setup-experiment

CRITICAL E005: Suspicious download URL detected in skill instructions.

• Suspicious download URL detected (high risk: 0.70). These URLs point to a main site plus a direct install script (https://cli.promptingco.com/install.sh) hosted on a differently named but similar domain—curl|bash from an external .sh is a common malware vector and the domain mismatch raises typosquatting/ownership uncertainty, so treat it as potentially risky unless you can verify the domains are owned by the same trusted organization.

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

• Third-party content exposure detected (high risk: 0.85). The required workflow ingests outsider-authored free text when it performs Step B1/B2 “web-search for … docs” and “fetch docs from the location captured in Step B1,” then uses that fetched page text to propose tasks (Path B Step B1 Step B2), which is public web content authored by third parties and becomes LLM-readable context.

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

• Potentially malicious external URL detected (high risk: 0.80). The skill explicitly instructs running remote install code via curl -fsSL https://cli.promptingco.com/install.sh | bash (which fetches and executes remote code at runtime), and it also fetches a user-provided docs URL during Step B1/B2 to generate task prompts (the fetched docs directly shape prompts), so there are runtime external dependencies that execute code or control prompts.