nuclei-template-writer

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

• Third-party content exposure detected (medium risk: 0.30). The skill’s runtime workflow ingests the user’s free-text “finding” input (HTTP request/response pairs or descriptions) and extracts arbitrary strings/payloads/errors from it to generate the Nuclei template.

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

• Secret detected (high risk: 1.00). I flagged a literal JWT-like token in references/vuln-classes.md because it is a concrete token string (forged_token = "eyJhbGciOiJub25lIiwidHlwIjoiSldUIn0.eyJzdWIiOiIxIiwicm9sZSI6ImFkbWluIn0.") rather than a placeholder or truncated example. Although presented as an example for an "alg:none" JWT forgery, this is a real, usable token value that could grant access on vulnerable servers that accept alg:none tokens — therefore it meets the definition of a secret/credential to flag.

Ignored potential matches and why:

• dXNlcm5hbWU6cGFzc3dvcmQ= (base64 of "username:password") and similar base64/example credentials: documentation/example placeholders — low-entropy / instructional.
• "Test1234", "test@test.com", "admin", "password", and other simple strings shown as run examples: obvious examples/placeholders — ignore.
• Truncated values like "eyJ..." and "token=eyJ..." appearing in examples and run snippets: truncated placeholders — ignore.
• Regex patterns (e.g., AKIA..., AIza...) and template variables such as "REPLACE_ME": patterns/placeholders, not actual secrets.