Executes code in a browser-based sandbox environment, supporting frameworks like React-Vite for building and previewing web applications
The server has only one tool, 'run_vibe_sandbox', with a name that violates verb_noun convention and describes multiple unrelated responsibilities. The tool name uses 'run' (generic verb) instead of a domain-specific action. The description is minimal and lacks critical details about when to use it, prerequisites, and expected return structure. The input schema is visible but parameters lack descriptions entirely, violating the core requirement that every parameter must explain what it controls. The output schema is not documented in the source code, the client returns a hardcoded dict with no specification of valid fields, data types, or constraints. Error handling is absent; the mock implementation shows no error scenarios or recovery guidance. Security is not addressed: the tool appears to make external HTTP requests to a sandbox service but exposes no mechanism for managing credentials or validating untrusted input. The tool also violates single responsibility: 'run_vibe_sandbox' both executes code AND generates a preview URL AND reports build status, these should be separate tools.
Executes code in a browser-based sandbox environment and returns preview URL and build status
Tool name uses generic verb 'run' instead of domain-specific action. 'run_vibe_sandbox' should be 'execute_code_in_sandbox' or 'create_web_preview' to convey intent clearly.
Input parameters 'natural_language_prompt' and 'framework' lack descriptions.
Output schema not documented. The tool returns a dict with fields 'sandbox_preview_url', 'bundle_size_kb', 'build_status' but no formal schema, type info, or constraints are visible. LLMs cannot plan downstream calls or validate results.
Tool violates single responsibility principle. 'run_vibe_sandbox' performs three distinct actions: (1) execute code, (2) generate preview URL, (3) report build status. Should be split into separate tools or clearly document this composition.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 23 | 2026-07-28+ | v2 |
No error handling or recovery guidance. The mock returns only success cases. What happens if the sandbox fails? If the build times out? If the framework is unsupported? No error codes, messages, or retry guidance provided.
Parameter 'framework' has no enum or constraint. LLMs will hallucinate invalid values. Should document allowed frameworks: ['React-Vite', 'Vue-Vite', 'Svelte-Vite', ...] or similar.
Tool description is 143 characters, borderline short, but lacks critical information: When should LLM call this vs alternatives? What prerequisites exist? What does 'preview URL' contain? Does build_status imply the code is executing or just compiled?