app-ai-guardrails
Pass
Audited by Gen Agent Trust Hub on Jul 21, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: Fetches installation scripts and project templates from official and well-known service providers, such as Astral (astral.sh) for Python tooling and the official Spring Initializr (start.spring.io) for Java projects. The skill instructions mandate that installers be downloaded for user inspection rather than directly piped to a shell.
- [REMOTE_CODE_EXECUTION]: Installs AI agent skills and development tools from external repositories, including verified sources from the author, GitHub, and Vercel Labs. These operations are performed using standard package managers (npx, brew, go, cargo) and include explicit safety checkpoints where the agent must obtain user consent and treat the fetched content as untrusted data.
- [COMMAND_EXECUTION]: Executes standard framework CLI commands and package management tasks (npm, uv, go, cargo, gradle) to initialize projects, run linting gates, and verify test coverage. These commands are necessary for the skill's primary function of scaffolding and verifying application guardrails.
- [PROMPT_INJECTION]: Recognizes the potential for indirect prompt injection via third-party agent skills. It implements a defensive 'M2 posture' which instructs the agent to treat all fetched content as data rather than instructions, preventing external files from overriding the skill's phase system or safety constraints.
Audit Metadata