sdk-package-publishing

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

• Secret detected (high risk: 1.00). The potential match PYPI_TOKEN: pypi-AgEIcHlwaS5vcmcCJDk3ZmQ2YjRhLTFiMmMtNGQ1ZS04ZjAwLTEyMzQ1Njc4OWFiYw appears within the eval prompt file (evals/evals.json) under prompt 23. This prompt describes a simulated security incident where a user is pasting a leaked token from CI to ask for incident response help.

Although this string contains the standard prefix and high-entropy structure of a real PyPI token (pypi-AgEI...), it is explicitly part of an evaluation dataset test case representing a credential leakage incident. However, looking at the instruction: "Include: Actual API keys... EXCEPTION: a literal redaction marker of the form 'REDACTED_SECRET_' is NOT a placeholder -- it marks where a real credential was found and stripped before scanning, so always flag it."

Since this string is an active-looking credential included in a user-provided prompt (eval prompt), and not a redaction marker, it meets the definition of an actual credential leak embedded in the documentation/test suite.