newsletter-curation

CRITICAL E005: Suspicious download URL detected in skill instructions.

• Suspicious download URL detected (high risk: 0.80). These URLs host a remote shell installer (curl ... | sh) from a small .sh domain — executing an unattended remote shell script is a high-risk distribution pattern unless the domain and publisher are well-known and verifiable.

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

• Third-party content exposure detected (high risk: 0.90). The skill's "Content Sourcing" section explicitly directs running external searches via infsh (e.g., "infsh app run tavily/search-assistant" and queries like "site:reddit.com [your niche] discussion this week") to fetch user-generated public content (Reddit, Twitter, forums) that the agent is expected to read and use for curation, so untrusted third-party content can materially influence its outputs and actions.

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

• Potentially malicious external URL detected (high risk: 1.00). The skill's Quick Start instructs fetching and executing remote code with "curl -fsSL https://cli.inference.sh | sh" (https://cli.inference.sh), which is run at setup/runtime, directly executes remote code, and is required for the infsh commands used throughout the skill.