functional-natural.json

[
  {
    "id": "multi-file-search",
    "_comment": "Same intent as functional.json but no 'use codeact' hint. Cross-references TODOs against import lines, which is awkward in one bash one-liner — codeact should win naturally.",
    "prompt": "For each Python file in src/, list its TODO comments AND the modules it imports. Show file -> {todos: [...], imports: [...]} as JSON.",
    "assertions": {
      "codeact_invoked": true,
      "output_contains": ["app.py", "TODO"]
    }
  },
  {
    "id": "batch-count",
    "_comment": "Multi-dimensional aggregation per file (lines, TODOs, function defs) plus sort. Hard to do without a loop — natural fit for codeact.",
    "prompt": "For each Python file under src/, count its total lines, number of TODO comments, and number of `def` function definitions. Show all three as a table sorted by total lines descending.",
    "assertions": {
      "codeact_invoked": true,
      "output_contains": ["app.py", "lines"]
    }
  },
  {
    "id": "json-validate",
    "_comment": "Validate + aggregate (top-level key count for valid ones). Multi-file scan with per-file analysis — codeact fits.",
    "prompt": "Check every JSON file in config/ for valid syntax. For each valid file also report the number of top-level keys. For invalid files report the parse error. Combine into one summary table.",
    "assertions": {
      "codeact_invoked": true,
      "output_contains": ["broken.json"]
    }
  },
  {
    "id": "single-file-read",
    "_comment": "Negative test: trivial single-file read should NOT trigger codeact even when no hint is given.",
    "prompt": "Read README.md and tell me what it says in one sentence.",
    "assertions": {
      "codeact_invoked": false,
      "output_contains": ["README"]
    }
  }
]