What causes "Agent exited with status: {}" in nikivdev/code?

Re-run the agent command directly (e.g. `claude ...` or `gen ...`) to see its stderr/stdout.

Agent exited with status: {}

Error message

Agent exited with status: {}

What it means

After building the full prompt and dispatching to the chosen backend (Claude Code, gen, etc.), run_agent checks the child process exit status. A non-zero status means the agent backend ran but failed, so run_agent bails with 'Agent exited with status: <status>'. The message carries only the exit code, not the backend's output.

Source

Thrown at src/agents.rs:909

    println!("Invoking {} agent...\n", agent);

    let (tool, model) = get_agent_config();
    let status = match tool.as_str() {
        "claude" => invoke_claude(&full_prompt)?,
        "opencode" => invoke_opencode(&full_prompt, model.as_deref())?,
        _ => {
            let gen_loc = find_gen().ok_or_else(|| {
                anyhow::anyhow!(
                    "gen not found. Install with:\n  cd {} && f install\n  # or set GEN_REPO env var",
                    gen_repo_hint()
                )
            })?;
            invoke_gen(&gen_loc, &full_prompt)?
        }
    };

    if !status.success() {
        bail!("Agent exited with status: {}", status);
    }

    Ok(())
}

/// Invoke Claude Code with a prompt.
fn invoke_claude(prompt: &str) -> Result<std::process::ExitStatus> {
    Command::new("claude")
        .args(["-p", prompt])
        .stdin(Stdio::inherit())
        .stdout(Stdio::inherit())
        .stderr(Stdio::inherit())
        .status()
        .context("failed to run claude")
}

/// Invoke opencode with a prompt and optional model.
fn invoke_opencode(prompt: &str, model: Option<&str>) -> Result<std::process::ExitStatus> {

View on GitHub (pinned to a747e741ae)

Solutions

Re-run the agent command directly (e.g. `claude ...` or `gen ...`) to see its stderr/stdout.
Authenticate the backend: run its login flow or set the required API key env var (e.g. ANTHROPIC_API_KEY).
Verify the backend binary is installed and up to date.
Retry with a smaller/simpler prompt if the backend crashed on input size.

Example fix

// before
$ f agents run claude "review"
Error: Agent exited with status: exit status: 1
// after
$ claude --version && claude login   # fix backend auth/install
$ f agents run claude "review"       # now succeeds

Defensive patterns

Strategy: retry

Validate before calling

command -v claude >/dev/null && claude --version >/dev/null 2>&1 \
  && f agents run claude "prompt" || echo "backend agent not ready"

Type guard

fn backend_ready(cmd: &str) -> bool {
    std::process::Command::new(cmd).arg("--version").output()
        .map(|o| o.status.success()).unwrap_or(false)
}

Try / catch

match run_agent(agent, prompt) {
    Err(e) if e.to_string().starts_with("Agent exited with status") => {
        eprintln!("Backend '{agent}' failed; run it directly for details.");
        // optionally retry after fixing auth/install
    }
    other => other?,
}

Prevention

Authenticate backend CLIs (login flows, API key env vars) before use.
Verify the backend binary is installed and current.
For flaky/network backends, retry with backoff.
Check backend logs by running the underlying command directly on failure.

When it happens

Trigger: The invoked agent backend (claude, gen, or a custom binary) exits non-zero — e.g. auth failure, missing API key, crashed model CLI, or the prompt caused a fatal backend error.

Common situations: Expired or missing API credentials for the backend agent CLI; backend not installed despite the name resolving; network failure reaching the model provider; backend OOM/crash on a large prompt.

Related errors

AI-assisted analysis of nikivdev/code@a747e741ae (2026-09-01). Data as JSON: /api/errors/064bb05d4ea2b4f8. Report an issue: GitHub.