The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/generate-tts.mjs is vulnerable to command injection through unvalidated user input. It uses execSync to invoke ffmpeg and ffprobe while interpolating variables inside double-quoted strings. The --output-dir argument and segment id fields from the input JSON are used to construct file paths that are then passed to shell commands. An attacker providing a malicious ID such as '; touch /tmp/pwned; ' can execute arbitrary commands on the host system.
[EXTERNAL_DOWNLOADS]: The script scripts/generate-tts.mjs automatically installs the @google/genai npm package if it is missing. While the package is from a well-known service, performing automated installations without user verification is a security concern.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection. It accepts user-controlled content and interpolates it directly into the prompt sent to the AI model without sanitization. Ingestion points include the --text, --director-notes, and --segments inputs. The skill lacks explicit boundary markers or instructions to ignore embedded commands, and it possesses significant capabilities including shell execution.

gemini-tts