fix-hard

HIGH W007: Insecure credential handling detected in skill instructions.

• Insecure credential handling detected (high risk: 0.90). The prompt mandates providing "file:line" evidence and writing proof traces for every claim, which forces the agent to include verbatim file contents (and thus any embedded API keys, tokens, or passwords) in its outputs, creating an exfiltration risk.

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

• Third-party content exposure detected (high risk: 0.90). The workflow explicitly requires using a "researcher" subagent to "research quickly about the root causes on the internet" (see "Fix the issue" step 2), which means the agent will fetch and interpret open/public web content that can influence diagnosis and actions.