The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill provides instructions to install a command-line tool using curl -fsSL https://cli.inference.sh | sh. This is a high-risk pattern that executes a remote shell script directly in the terminal, allowing for potential code execution from an external source.
[EXTERNAL_DOWNLOADS]: The skill initiates downloads from cli.inference.sh and dist.inference.sh for the tool installation and binary retrieval. It also references external skills via npx skills add which downloads code from remote registries.
[COMMAND_EXECUTION]: The skill requires the Bash tool to execute infsh commands as defined in the allowed-tools metadata field. These commands are used to interact with various AI models.
[INDIRECT_PROMPT_INJECTION]: The skill takes untrusted user input (prompts) and interpolates them into shell commands for the image generation tool.
Ingestion points: User-provided prompts are placed within the --input argument of infsh app run commands.
Boundary markers: The skill uses JSON string structures (e.g., '{"prompt": "..."}') to encapsulate the input, which provides structural delimitation but does not inherently sanitize the content for instructions.
Capability inventory: The agent has the capability to execute infsh commands via the Bash tool, which can interact with multiple external AI providers.
Sanitization: There are no explicit instructions for the agent to sanitize or validate the user-provided prompt before interpolation into the command line.

ai-image-generation