The Agent Skills Directory

[Indirect Prompt Injection] (LOW): The skill's primary purpose is processing external text for synthesis, creating a natural ingestion point for untrusted data.
Ingestion points: SSMLProcessor.process(ssml) and SentenceProcessor.synthesize_natural(text) take external strings.
Boundary markers: None present in the example code.
Capability inventory: The skill writes to the filesystem (_save_audio, AudioConverter.convert) and likely invokes ffmpeg via the pydub library.
Sanitization: While the TTSContentFilter masks PII (passwords, keys), it does not sanitize for embedded instructions that could influence the agent's behavior.
[Resource Exhaustion] (LOW): The RateLimitedTTS implementation is a proactive security measure to prevent denial-of-service (DoS) via synthesis abuse.
[Insecure Functionality] (LOW): The SecureAudioOutput class uses tempfile.mktemp(). This function is deprecated and insecure as it is vulnerable to race conditions where a malicious actor could create a file at the returned path before the application does. tempfile.mkstemp() or NamedTemporaryFile should be used instead.
[False Positive] (INFO): The automated scanner flagged logger.info as a malicious URL. This is a false positive; the string is a standard Python logging method call and not a network resource.

text-to-speech