Recipe catalog / instruction-clarity

Grade instruction clarity

How unambiguous is instruction for a delegate who has only context, on a five-level rubric?

You need to grade an instruction before sending it to a sub-agent or teammate, so an orchestrator can rewrite it or ask a question instead of letting the delegate guess.

Explore this recipe interactively · Source and implementation guide

Use instruction-clarity 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 { instructionClarity } from 'jev-recipes/instruction-clarity';

const result = await instructionClarity({
  "instruction": "Clean up the flaky tests in the payments module and make sure CI is green.",
  "context": "The payments module has 214 tests across 18 files. Three tests were quarantined last week with a skip marker and a link to issue #882. CI currently fails on main because of an unrelated lint error in the docs build.",
  "minConfidence": 0.8
});
console.log(result);

Input contract

{
  "input": {
    "$schema": "https://json-schema.org/draft/2020-12/schema",
    "type": "object",
    "properties": {
      "instruction": {
        "type": "string"
      },
      "context": {
        "type": "string"
      },
      "minConfidence": {
        "type": "number",
        "minimum": 0,
        "maximum": 1
      }
    },
    "required": [
      "instruction"
    ]
  },
  "result": {
    "$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
        }
      },
      "clarity": {
        "type": "string",
        "enum": [
          "unusable",
          "ambiguous",
          "gappy",
          "clear",
          "precise"
        ]
      }
    },
    "required": [
      "model",
      "usage",
      "status",
      "score",
      "level",
      "confidence",
      "probabilities",
      "clarity"
    ],
    "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 instruction-clarity.

{
  "model": "demo-fixture",
  "usage": {
    "input_tokens": 0,
    "output_tokens": 0
  },
  "status": "ready",
  "score": 1.99,
  "level": 2,
  "confidence": 0.83,
  "probabilities": {
    "0": 0.01,
    "1": 0.08,
    "2": 0.83,
    "3": 0.07,
    "4": 0.01
  },
  "clarity": "gappy"
}

Evaluation evidence

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

Limitations

• Grades ambiguity from the delegate's point of view using only instruction and context. It cannot know what the delegate already knows unless context says so.
• Reports how clear the instruction is, not whether the instructed work is sensible, safe, or within the delegate's abilities.

Related recipes

• query-specificity: Use query-specificity to grade how narrowly a search query pins down what is wanted, rather than how clearly an instruction directs work.
• clarify: Use clarify to list the specific questions a delegate would need answered when the instruction grades below clear.