nmt-analyze-interviews

CRITICAL E004: Prompt injection detected in skill instructions.

• Potential prompt injection detected (medium risk: 0.60). The skill includes an out-of-scope "update check" that instructs the agent to read a local file and run a curl command (potentially exfiltrating .nmt-version) after finishing, which is a hidden/extra external-execution step beyond the interview-analysis purpose.

HIGH W007: Insecure credential handling detected in skill instructions.

• Insecure credential handling detected (high risk: 0.90). The skill explicitly requires keeping user-supplied interview quotes verbatim and returning 2–3 anchor quotes per Core Job from files the agent reads, so any secrets present in those files (API keys, tokens, passwords) would be reproduced in the LLM's output—creating an exfiltration risk.

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

• Third-party content exposure detected (medium risk: 0.30). In nmt-analyze-interviews STAGE 3, the orchestrator spawns a distiller subagent per user-provided interview file and each distiller reads that file content (transcripts/notes/support/chat logs/survey open-ends) to extract verbatim expected-outcome clauses and anchor quotes.