The Agent Skills Directory

Unverifiable Dependencies & Remote Code Execution (CRITICAL): The installation instructions in SKILL.md and references/authentication.md recommend running curl -fsSL https://cli.inference.sh | sh. This is a classic piped-to-shell RCE pattern from a domain not listed in the trusted sources, allowing unverified code execution in the user's environment.\n- Command Execution (HIGH): The skill requests broad Bash(infsh *) permissions. This tool capability allows the agent to execute any sub-command of the infsh CLI, which includes authenticated actions such as infsh login, posting to social media (x/post-tweet), and managing cloud tasks.\n- Indirect Prompt Injection (LOW): The skill presents a significant surface for indirect prompt injection by processing untrusted user prompts through over 150 third-party AI applications. \n
Ingestion points: Data enters via the --input flag in infsh app run and infsh app test commands.\n
Boundary markers: None identified. No delimiters or instructions are provided to the agent to segregate user data from system instructions.\n
Capability inventory: Full Bash execution for the infsh binary, network access to numerous AI providers, and social media integration (Twitter/X).\n
Sanitization: None documented. Content is passed directly as JSON input to remote models.\n- Privilege Escalation (MEDIUM): In references/cli-reference.md, shell completion commands suggest writing directly to /etc/bash_completion.d/, which typically requires elevated (root) privileges.

agent-tools