_EXECUTION]: The fpef-analyze.sh script is vulnerable to command injection. User-supplied input from the interactive mode is interpolated into an unquoted bash heredoc (<< EOF), which allows for arbitrary shell command execution via subshell expansion (e.g., entering $(id) as a problem description).
[COMMAND_EXECUTION]: The skill uses insecure string interpolation within a python3 -c call in the get_config_value function. User-controlled variables are placed directly inside single quotes, enabling arbitrary Python code execution if the input contains a single quote followed by malicious code.
[COMMAND_EXECUTION]: The generated diagnostic script (collect-evidence.sh) utilizes sudo to copy system-wide logs from /var/log. This requires elevated permissions and grants the tool access to sensitive system information that may contain private data or PII.
[DATA_EXFILTRATION]: The automated evidence collection process executes the env command and redirects the output to a log file. Environment variables are a primary location for storing sensitive credentials, API keys, and configuration secrets, making this a significant data exposure risk.
[PROMPT_INJECTION]: Untrusted user inputs are directly embedded into generated markdown reports and configuration files without sanitization or the use of boundary markers. This creates a surface for indirect prompt injection if these reports are later parsed by an AI agent.

FPEF Evidence Analyzer