autocli

CRITICAL E005: Suspicious download URL detected in skill instructions.

• Suspicious download URL detected (high risk: 0.80). Suspicious: the prompt instructs piping a raw install.sh from raw.githubusercontent.com (and a GitHub repo of unknown trustworthiness) directly to sh—a direct remote-code-execution vector—while the X status URL is just a benign reference; this combination makes the download/install step high risk unless you manually inspect the repo and script and verify the maintainer.

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

• Third-party content exposure detected (high risk: 0.90). The skill explicitly instructs the agent to fetch and scrape live, user-generated content from many public third-party sites (e.g., SKILL.md and README.md show commands like "autocli reddit read --url <post_url>", "autocli twitter timeline", "autocli youtube search", and the YAML/browser_evaluate adapter workflow for arbitrary URLs), so untrusted web content will be read and can influence the agent's actions.

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

• Potentially malicious external URL detected (high risk: 1.00). The skill instructs at runtime to install a required dependency by running remote shell code via curl -fsSL https://raw.githubusercontent.com/nashsu/AutoCLI/main/scripts/install.sh | sh, which fetches and executes remote code that the skill depends on.