The Agent Skills Directory

[REMOTE_CODE_EXECUTION] (CRITICAL): The skill configuration and documentation (SKILL.md, references/authentication.md) promote installation via curl -fsSL https://cli.inference.sh | sh. This pattern is a verified remote code execution vulnerability when used with untrusted domains, as it allows a remote server to execute arbitrary code on the user's machine.
[COMMAND_EXECUTION] (HIGH): The skill defines allowed-tools: Bash(infsh *) in SKILL.md. This grants the AI agent broad authority to execute any subcommand of the infsh CLI, including those that manage cloud deployments, handle authentication (infsh login), and modify system configurations.
[EXTERNAL_DOWNLOADS] (HIGH): The skill depends on binaries and scripts from dist.inference.sh and external skill sets added via npx skills add. These sources are not within the trusted scope defined for AI agent skills, posing a significant supply chain integrity risk.
[PROMPT_INJECTION] (LOW): The skill is vulnerable to Indirect Prompt Injection (Category 8). It ingests untrusted data via the --input flag in the infsh app run tool (documented in SKILL.md and references/running-apps.md). No boundary markers or sanitization procedures are documented to prevent embedded instructions in input data from hijacking the agent's logic.

agent-tools