The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill documents methods for ingesting untrusted external audio data from URLs and microphone input, which represents a potential attack surface for indirect prompt injection if the resulting data is later processed by an LLM without proper sanitization.
Ingestion points: The skill utilizes decodeAudioData (in audio.md) to load remote audio files and AudioRecorder (in recording.md) to capture microphone input.
Boundary markers: There are no explicit boundary markers or instructions to ignore embedded signals within the provided code patterns.
Capability inventory: The skill provides capabilities for writing to the local filesystem via AudioRecorder and performing network read operations via StreamerNode and decodeAudioData.
Sanitization: No specific audio data sanitization or validation logic is included in the provided patterns.
[EXTERNAL_DOWNLOADS]: The documentation includes examples that demonstrate fetching audio assets from remote URLs.
Evidence: audio.md contains decodeAudioData('https://example.com/audio.mp3') and playback.md includes streamer.initialize('https://example.com/stream.m3u8'). These are noted as standard placeholders for developer implementation.
[COMMAND_EXECUTION]: The skill provides instructions for package installation and code generation using common CLI tools.
Evidence: worklets.md recommends running npm install react-native-worklets and describes using npx rn-audioapi-custom-node-generator create -o ./ to scaffold custom native nodes.
[DYNAMIC_EXECUTION]: The skill leverages Audio Worklets to execute custom JavaScript logic on specialized high-performance runtimes.
Evidence: Use of WorkletNode, WorkletSourceNode, and WorkletProcessingNode in worklets.md to run procedural synthesis and real-time effects.

audio