linkfox-ruiguan-text-trademark-detection

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

• Third-party content exposure detected (medium risk: 0.30). The runtime workflow posts user-provided productTitle/productText (free text) to the POST /ruiguan/textTrademarkDetection API and then ingests and prints the resulting JSON (including blacklistTrademarks/whitelistTrademarks) into stdout and session files.

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

• Secret detected (high risk: 1.00). I reviewed the skill files for literal, high-entropy values that could grant access.

Findings:

• scripts/onboarding.py lines 48-49 contain a high-entropy base64url-like literal assigned to _LOGIN_FIXED_UID ("eyJhX2lkIjoiNmEyMmM4YjA1YmM5MTZhIiwiZF9pZCI6IiJ9"). This is not a placeholder (it's a concrete encoded value) and is used as a fallback UID header for login flows, so it may function as an authentication-related token and should be treated as a secret.

Ignored items:

• Environment variable names and curl examples that reference $LINKFOXAGENT_API_KEY / $LINKFOX_AGENT_API_KEY are placeholders (references/api.md and other files) and intentionally instruct using env vars, so they are not flagged.
• Other strings (URLs, user-agent, numeric IDs, example JSON) are not high-entropy credentials and were ignored per the rules.