nucel-agent-claude-code

Claude Code provider for the Nucel Agent SDK — a subprocess wrapper for Anthropic's claude CLI.

Features

• Streaming JSONL protocol — parses system/init, assistant, result, and rate-limit events line-by-line
• Session resume — uses the official --resume <session_id> CLI flag
• Multi-turn — keeps the subprocess alive and writes prompts to stdin (start_interactive mode)
• Token tracking — input / output tokens reported by the CLI
• Budget enforcement — passes --max-budget-usd to the CLI and guards client-side before/after every query
• Timeout protection — configurable timeout per query (default: 10 minutes)
• Permission modes — Prompt, AcceptEdits, BypassPermissions, RejectAll → mapped to the official --permission-mode values
• max_turns control — passes --max-turns <n> (defaults to 1 for the initial spawn)
• Graceful shutdown — SIGTERM then SIGKILL fallback

How it works

Spawns the claude CLI in print mode with streaming JSON output:

claude \
  --model <model> \
  --permission-mode <mode> \
  --max-budget-usd <amount> \
  --system-prompt <prompt> \
  -p "<user prompt>" \
  --output-format stream-json \
  --verbose \
  --max-turns <n>

Then parses the JSONL stream and extracts:

• Text content from assistant messages
• Cost from result.total_cost_usd
• Token usage from result.usage
• Session ID from system/init (for resume)

For session resume, --resume <session_id> is added.

CLI flag mapping

Reference: Claude Code CLI docs.

Usage

[dependencies]
nucel-agent-sdk = "0.1"
tokio = { version = "1", features = ["full"] }

use nucel_agent_sdk::{ClaudeCodeExecutor, AgentExecutor, PermissionMode, SpawnConfig};
use std::path::Path;

#[tokio::main]
async fn main() -> Result<(), Box<dyn std::error::Error>> {
    let executor = ClaudeCodeExecutor::new();
    // Or supply an API key explicitly:
    // let executor = ClaudeCodeExecutor::with_api_key(std::env::var("ANTHROPIC_API_KEY")?);

    // Check availability
    let avail = executor.availability();
    if !avail.available {
        eprintln!("{}", avail.reason.unwrap_or_default());
        // → "`claude` CLI not found. Install: npm install -g @anthropic-ai/claude-code"
        return Ok(());
    }

    let session = executor.spawn(
        Path::new("/my/repo"),
        "Fix the failing tests in src/lib.rs",
        &SpawnConfig {
            model: Some("claude-opus-4-6".into()),
            budget_usd: Some(5.0),
            permission_mode: Some(PermissionMode::AcceptEdits),
            max_turns: Some(10),
            ..Default::default()
        },
    ).await?;

    let resp = session.query("Did the tests pass?").await?;
    println!("{}", resp.content);

    let cost = session.total_cost().await?;
    println!("Total cost: ${:.4}", cost.total_usd);

    session.close().await?;
    Ok(())
}

Capabilities

session_resume:     true   (via --resume <id>)
token_usage:        true
mcp_support:        true
autonomous_mode:    true
structured_output:  false

Requirements

• claude CLI on PATH: npm install -g @anthropic-ai/claude-code
• A valid ANTHROPIC_API_KEY env var or a logged-in Claude Max / Pro subscription (via claude login)

Source

• Repo: https://github.com/nucel-dev/agent-sdk
• Docs: https://docs.rs/nucel-agent-claude-code

License

Apache-2.0