The Agent Skills Directory

[REMOTE_CODE_EXECUTION] (CRITICAL): The skill instructions include curl -fsSL https://cli.inference.sh | sh. This is a piped remote execution pattern that downloads and runs a script from a non-trusted source without integrity checks.
[EXTERNAL_DOWNLOADS] (HIGH): The skill relies on binaries and scripts hosted at inference.sh, which is not included in the Trusted External Sources list. This creates a dependency on an unverified third-party infrastructure.
[COMMAND_EXECUTION] (MEDIUM): The skill makes extensive use of the Bash tool to execute infsh commands and uses npx to dynamically install additional skills, which increases the potential impact of any compromised dependency.
[INDIRECT_PROMPT_INJECTION] (LOW): The skill ingests untrusted data from external URLs for images and audio files without input sanitization or boundary markers. While this is the intended use case, it presents a surface for tool output poisoning.
Ingestion points: SKILL.md examples show user-provided URLs passed as JSON strings to infsh CLI inputs.
Boundary markers: Absent. The agent is instructed to pass raw strings directly into CLI arguments.
Capability inventory: The skill has access to the Bash tool, allowing file system interaction and network requests.
Sanitization: Absent. There is no evidence of URL validation or shell argument escaping.

ai-avatar-video