Class CodexRunner

evals/buffbench/runners/codex.ts:5–143

import type { Runner, RunnerResult, AgentStep } from './runner'

export class CodexRunner implements Runner {
  private cwd: string
  private env: Record<string, string>

  constructor(cwd: string, env: Record<string, string> = {}) {
    this.cwd = cwd
    this.env = env
  }

  async run(prompt: string): Promise<RunnerResult> {
    const steps: AgentStep[] = []
    let totalCostUsd = 0

    return new Promise((resolve, reject) => {
      // Codex CLI uses the prompt as a positional argument
      // Use exec subcommand with --full-auto for automatic execution
      // --full-auto enables -a on-failure and --sandbox workspace-write
      // Use --json for structured output that we can parse
      const args = [
        'exec',
        '--full-auto',
        '--json',
        '-m',
        'gpt-5.1-codex',
        prompt,
      ]

      console.log(`[CodexRunner] Running: codex ${args.join(' ')}`)

      const child = spawn('codex', args, {
        cwd: this.cwd,
        env: {
          ...process.env,
          ...this.env,
          CODEX_API_KEY: process.env.OPENAI_API_KEY || this.env.OPENAI_API_KEY,
        },
        // Use 'ignore' for stdin to prevent the CLI from waiting for input
        stdio: ['ignore', 'pipe', 'pipe'],
      })

      let _stdout = ''
      let stderr = ''

      child.stdout.on('data', (data: Buffer) => {
        const chunk = data.toString()
        _stdout += chunk
        process.stdout.write(chunk)

        // Codex outputs events as JSON lines in some modes
        const lines = chunk.split('\n').filter((line) => line.trim())
        for (const line of lines) {
          try {
            const event = JSON.parse(line)
            if (event.type === 'message') {
              steps.push({
                type: 'text',
                text: event.content || event.message || '',
              })