agent-tools

CRITICAL E005: Suspicious download URL detected in skill instructions.

• Suspicious download URL detected (high risk: 0.90). The package installer is delivered via a curl | sh from cli.inference.sh and pulls binaries/checksums directly from dist.inference.sh (direct binary distribution and piping an installer are high-risk behaviors unless you fully trust and verify the publisher), even though the package claims checksum/Sigstore verification this pattern is commonly used to distribute malicious executables.

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

• Third-party content exposure detected (medium risk: 0.30). The skill’s runtime path ingests outsider-authored free text via the CLI’s belt app run <app> --input '<json>' / --input input.json fields (e.g., prompt, query, text) which are user-supplied strings that the selected app workflow consumes.

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

• Potentially malicious external URL detected (high risk: 1.00). The skill directly instructs running a remote installer via "curl -fsSL https://cli.inference.sh | sh" and fetching/executing binaries via https://dist.inference.sh/cli/manifest.json and https://dist.inference.sh/cli/checksums.txt, which fetches and executes remote code at runtime.