nucel-agent-codex

Codex provider for Nucel Agent SDK — subprocess wrapper for the OpenAI codex CLI.

Features

• JSONL parsing — item.completed (agent_message) and turn.completed events
• Token tracking — input/output tokens from usage reports
• Budget enforcement — automatic cost checks before each query
• Structured output — supports JSON Schema output (Codex-specific)
• Model selection — configurable model per query

How it works

Spawns codex exec --experimental-json "<prompt>" as a subprocess, parses the JSONL output, and extracts:

• Agent text responses from item.completed → agent_message events
• Token usage from turn.completed → usage fields

Usage

[dependencies]
nucel-agent-sdk = "0.1"

use nucel_agent_sdk::{CodexExecutor, AgentExecutor, SpawnConfig};

let executor = CodexExecutor::new();

// Check availability
let avail = executor.availability();
if !avail.available {
    eprintln!("Install: npm install -g @openai/codex");
    return;
}

// Spawn and query
let session = executor.spawn(
    std::path::Path::new("/my/repo"),
    "Fix the failing tests",
    &SpawnConfig {
        model: Some("gpt-5-codex".into()),
        budget_usd: Some(3.0),
        ..Default::default()
    },
).await?;

let resp = session.query("Now add more tests").await?;
println!("{}", resp.content);

session.close().await?;

CLI Requirements

• codex CLI installed: npm install -g @openai/codex
• Valid CODEX_API_KEY or OpenAI API key

License

Apache-2.0