video
Fail
Audited by Gen Agent Trust Hub on Jul 2, 2026
Risk Level: HIGHCOMMAND_EXECUTIONREMOTE_CODE_EXECUTIONEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: Multiple scripts (
scripts/tts-providers/gemini.mjs,scripts/tts-providers/supertone.mjs, andscripts/tts-providers/supertonic.mjs) useexecSyncto execute shell commands for media processing (e.g.,ffmpeg). These commands are constructed using string concatenation with theidfield from the user-provided timeline JSON. A malicious timeline file can include shell metacharacters in theidfield to execute arbitrary commands. - [REMOTE_CODE_EXECUTION]: The command injection vulnerability mentioned above provides a direct path for remote code execution. An attacker who can influence the content of a timeline or script file processed by the agent can gain control over the execution environment.
- [EXTERNAL_DOWNLOADS]: The skill includes scripts (
scripts/ensure-remotion.mjsandsetup-remotion.sh) that automatically download and install numerous NPM/pnpm packages. This pattern introduces significant supply chain risk, as the skill depends on the security and availability of external registries and third-party code at runtime. - [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection. It is designed to ingest and process complex JSON data (
timeline.json,script.json) from the workspace. If this data is sourced from an untrusted location (e.g., a pull request or an external repository), it can manipulate the agent's behavior or exploit the command execution vulnerabilities. - [DATA_EXPOSURE]: The skill relies on sensitive environment variables (
GEMINI_API_KEY,SUPERTONE_API_KEY) for its primary functionality. While it does not hardcode them, it promotes a workflow where these secrets are actively used in a potentially compromised environment.
Recommendations
- AI detected serious security threats
Audit Metadata