independent-review

CRITICAL E005: Suspicious download URL detected in skill instructions.

• Suspicious download URL detected (high risk: 0.90). Several entries are direct links to installer scripts (.sh/.ps1) and others are deceptive "localhost" variants that are commonly used to bypass loopback checks — both are high-risk patterns for distributing malware or tricking install checks.

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

• Third-party content exposure detected (medium risk: 0.30). Outsider-authored text is ingested at runtime via independent_review.sh reading a caller-provided <file> or stdin (CONTENT="$([ "$FILE" = "-" ] && cat || cat -- "$FILE")") and sending it inside the ${CONTENT} section of the prompt to external reviewers (Codex CLI and ollama-cloud), so an attacker can supply arbitrary free text to be reviewed.

MEDIUM W012: Unverifiable external dependency detected (runtime URL that controls agent).

• Potentially malicious external URL detected (high risk: 0.90). The onboarding/reference includes curl|sh and PowerShell one-liners that the agent will pull and present at runtime (e.g. https://chatgpt.com/codex/install.sh), which if run fetch and execute remote install scripts — a high-confidence runtime instruction to execute remote code.