The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill references external resources for setup and installation.
Fetches installation instructions from a remote repository: https://raw.githubusercontent.com/inference-sh/skills/refs/heads/main/cli-install.md.
Instructs the user or agent to install a CLI tool using npx skills add belt-sh/cli.
[COMMAND_EXECUTION]: The skill is designed to interact with the local system using the Bash tool to execute belt CLI commands.
Examples include belt login, belt app run <model> --input 'JSON', and belt app list.
The allowed-tools frontmatter correctly limits access to the Bash tool specifically for belt commands.
[INDIRECT_PROMPT_INJECTION]: The skill has an attack surface for indirect prompt injection through user-provided input.
Ingestion points: User-provided strings are used as prompts in the --input argument of shell commands (e.g., in SKILL.md).
Boundary markers: The skill uses single quotes and JSON formatting to delimit user input, which provides basic separation but may be bypassed by complex inputs containing single quotes.
Capability inventory: The skill can execute shell commands via the Bash tool and perform network operations via the belt CLI.
Sanitization: There is no explicit evidence of sanitization or escaping logic within the instruction file for the interpolated prompt data.

ai-image-generation