The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill utilizes a remote installation pattern via curl -fsSL https://cli.inference.sh | sh. While this originates from the vendor's own domain, it involves executing unverified code from a remote source.
[REMOTE_CODE_EXECUTION]: The manual installation instructions involve dynamically fetching binaries from dist.inference.sh using curl and grep on a manifest file, which is then executed as a local CLI tool.
[COMMAND_EXECUTION]: The skill relies on the infsh command to perform its primary functions, including running AI models and managing cloud tasks.
[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface when processing user-supplied data.
Ingestion points: Data enters the agent context through the --input flag, which accepts JSON files or inline JSON strings (e.g., in SKILL.md and references/running-apps.md).
Boundary markers: There are no explicit boundary markers or instructions telling the agent to ignore embedded commands within the input JSON.
Capability inventory: The skill possesses the Bash(infsh *) capability, which allows it to execute commands that interact with external AI services.
Sanitization: No sanitization or validation of the input JSON structure is described within the skill's instructions.

agent-tools