remotion-best-practices
Warn
Audited by Gen Agent Trust Hub on Feb 19, 2026
Risk Level: MEDIUMEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONCREDENTIALS_UNSAFE
Full Analysis
- [EXTERNAL_DOWNLOADS] (MEDIUM): The
rules/transcribe-captions.mdfile contains instructions to download the Whisper.cpp binary and language models to the local project directory. Although this is a functional requirement for the transcription feature, downloading and executing binaries that are not managed through standard package registries (and are from sources outside the system's trusted list) poses a medium risk. - [COMMAND_EXECUTION] (MEDIUM): Several rule files (
rules/ffmpeg.md,rules/transcribe-captions.md,rules/voiceover.md) provide templates for executing shell commands viabunx,npx, or Node.jsexecSyncto perform operations like video re-encoding and speech-to-text processing. - [CREDENTIALS_UNSAFE] (LOW): The
rules/voiceover.mdrule instructs the agent to request anELEVENLABS_API_KEYfrom the user and store it in a.envfile. While it uses placeholder text, this pattern creates a surface for the accidental exposure of sensitive API secrets if the environment or configuration files are mishandled. - [DATA_EXFILTRATION] (LOW): Network operations using
fetch()are present inrules/voiceover.md,rules/compositions.md, andrules/display-captions.mdto communicate with external APIs (ElevenLabs) or retrieve remote assets. These are used for the skill's primary purpose but represent a path for data flow to non-whitelisted domains. - [INDIRECT_PROMPT_INJECTION] (LOW): The skill is designed to process external, potentially untrusted media files and data structures.
- Ingestion points: Video/audio metadata retrieval in
rules/get-video-duration.mdandrules/get-video-dimensions.md, and subtitle parsing inrules/display-captions.md. - Boundary markers: Absent; the skill does not suggest delimiters or warnings when processing external data.
- Capability inventory: File system access (
fs.writeFileSync), network requests (fetch), and shell execution (execSync). - Sanitization: No explicit sanitization or validation of the content of external files (like captions or video metadata) is documented before processing.
Audit Metadata