israeli-heritage-explorer

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

• Third-party content exposure detected (medium risk: 0.30). The skill’s required runtime workflow ingests user-authored free text by turning the user’s requested research scope/topic into NLI OpenLibrary API query clauses (Step 1 → Step 2) and then processes the resulting metadata into the report (Step 3-4) via https://api.nli.org.il/openlibrary/search.

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

• Secret detected (high risk: 1.00). I flagged a literal, high-entropy API key found in evidence.json. The file contains the string "DVQyidFLOAjp12ib92pNJPmflmB5IessOq1CJQDK" in a raw_snippet for the claim "guest-key-exists" — this is not a placeholder like YOUR_KEY, it appears to be an actual guest API key and meets the definition of a secret.

Ignored items: example placeholders such as "YOUR_KEY" in curl examples (references/nli-api-reference.md) are documentation placeholders and not flagged. The script uses an environment variable (NLI_API_KEY) rather than a hardcoded key, so nothing in the scripts besides the evidence.json literal key is flagged. Numeric system identifiers and IIIF docids are not secrets.