The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill metadata (SKILL.md) specifies the download of binary runtime archives and voice models from a third-party GitHub repository (k2-fsa/sherpa-onnx) across macOS, Linux, and Windows platforms.- [REMOTE_CODE_EXECUTION]: The skill implementation downloads pre-compiled binary executables from a remote source which are then executed by the agent via a provided wrapper script.- [COMMAND_EXECUTION]: The Node.js wrapper script (bin/sherpa-onnx-tts) utilizes spawnSync to execute the downloaded sherpa-onnx-offline-tts binary with arguments including model paths and user-provided text.- [DYNAMIC_EXECUTION]: The wrapper script dynamically modifies the runtime environment at execution time by prepending the library directory to search paths (LD_LIBRARY_PATH on Linux, DYLD_LIBRARY_PATH on macOS, and PATH on Windows) to facilitate the loading of shared object dependencies.- [INDIRECT_PROMPT_INJECTION]: The skill processes external text input which is passed directly to a subprocess for synthesis.
Ingestion points: The text argument in bin/sherpa-onnx-tts allows for arbitrary text input.
Boundary markers: Absent; the script does not use delimiters or instructions to prevent the processing of embedded directives.
Capability inventory: Subprocess execution via spawnSync in bin/sherpa-onnx-tts across all files.
Sanitization: Absent; the script performs basic string joining and trimming but no sanitization of input that might impact the underlying binary execution.

sherpa-onnx-tts