echo-feedback

HIGH W007: Insecure credential handling detected in skill instructions.

• Insecure credential handling detected (high risk: 0.80). The prompt asks the agent to ingest raw support tickets/CSV/interview dumps and to reproduce "representative quotes" verbatim, so if those inputs contain API keys, tokens, or passwords the LLM would be required to output secrets directly.

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

• Third-party content exposure detected (high risk: 0.85). Outsider free text is ingested at runtime via Step 1 “Accept any of the following as input” (e.g., support ticket exports, NPS verbatims, app store reviews, churn survey responses, social/community posts, and call notes), which are authored by non-operating users and then used for clustering/“Representative quotes” in the LLM context.