vueuse-functions
Pass
Audited by Gen Agent Trust Hub on Nov 3, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill documents the experimental
useWebMCPcomposable which facilitates the registration of JavaScript functions as tools callable by AI agents. This creates a surface for indirect injection where a malicious web page could expose dangerous tools to the agent. - Ingestion points: Tools registered via
useWebMCPas described inreferences/useWebMCP.md. - Boundary markers: Documentation is absent regarding the use of delimiters to distinguish between data and instructions in tool results.
- Capability inventory: The composable allows any local function to be exposed as a discoverable tool for the agent, potentially including functions with side effects like data modification or navigation.
- Sanitization: No explicit guidance is provided for sanitizing inputs received from the agent or outputs returned to the agent.
- [COMMAND_EXECUTION]: The skill provides documentation for
useIpcRendererinreferences/useIpcRenderer.md. This allows communication with an Electron app's main process, which often has full system access. While a standard API for Electron developers, it provides a high-privilege communication channel that an agent could misuse if the target application lacks proper security boundaries. - [EXTERNAL_DOWNLOADS]: The documentation includes examples of fetching external scripts and data from well-known and generally trusted service providers:
- Fetches embedded player scripts from Twitch (
player.twitch.tv) inreferences/useScriptTag.md. - Downloads utility libraries from the Cloudflare CDN (
cdnjs.cloudflare.com) inreferences/useWebWorkerFn.md. - Accesses placeholder data from JSONPlaceholder (
jsonplaceholder.typicode.com) inreferences/useAsyncState.md.
Audit Metadata