building-inferencesh-apps
Pass
Audited by Gen Agent Trust Hub on Aug 6, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill recommends downloading the
beltCLI tool directly from the vendor's domain athttps://cli.inference.sh. - [EXTERNAL_DOWNLOADS]: The documentation references installation scripts for external developer tools from well-known providers, including
uvfromastral.sh,fnmfromfnm.vercel.app, andnvmfrom GitHub's raw content domain. - [REMOTE_CODE_EXECUTION]: The installation procedures for the platform CLI and supporting utilities utilize piped shell execution (e.g.,
curl | sh,curl | bash, and PowerShelliex), which is a common practice for developer tool installers from these sources. - [COMMAND_EXECUTION]: The skill instructs the agent to use the
beltCLI for critical workflows, such as initializing projects, running local tests, and deploying code to the inference.sh cloud. - [PROMPT_INJECTION]: The skill processes user-defined project files (such as
inf.yml,inference.py, andinference.js) which serves as an indirect prompt injection surface. - Ingestion points: Project configuration and logic files in the
my-app/directory (SKILL.md). - Boundary markers: None identified in the provided instructions.
- Capability inventory: Execution of shell commands via the CLI, file system modifications, and network requests via included code examples (SKILL.md, references/python-app-logic.md).
- Sanitization: There is no explicit sanitization or validation of the content within these project files before the agent processes them.
Audit Metadata