Function buildPluginScenarios

plugins/plugin-eval/src/core/benchmark.js:144–177

(target)

}

function buildPluginScenarios(target) {
  return [
    {
      id: "entrypoint-routing",
      title: "Entrypoint routing",
      purpose: "Check whether the plugin routes a representative request through the right capability and finishes the task in the workspace.",
      userInput: `Use the local Codex plugin "${target.name}" if it helps. Handle a representative request for this plugin in the current workspace and finish the task.`,
      successChecklist: [
        "The run uses the right plugin capability.",
        "The result changes the workspace instead of only describing work.",
      ],
    },
    {
      id: "multi-skill-follow-up",
      title: "Multi-capability follow-up",
      purpose: "Measure whether a second request stays coherent when the plugin needs a different angle.",
      userInput: `Use the local Codex plugin "${target.name}" if it helps. Complete a follow-up task in the current workspace that needs a different angle from the same plugin, and keep the result cohesive.`,
      successChecklist: [
        "The task completes end to end.",
        "The run stays coherent instead of redoing unrelated work.",
      ],
    },
    {
      id: "plugin-boundary",
      title: "Plugin boundary",
      purpose: "Check whether the plugin narrows scope when the request is a weak match.",
      userInput: `This task is a weak match for the local Codex plugin "${target.name}". Handle the appropriate slice in the workspace, leave unrelated work alone, and explain the boundary clearly.`,
      successChecklist: [
        "The run narrows scope instead of forcing a bad fit.",
        "Any edits stay justified by the request.",
      ],
    },
  ];
}

async function createStarterBenchmarkConfig(target, options = {}) {
  if (!["skill", "plugin"].includes(target.kind)) {