The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructs the agent to download and execute an installation script directly from the vendor's domain at https://cli.inference.sh using a curl | sh pipeline to install the belt CLI tool.
[DATA_EXFILTRATION]: The belt CLI features an automatic file detection mechanism that identifies local file paths provided in input parameters and uploads the contents of those files to the inference.sh cloud infrastructure for processing. This poses a significant risk of accidental or malicious exfiltration of sensitive files, such as environment variables or credentials, if they are passed as inputs to the tool.
[COMMAND_EXECUTION]: The skill requires access to the Bash tool to perform CLI installation, authentication, and execution of various AI tasks through the belt command-line utility.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection as it processes external prompts and data for delivery to third-party AI models. 1. Ingestion points: Input parameters for the belt app run command found in SKILL.md and references/running-apps.md. 2. Boundary markers: No delimiters or explicit instructions to ignore embedded commands are present. 3. Capability inventory: Command execution via the Bash tool as defined in the skill's frontmatter. 4. Sanitization: No sanitization or validation of input data is documented before the content is passed to the execution environment.

agent-tools