The Agent Skills Directory

Remote Code Execution (CRITICAL): In SKILL.md, the skill directs the user or agent to execute curl -fsSL https://cli.inference.sh | sh. This is a classic piped remote execution pattern where a script is downloaded and immediately run. Since inference.sh is not a Trusted External Source, this constitutes a critical security risk as the script could perform any action the user has permissions for.\n- Command Execution (HIGH): The skill utilizes the Bash tool to execute the infsh CLI and other processes. Running arbitrary binaries downloaded at runtime is a high-risk activity that provides a path for malicious command execution.\n- Indirect Prompt Injection (LOW):\n
Ingestion points: Untrusted text and document content are ingested via the --input parameters of the infsh and openrouter tools in SKILL.md.\n
Boundary markers: While data is structured in JSON, there are no delimiters or instructions provided to the agent or tools to ignore potentially malicious embedded instructions in the text content.\n
Capability inventory: The skill has access to shell execution (Bash) and network-connected APIs, making it a viable target for exploitation.\n
Sanitization: No evidence of input sanitization or verification is present in the provided script or documentation.\n- External Downloads (MEDIUM): The skill suggests installing additional dependencies via npx skills add inference-sh/skills@..., which introduces unverifiable third-party content into the agent's environment.

ai-podcast-creation