mention-radar

HIGH W007: Insecure credential handling detected in skill instructions.

• Insecure credential handling detected (high risk: 0.90). The skill instructs the agent to perform direct curl calls that include the XAI_API_KEY in an Authorization header (passed on the command line), forcing the agent to embed/use the secret in generated shell commands and tool calls — a high exfiltration risk.

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

• Third-party content exposure detected (medium risk: 0.30). Outsider-authored text is ingested at runtime via the skill’s external mention collection: it runs an x_search query through X.AI to retrieve other people’s X post full text (step 2, Path A) and also uses WebSearch to fetch/search and ingest snippets/content from public platforms like Reddit/GitHub Discussions/HN (step 2, Path B) for later categorization.