The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill provides instructions for a manual installation using a shell script downloaded via curl from the vendor's domain (cli.inference.sh). While this involves downloading and executing remote code, the instructions include explicit steps for the agent to review the script's content using cat before execution. The source domain is associated with the skill author's infrastructure.
[EXTERNAL_DOWNLOADS]: The skill references several external sources for installation, including official package registries (Homebrew, Scoop, NPM) and the vendor's binary distribution servers (dist.inference.sh). These are standard and expected delivery mechanisms for CLI-based tooling.
[COMMAND_EXECUTION]: The skill's primary functionality is delivered through the execution of the belt CLI. The allowed-tools configuration in the frontmatter correctly limits the agent's scope to specific CLI commands and package managers required for setup.
[INDIRECT_PROMPT_INJECTION]: The skill identifies ingestion points for untrusted data, such as semantic search results (belt knowledge search) and user-provided input files for AI apps (belt app run --input). While the skill relies on the CLI's schema validation, it does not explicitly define prompt boundary markers or sanitization steps to handle instructions that might be embedded in the data processed by the agent.

belt