{
  "skill_name": "create-goal",
  "evals": [
    {
      "id": 1,
      "name": "vague-idea-build-faster",
      "prompt": "I want to make the build faster",
      "expected_output": "A structured goal with an objective about build time reduction, success criteria with measurable targets, verification evidence, scope boundaries, iteration policy, completion audit, and blocked stop condition.",
      "assertions": [
        "The output contains a section titled '## Goal'",
        "The output contains '**Objective:**' followed by a single sentence",
        "The objective includes a specific numeric target with units (e.g. '12 minutes', '50%', '1000 requests')",
        "The output contains '**Success Criteria:**' with at least 2 checkbox items '- [ ]'",
        "The output contains '**Verification Evidence:**' with concrete evidence types",
        "The output contains '**Scope:**' with both '**In scope:**' and '**Out of scope:**' subsections",
        "The output contains '**Iteration Policy:**'",
        "The output contains '**Completion Audit:**'",
        "The output contains '**Blocked Stop Condition:**'",
        "The output contains clarifying questions before the goal draft"
      ]
    },
    {
      "id": 2,
      "name": "specific-feature-oauth",
      "prompt": "Add OAuth2 login with Google to the existing auth system. Must support refresh tokens and handle session expiry gracefully.",
      "expected_output": "A goal with a specific objective about OAuth2 integration, success criteria covering refresh tokens and session expiry, and clear scope boundaries.",
      "assertions": [
        "The output contains '**Objective:**' followed by a single sentence",
        "The objective mentions OAuth2 or Google login",
        "The success criteria include refresh tokens",
        "The success criteria include session expiry handling",
        "The scope explicitly states what auth providers are in scope",
        "The blocked stop condition mentions API credentials or missing dependencies"
      ]
    },
    {
      "id": 3,
      "name": "drifting-goal-refactor",
      "prompt": "I started refactoring the database layer but now I'm not sure what the goal is anymore. I just know the current code is messy.",
      "expected_output": "A goal that helps the user re-establish scope and boundaries for an in-progress refactor, with clear success criteria and out-of-scope items.",
      "assertions": [
        "The output contains '**Objective:**' followed by a single sentence",
        "The objective does not assume specific technologies unless stated",
        "The scope has explicit '**Out of scope:**' items to prevent drift",
        "The completion audit emphasizes mapping requirements to existing changes",
        "The blocked stop condition addresses partially completed work"
      ]
    },
    {
      "id": 4,
      "name": "edge-case-too-vague",
      "prompt": "make it better",
      "expected_output": "Either clarifying questions that expose the vagueness, or a goal that forces specificity by asking for metrics and boundaries.",
      "assertions": [
        "The output contains clarifying questions",
        "At least one question asks 'what does better mean' or similar",
        "The objective does not use the word 'better' without a metric",
        "If a goal is drafted, the success criteria are pass/fail verifiable"
      ]
    },
    {
      "id": 5,
      "name": "research-goal",
      "prompt": "Research whether we should use PostgreSQL or MySQL for the new analytics pipeline. We need to handle 10M events/day.",
      "expected_output": "A goal with a research-oriented objective, success criteria that include evidence-based recommendations, and clear constraints on the decision.",
      "assertions": [
        "The objective is a single sentence about comparing PostgreSQL and MySQL",
        "The success criteria include a recommendation with justification",
        "The verification evidence includes benchmarks or load tests",
        "The scope limits the research to the analytics pipeline use case",
        "The constraints mention the 10M events/day requirement"
      ]
    }
  ]
}