The Agent Skills Directory

[COMMAND_EXECUTION] (HIGH): Shell command injection vulnerability in scripts/clone-voice.sh. The script constructs a curl command string using variables like $VOICE_NAME, $DESCRIPTION, and $file without any sanitization or shell escaping, and then executes the resulting string using eval. An attacker providing a malicious name (e.g., containing ; whoami) can execute arbitrary shell commands.
Evidence: RESPONSE=$(eval "$CURL_CMD") in scripts/clone-voice.sh.
[COMMAND_EXECUTION] (MEDIUM): Potential python code injection in examples/professional-cloning/clone-workflow.sh. Filenames found on the local filesystem are directly interpolated into a python code string passed to python -c. A filename containing a single quote could terminate the intended string and execute malicious Python code.
Evidence: duration=$(python -c "from pydub import AudioSegment; print(len(AudioSegment.from_file('$file')) / 1000.0)" ...) in examples/professional-cloning/clone-workflow.sh.
[COMMAND_EXECUTION] (MEDIUM): Runtime script generation and execution in examples/professional-cloning/clone-workflow.sh. The workflow generates a monitoring script (monitor_${VOICE_ID}.sh) and uses chmod +x to make it executable. This dynamic code creation increases the attack surface and risk of persistence or unauthorized execution if the generation logic is compromised.
Evidence: cat > "monitor_${VOICE_ID}.sh" <<'MONITOR_SCRIPT' ... followed by chmod +x "monitor_${VOICE_ID}.sh" in examples/professional-cloning/clone-workflow.sh.

voice-processing