The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill provides instructions in rules/transcribe-captions.md to download and install the Whisper.cpp binary using the @remotion/install-whisper-cpp package, followed by its execution for transcription tasks.
[COMMAND_EXECUTION]: In rules/transcribe-captions.md, the skill includes code examples that utilize child_process.execSync to run ffmpeg shell commands for audio conversion.
[EXTERNAL_DOWNLOADS]: The skill facilitates downloading external assets and binaries. This includes fetching Lottie animations from LottieFiles and downloading Whisper.cpp models.
[DATA_EXFILTRATION]: Several rules (e.g., rules/calculate-metadata.md, rules/compositions.md, rules/lottie.md) demonstrate fetching data from URLs provided via component properties (e.g., props.dataUrl, props.videoId). This pattern creates a potential vector for data exfiltration if an attacker-controlled URL is supplied to the agent.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted external data, such as .srt subtitle files, JSON caption files, and video metadata from remote URLs. This creates a surface for indirect prompt injection where malicious instructions embedded in these files could influence the agent's behavior during processing.
Ingestion points: rules/import-srt-captions.md (SRT parsing), rules/display-captions.md (JSON fetching), rules/calculate-metadata.md (metadata fetching).
Boundary markers: Absent in code examples; the agent is instructed to process the data directly.
Capability inventory: Includes network fetching (fetch), command execution (execSync), and binary installation.
Sanitization: No explicit sanitization or validation logic is shown for the content of the external files being processed.

remotion-best-practices