text-to-speech

HIGH W007: Insecure credential handling detected in skill instructions.

• Insecure credential handling detected (high risk: 1.00). The prompt embeds a literal TrustedClientToken-like token in the WebSocket endpoint URL, which would require the agent to include that secret verbatim in connection code or commands, creating an exfiltration risk.

HIGH W008: Secret detected in skill content (API keys, tokens, passwords).

• Secret detected (high risk: 1.00). I scanned the document for literal, high-entropy credentials. The WebSocket endpoint line

wss://speech.platform.bing.com/consumer/speech/synthesize/readaloud/edge/v1?TrustedClientToken=6A5AA1D4EAFF4E9FB37E23D68491D6F4&ConnectionId=[UUID]

contains a 32-character hex-like token (TrustedClientToken=6A5AA1D4EAFF4E9FB37E23D68491D6F4). This is a literal, non-truncated, random-looking value in a query parameter and meets the definition of a secret (high-entropy token). I flagged it.

Ignored items and rationale:

• The earlier truncated endpoint occurrences with "..." were treated as redacted/truncated and ignored per the rules.
• The ConnectionId is shown as [UUID] (placeholder) and thus ignored.
• Other strings in the doc are descriptive, simple examples, or explicit placeholders (e.g., voice IDs, settings) and do not meet the high-entropy secret criterion.