A writer and a reviewer

Give one model a brief and another model the job of checking it. This file has a Claude seat write a function and its test, runs the test with Node, and has a Codex seat read the files and the test result. If the reviewer sends the work back, the writer runs again with the findings, once.

npm install @obversa/runtime @obversa/teams @obversa/engine-claude-cli @obversa/engine-codex

The file

Change the brief, the file list and the test command to your own. Run it from the directory the work belongs in.

import { ClaudeCliEngine } from '@obversa/engine-claude-cli';
import { CodexEngine } from '@obversa/engine-codex';
import { run } from '@obversa/runtime';
import { writerReviewerPair } from '@obversa/teams';

const workspace = process.cwd();
const writer = {
  engine: new ClaudeCliEngine({
    defaultModel: 'claude-sonnet-4-5',
    permissionMode: 'bypassPermissions',
  }),
  identity: {
    adapter: 'claude-cli',
    provider: 'anthropic',
    modelFamily: 'claude',
    model: 'claude-sonnet-4-5',
  },
};
const reviewer = {
  engine: new CodexEngine({
    defaultModel: 'gpt-5.6-luna',
    permissionMode: 'bypassPermissions',
  }),
  identity: {
    adapter: 'codex',
    provider: 'openai',
    modelFamily: 'gpt',
    model: 'gpt-5.6-luna',
  },
};
const team = writerReviewerPair({
  brief: 'Write a pure add(a, b) function in src/add.mjs with a Node test in test/add.test.mjs.',
  workspace,
  files: ['src/add.mjs', 'test/add.test.mjs'],
  test: { command: 'node', args: ['--test', 'test/add.test.mjs'] },
  writer,
  reviewer,
});
const result = await run(team, { cwd: workspace });

console.log(JSON.stringify(result.outcome, null, 2));

What a run printed

The output below is from one real run of this file, with the two command line tools signed in on the machine.

{
  "status": "pass",
  "summary": "dag \"writer-reviewer-pair\": all 3 node(s) green",
  "data": {
    "writer": {
      "status": "pass",
      "summary": "Created pure add(a, b) function in src/add.mjs and comprehensive test suite in test/add.test.mjs. All 5 tests pass, covering positive numbers, negative numbers, mixed signs, zero, and decimals."
    },
    "test": {
      "status": "pass",
      "summary": "`node` exited 0"
    },
    "reviewer": {
      "status": "pass",
      "summary": "The required pure add function and Node test are present and pass."
    }
  }
}

The writer left two files behind. src/add.mjs:

/**
 * Pure function that adds two numbers.
 * @param {number} a - First number
 * @param {number} b - Second number
 * @returns {number} Sum of a and b
 */
export function add(a, b) {
  return a + b;
}

and test/add.test.mjs:

import { describe, it } from 'node:test';
import assert from 'node:assert/strict';
import { add } from '../src/add.mjs';

describe('add', () => {
  it('adds two positive numbers', () => {
    assert.strictEqual(add(2, 3), 5);
  });

  it('adds two negative numbers', () => {
    assert.strictEqual(add(-2, -3), -5);
  });

  it('adds positive and negative numbers', () => {
    assert.strictEqual(add(5, -3), 2);
  });

  it('adds zero', () => {
    assert.strictEqual(add(0, 5), 5);
    assert.strictEqual(add(5, 0), 5);
  });

  it('adds decimal numbers', () => {
    assert.strictEqual(add(1.5, 2.5), 4);
  });
});

The reviewer wrote its decision to reviews/reviewer.json:

{
  "status": "pass",
  "summary": "The required pure add function and Node test are present and pass.",
  "findings": [
    {
      "evidence": "src/add.mjs exports add(a, b) and returns only a + b, with no side effects."
    },
    {
      "evidence": "test/add.test.mjs uses Node's built-in test and assert modules and covers positive, negative, mixed-sign, zero, and decimal inputs."
    },
    {
      "evidence": "node --test test/add.test.mjs completed with 5 passing tests and 0 failures."
    }
  ]
}

The team’s shape

step	does	done when
writer	Write the code and its test from the brief.	The files named in the brief exist in the workspace.
test	Run the test command against the written files.	The test command exits 0.
reviewer	Read the code and the test result and accept or send the work back.	A different model family has accepted the change.

The reviewer runs after the test, so it reads a result and not a promise. A rejection is a revision request aimed at writer, and maxKickbacks (default 1) bounds how many times the pair goes round.

Gotchas

The two seats must be different model families. The team refuses to build when writer.identity.modelFamily equals the reviewer’s. The check reads the identity you declare, so declare what runs.
A seat is an engine and its identity. The plugin runs the model; the identity is the record of what ran, and it is what the run record and the family check use.
The writer needs permission to write. A Claude or Codex seat started in a read-only mode returns a plan and no files, and the writer step then fails by name for the first missing file.
The test command runs in the workspace. Its arguments are relative to the directory you pass as workspace, which the file above sets to the current directory.
A writing seat can write anywhere the process can. The files above start their Claude and Codex seats with permission prompts off, which is what lets a model write files. Run them in a directory you are willing to let a model change, and give a reviewing seat a read-only mode if its plugin has one.

Source

The file is examples/teams/writer-reviewer-pair.ts. The teams package page lists every input.

Start

Concepts

Examples

Reviewing

Recording runs

Driving runs

Writing workflow shapes

Memory

Workspace

Core packages

Plugins

A writer and a reviewer

The file

What a run printed

The team’s shape

Gotchas

Source

​The file

​What a run printed

​The team’s shape

​Gotchas

​Source

The file

What a run printed

The team’s shape

Gotchas

Source