user-research-synthesis

HIGH W007: Insecure credential handling detected in skill instructions.

• Insecure credential handling detected (high risk: 0.90). The prompt mandates using verbatim quotes and appending raw interview observations (and accepts pasted/uploaded transcripts), which means any secrets contained in those inputs (API keys, passwords, tokens) would be reproduced exactly in generated outputs and saved files—creating a high exfiltration risk.

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

• Third-party content exposure detected (high risk: 0.85). The skill’s runtime workflow ingests “raw user interview notes, transcripts, and observations” that the operating user uploads/pastes (including “Slack messages from customer channels” and “customer support tickets”), which are outsider-authored free-form text and can be fed into the agent’s LLM context via the “Upload or paste… You can upload multiple files or just paste everything into this chat” step.

MEDIUM W021: Hidden or invisible Unicode characters detected (potential obfuscation or prompt injection).

• Hidden Unicode characters detected (1 type(s) found)