web-search

CRITICAL E005: Suspicious download URL detected in skill instructions.

• Suspicious download URL detected (high risk: 0.75). Most URLs are documentation/article pages (including example.com placeholders) and an image, but the presence of a direct installer URL (https://cli.inference.sh) used with a "curl ... | sh" pattern and a personal cloud file host (cloud.inference.sh) makes these a moderate-to-high risk because remote shell scripts or hosted files can deliver malware if the source is not fully trusted and verifiable.

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

• Third-party content exposure detected (high risk: 1.00). The SKILL.md explicitly instructs using inference.sh apps (e.g., tavily/extract and exa/extract) to fetch and extract content from arbitrary URLs (see the "Tavily Extract" and "Exa Extract" examples and the "Workflow: Extract + Summarize" that feeds extracted webpage content into LLMs), which exposes the agent to untrusted public web content that could inject instructions.

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

• Potentially malicious external URL detected (high risk: 0.90). The Quick Start runs "curl -fsSL https://cli.inference.sh | sh", which fetches and executes remote installer code at runtime from https://cli.inference.sh and the skill depends on the infsh CLI to run apps, so this is a runtime external dependency that executes remote code.