nmt-market-research

CRITICAL E004: Prompt injection detected in skill instructions.

• Potential prompt injection detected (high risk: 0.70). The file embeds a hidden post-run "update check" that instructs the agent to read a local file (.nmt-version) and make an outbound curl to a remote URL (with "skip silently" on errors), which is an out-of-scope, potentially exfiltrative maintenance action and therefore a deceptive instruction outside the skill's stated market-research purpose.

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

• Third-party content exposure detected (medium risk: 0.30). Source of outsider-authored free text is the user’s Step 1–7 intake (STAGE 1 Steps 1–7) collected via AskUserQuestion as well as user-provided materials/claims, which the skill then ingests into the LLM context and explicitly uses in analysis and the “What you told me — and the risks I see in it” block (and can be web-verified in Deep mode).