Evals

1{
2  "skill_name": "add-webmcp",
3  "evals": [
4    {
5      "id": 1,
6      "prompt": "Use add-webmcp on this Next.js commerce app. Analyze its routes, forms, server actions, and Zod schemas, then expose product search and checkout features to agents.",
7      "expected_output": "A traced capability inventory, first-party WebMCP integration for safe tasks, and Stagehand discovery/invocation evidence that preserves checkout confirmation.",
8      "assertions": [
9        "The implementation traces UI handlers through validation, authorization, and side effects before choosing tools",
10        "Search is marked read-only while final checkout remains behind an application-enforced confirmation boundary",
11        "The implementation reuses existing schemas and business logic instead of duplicating them in browser code",
12        "Stagehand discovers every expected tool and invokes only safe synthetic fixtures"
13      ]
14    },
15    {
16      "id": 2,
17      "prompt": "Make this Vite React support portal agent-ready with WebMCP. It has a ticket search, a save-draft action, and a send-reply form.",
18      "expected_output": "Tools for search and draft saving plus a guarded reply workflow, integrated at the React root and validated through Stagehand.",
19      "assertions": [
20        "The tool contracts mirror the portal's existing validators and current-user permissions",
21        "Search and draft fixtures are invoked against disposable data",
22        "Sending a reply is not autosubmitted and is discovery-only unless an authorized sandbox exists",
23        "Untrusted ticket or user content is annotated and treated as data"
24      ]
25    },
26    {
27      "id": 3,
28      "prompt": "I only have a live URL and want a standalone webmcp.init.js without modifying source code.",
29      "expected_output": "A routing decision to use webmcp-gen rather than changing an unavailable codebase with add-webmcp.",
30      "assertions": [
31        "The skill does not pretend to analyze source code that was not provided",
32        "The request is routed to the site-specific init-script workflow"
33      ]
34    }
35  ]
36}