Recipe catalog / plan-completeness

Grade plan completeness

How completely does plan cover what task requires, on a five-level rubric?

You need to check an agent-written plan against the task before execution starts, so missing requirements are caught while they are cheap to add.

Explore this recipe interactively · Source and implementation guide

Use plan-completeness in TypeScript

Install with npm install jev-recipes. Requires Node.js 22.9 or newer and ES modules. Set TYPESAFE_API_KEY in your server environment for live calls, which send input to TypeSafe and use API quota. See the installation guide.

import { planCompleteness } from 'jev-recipes/plan-completeness';

const result = await planCompleteness({
  "task": "Add an Export as CSV button to the reports page. The export must respect the filters currently applied and must work for guest accounts as well as members.",
  "plan": "1. Add an Export as CSV button beside the existing PDF export control. 2. Create a /reports/export.csv endpoint that serializes the current report query. 3. Pass the active filter parameters from the page to the new endpoint. 4. Add an integration test that exports a filtered report and checks the rows match the filtered view.",
  "minConfidence": 0.8
});
console.log(result);

Input contract

{
  "input": {
    "$schema": "https://json-schema.org/draft/2020-12/schema",
    "type": "object",
    "properties": {
      "task": {
        "type": "string"
      },
      "plan": {
        "type": "string"
      },
      "minConfidence": {
        "type": "number",
        "minimum": 0,
        "maximum": 1
      }
    },
    "required": [
      "task",
      "plan"
    ]
  },
  "result": {
    "$schema": "https://json-schema.org/draft/2020-12/schema",
    "type": "object",
    "properties": {
      "model": {
        "type": "string"
      },
      "usage": {
        "type": "object",
        "properties": {
          "input_tokens": {
            "type": "integer",
            "minimum": 0,
            "maximum": 9007199254740991
          },
          "output_tokens": {
            "type": "integer",
            "minimum": 0,
            "maximum": 9007199254740991
          }
        },
        "required": [
          "input_tokens",
          "output_tokens"
        ],
        "additionalProperties": false
      },
      "status": {
        "type": "string",
        "enum": [
          "ready",
          "review"
        ]
      },
      "score": {
        "type": "number",
        "minimum": 0
      },
      "level": {
        "type": "integer",
        "minimum": 0,
        "maximum": 9007199254740991
      },
      "confidence": {
        "type": "number",
        "minimum": 0,
        "maximum": 1
      },
      "probabilities": {
        "type": "object",
        "propertyNames": {
          "type": "string"
        },
        "additionalProperties": {
          "type": "number",
          "minimum": 0,
          "maximum": 1
        }
      },
      "completeness": {
        "type": "string",
        "enum": [
          "none",
          "partial",
          "mostly",
          "nearly",
          "complete"
        ]
      }
    },
    "required": [
      "model",
      "usage",
      "status",
      "score",
      "level",
      "confidence",
      "probabilities",
      "completeness"
    ],
    "additionalProperties": false
  }
}

Saved example result

This hand-authored response demonstrates the contract. It is not a model accuracy measurement. Run it without an API key: npx jev-recipes demo plan-completeness.

{
  "model": "demo-fixture",
  "usage": {
    "input_tokens": 0,
    "output_tokens": 0
  },
  "status": "ready",
  "score": 2.06,
  "level": 2,
  "confidence": 0.82,
  "probabilities": {
    "0": 0.01,
    "1": 0.05,
    "2": 0.82,
    "3": 0.11,
    "4": 0.01
  },
  "completeness": "mostly"
}

Evaluation evidence

Use the evaluation guide to measure this decision on your own labeled cases.

Limitations

• Grades coverage of stated and clearly implied requirements. It does not judge whether the planned steps would actually work.
• The top level requires a verification step. A plan that covers everything but never checks the result grades one level lower by design.

Related recipes

• answer-coverage: Use answer-coverage to check a finished draft against explicit questions rather than a plan against a task.
• clarify: Use clarify when the task itself is ambiguous, since a plan cannot cover requirements the task never made clear.