caffeine-app
Pass
Audited by Gen Agent Trust Hub on Nov 19, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill instructs the AI agent to execute shell commands to scaffold project files, manage dependencies, and build canisters. This involves using the
@caffeineai/clifor operations such ascaffeine build,caffeine install, andcaffeine auth login, as well as standard tools likepnpmandmops. - [EXTERNAL_DOWNLOADS]: The skill facilitates the installation of development tools and libraries from the public npm registry and the Motoko package manager (Mops). Downloaded resources include official vendor packages such as
@caffeineai/cliand@caffeineai/core-infrastructure, as well as well-known libraries like React, Vite, and various @radix-ui components. - [INDIRECT_PROMPT_INJECTION]: As a tool for creating application code based on user requests, this skill provides a surface for indirect instruction injection.
- Ingestion points: User descriptions and requirements for building a specific application.
- Boundary markers: Not explicitly included in the scaffolded project templates.
- Capability inventory: Subprocess execution for building and previewing projects through the
caffeineCLI. - Sanitization: The skill recommends delegating Motoko implementation details to the authoritative
writing-motokoskill to ensure secure coding patterns and prevent errors.
Audit Metadata