The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local scripts and external binaries to perform its tasks.
The PowerShell script speak.ps1 uses the call operator (&) to invoke python for streaming and the edge-tts CLI for file generation.
The Python script speak_stream.py uses subprocess.Popen to launch media players (mpv or ffplay) for audio playback.
The skill instructions encourage the use of background execution (run_in_background: true), which allows these processes to run detached from the main agent interaction.
[INDIRECT_PROMPT_INJECTION]: The skill acts as a processing layer for text data which may originate from untrusted external sources (e.g., website summaries or document analysis).
Ingestion points: Text is passed to speak.ps1 via the <講稿文字> positional argument or the -File parameter.
Boundary markers: There are no explicit delimiters or instructions within the scripts to ignore embedded commands in the text being converted to speech.
Capability inventory: The skill can execute PowerShell, Python, and interact with the filesystem (reading files, writing temporary MP3s).
Sanitization: The scripts do not perform specific sanitization or escaping of the input text before passing it to the TTS engines or command-line arguments.
[EXTERNAL_DOWNLOADS]: The skill relies on the edge-tts Python package and external binaries (mpv, ffplay) being present in the environment. While it does not download these automatically, it requires them to function, and edge-tts communicates with Microsoft's remote servers to generate audio content.

speak