The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses the belt CLI to interact with remote AI models. The allowed-tools frontmatter configuration restricts the Bash tool to only allow commands starting with belt, which is a security best practice for scoping agent capabilities.
[EXTERNAL_DOWNLOADS]: The skill references installation of the belt CLI via npx and provides documentation links to the official inference.sh GitHub repository. These are standard procedures for integrating with this service.
[INDIRECT_PROMPT_INJECTION]: The skill defines a surface for processing external data.
Ingestion points: User-controlled text is passed into the prompt field, and external URLs are passed into the images field of the JSON payload for the belt CLI in SKILL.md.
Boundary markers: The data is structured as JSON, but there are no specific instructions to the agent to treat the prompt content as untrusted or to ignore embedded instructions.
Capability inventory: The skill utilizes the belt command to perform remote image generation and editing tasks.
Sanitization: No explicit sanitization or filtering of the prompt or images input is described in the skill instructions.

p-image