The Agent Skills Directory

[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection. The script scripts/analyze_recording.py parses a metadata.json file located in the user-provided directory and outputs fields like title, appName, and speakers directly into the agent's context. An attacker could craft a directory containing a malicious metadata.json with instructions intended to hijack the agent's session.
Ingestion points: scripts/analyze_recording.py (reading metadata.json via json.loads).
Boundary markers: None present; data is interpolated directly into the printed report.
Capability inventory: scripts/analyze_recording.py performs file reading, temporary directory management (tempfile, shutil.rmtree), and executes external processes via subprocess.run.
Sanitization: No sanitization or escaping is performed on the metadata values before they are printed.
[COMMAND_EXECUTION]: The analysis script scripts/analyze_recording.py uses the subprocess module to call ffmpeg and ffprobe. It dynamically constructs the arguments based on the directory path provided by the user and the files discovered within that directory. Although it uses list-based arguments which mitigate shell injection, it remains a point of interaction with the system's execution environment.
[EXTERNAL_DOWNLOADS]: The skill requires the installation of multiple external Python dependencies (numpy, soundfile, scipy, pyloudnorm, pesq, pystoi, librosa) and system-level tools (ffmpeg, ffprobe). These are well-known and standard libraries for signal processing and audio analysis, but they represent a broad external dependency surface.

audio-quality-check