roboticus_cli/cli/admin/

setup.rs

use super::*;

// ── Setup wizard & starter skills ─────────────────────────────

/// Normalise a filesystem path for safe embedding in TOML double-quoted strings.
/// Converts Windows backslashes to forward slashes (valid on all platforms).
fn toml_safe_path(p: &str) -> String {
    p.replace('\\', "/")
}

pub const STARTER_SKILLS: &[(&str, &str)] = &[(
    "draft.md",
    r#"---
name: draft
description: Draft emails, messages, documents, or other written content
triggers:
  keywords: [draft, write, compose, email, letter, message, document]
priority: 6
---

Draft the requested content based on the user's description. Ask clarifying questions
if the audience, tone, or purpose is unclear. Default to a professional but approachable
tone unless told otherwise. Present the draft clearly and offer to revise.
"#,
)];

pub fn write_starter_skills(skills_dir: &std::path::Path) -> std::io::Result<usize> {
    let mut written = 0;
    for (filename, content) in STARTER_SKILLS {
        let path = skills_dir.join(filename);
        if !path.exists() {
            std::fs::write(&path, content)?;
            written += 1;
        }
    }
    Ok(written)
}

const APERTUS_8B_SUFFIX: &str = "apertus-8b-instruct:latest";
const APERTUS_70B_SUFFIX: &str = "apertus-70b-instruct:latest";

fn detect_system_ram_gb() -> Option<u64> {
    #[cfg(target_os = "linux")]
    {
        let output = std::process::Command::new("sh")
            .args(["-c", "awk '/MemTotal/ {print $2}' /proc/meminfo"])
            .output()
            .ok()?;
        if !output.status.success() {
            return None;
        }
        let kb = String::from_utf8_lossy(&output.stdout)
            .trim()
            .parse::<u64>()
            .ok()?;
        return Some(kb / 1024 / 1024);
    }

    #[cfg(target_os = "macos")]
    {
        let output = std::process::Command::new("sysctl")
            .args(["-n", "hw.memsize"])
            .output()
            .ok()?;
        if !output.status.success() {
            return None;
        }
        let bytes = String::from_utf8_lossy(&output.stdout)
            .trim()
            .parse::<u64>()
            .ok()?;
        return Some(bytes / 1024 / 1024 / 1024);
    }

    #[cfg(target_os = "windows")]
    {
        let output = std::process::Command::new("powershell")
            .args([
                "-NoProfile",
                "-Command",
                "(Get-CimInstance Win32_ComputerSystem).TotalPhysicalMemory",
            ])
            .output()
            .ok()?;
        if !output.status.success() {
            return None;
        }
        let bytes = String::from_utf8_lossy(&output.stdout)
            .trim()
            .parse::<u64>()
            .ok()?;
        return Some(bytes / 1024 / 1024 / 1024);
    }

    #[allow(unreachable_code)]
    None
}

fn aperture_options_for_provider(provider_prefix: &str, ram_gb: Option<u64>) -> Vec<String> {
    let mut options = vec![format!("{provider_prefix}/{APERTUS_8B_SUFFIX}")];
    if ram_gb.map(|v| v >= 64).unwrap_or(false) {
        options.push(format!("{provider_prefix}/{APERTUS_70B_SUFFIX}"));
    }
    options
}

fn has_hf_model_cache() -> bool {
    let home = roboticus_core::home_dir();
    let home_str = home.to_string_lossy().to_string();
    let hf_home = std::env::var("HF_HOME")
        .ok()
        .or_else(|| Some(format!("{home_str}/.cache/huggingface")));
    let hub_dir = match hf_home {
        Some(v) => std::path::PathBuf::from(v).join("hub"),
        None => return false,
    };
    if !hub_dir.exists() {
        return false;
    }
    std::fs::read_dir(&hub_dir)
        .ok()
        .map(|iter| {
            iter.filter_map(Result::ok)
                .any(|entry| entry.file_name().to_string_lossy().starts_with("models--"))
        })
        .unwrap_or(false)
}

fn has_ollama_models() -> bool {
    if which_binary("ollama").is_none() {
        return false;
    }
    let output = std::process::Command::new("ollama").arg("list").output();
    let Ok(out) = output else {
        return false;
    };
    if !out.status.success() {
        return false;
    }
    let line_count = String::from_utf8_lossy(&out.stdout).lines().count();
    line_count > 1
}

const LOCAL_MODEL_FRAMEWORKS: &[&str] = &[
    "sglang",
    "vllm",
    "docker",
    "ollama",
    "llama-server",
    "llama_cpp",
];

fn has_framework_in_path(path_var: &std::ffi::OsStr) -> bool {
    LOCAL_MODEL_FRAMEWORKS
        .iter()
        .any(|bin| crate::cli::which_binary_in_path(bin, path_var).is_some())
}

fn has_existing_local_model_stack() -> bool {
    let has_framework = std::env::var_os("PATH")
        .map(|p| has_framework_in_path(&p))
        .unwrap_or(false);
    has_framework || has_ollama_models() || has_hf_model_cache()
}

fn run_quick_personality_setup(
    workspace: &std::path::Path,
) -> Result<(), Box<dyn std::error::Error>> {
    let (DIM, BOLD, ACCENT, GREEN, YELLOW, RED, CYAN, RESET, MONO) = colors();
    let (OK, ACTION, WARN, DETAIL, ERR) = icons();
    use dialoguer::{Input, Select};

    let name: String = Input::new()
        .with_prompt("  Agent name")
        .default("Roboticus".into())
        .interact_text()?;

    let formality_options = vec!["formal", "balanced", "casual"];
    let formality_idx = Select::new()
        .with_prompt("  Communication style")
        .items(&formality_options)
        .default(1)
        .interact()?;

    let proactive_options = vec![
        "wait (only act when told)",
        "suggest (flag opportunities, ask first)",
        "initiative (act proactively)",
    ];
    let proactive_idx = Select::new()
        .with_prompt("  Proactiveness level")
        .items(&proactive_options)
        .default(1)
        .interact()?;
    let proactive_val = match proactive_idx {
        0 => "wait",
        2 => "initiative",
        _ => "suggest",
    };

    let domain_options = vec!["general", "developer", "business", "creative", "research"];
    let domain_idx = Select::new()
        .with_prompt("  Primary domain")
        .items(&domain_options)
        .default(0)
        .interact()?;

    let boundaries: String = Input::new()
        .with_prompt(
            "  Any hard boundaries? (topics/actions that are off-limits, or press Enter to skip)",
        )
        .allow_empty(true)
        .interact_text()?;

    let os_toml = roboticus_core::personality::generate_os_toml(
        &name,
        formality_options[formality_idx],
        proactive_val,
        domain_options[domain_idx],
    );
    let fw_toml = roboticus_core::personality::generate_firmware_toml(&boundaries);

    std::fs::create_dir_all(workspace)?;
    std::fs::write(workspace.join("OS.toml"), &os_toml)?;
    std::fs::write(workspace.join("FIRMWARE.toml"), &fw_toml)?;

    println!("  {OK} Personality configured for {BOLD}{name}{RESET} (OS.toml + FIRMWARE.toml)");

    Ok(())
}

pub fn cmd_setup() -> Result<(), Box<dyn std::error::Error>> {
    let (DIM, BOLD, ACCENT, GREEN, YELLOW, RED, CYAN, RESET, MONO) = colors();
    let (OK, ACTION, WARN, DETAIL, ERR) = icons();
    use dialoguer::{Confirm, Input, Select};

    println!("\n  {BOLD}Roboticus Setup Wizard{RESET}\n");
    println!("  This wizard will help you create an roboticus.toml configuration.\n");

    // Prerequisites: Go + gosh (plugin scripting engine)
    println!("  {BOLD}Checking prerequisites...{RESET}\n");
    let go_bin = which_binary("go");
    let has_go = go_bin.is_some();
    let has_gosh = which_binary("gosh").is_some();

    if !has_go {
        println!("  {WARN} Go is not installed (required for the gosh plugin engine).");
        println!(
            "     Install from {CYAN}https://go.dev/dl/{RESET} or: {MONO}brew install go{RESET}"
        );
        println!();
        let proceed = Confirm::new()
            .with_prompt(
                "  Continue without Go? (plugins won't work until Go + gosh are installed)",
            )
            .default(true)
            .interact()?;
        if !proceed {
            println!("\n  Setup paused. Install Go, then re-run {BOLD}roboticus init{RESET}.\n");
            return Ok(());
        }
    } else if !has_gosh {
        println!("  {OK} Go found");
        println!("  {WARN} gosh scripting engine not found.");
        let install_now = Confirm::new()
            .with_prompt("  Install gosh now via `go install`?")
            .default(true)
            .interact()?;
        if install_now {
            println!("  Installing gosh...");
            let result = if let Some(go_path) = go_bin.as_deref() {
                std::process::Command::new(go_path)
                    .args(["install", "github.com/drewwalton19216801/gosh@latest"])
                    .status()
            } else {
                Err(std::io::Error::new(
                    std::io::ErrorKind::NotFound,
                    "go binary not found",
                ))
            };
            match result {
                Ok(s) if s.success() => {
                    println!("  {OK} gosh installed successfully");
                }
                _ => {
                    println!("  {WARN} gosh installation failed. Install manually:");
                    println!(
                        "     {MONO}go install github.com/drewwalton19216801/gosh@latest{RESET}"
                    );
                }
            }
        } else {
            println!(
                "  Skipped. Install later: {MONO}go install github.com/drewwalton19216801/gosh@latest{RESET}"
            );
        }
    } else {
        println!("  {OK} Go found");
        println!("  {OK} gosh scripting engine found");
    }
    println!();

    // 1. Agent name
    let agent_name: String = Input::new()
        .with_prompt("  Agent name")
        .default("Roboticus".into())
        .interact_text()?;

    let offer_apertus_onboarding = !has_existing_local_model_stack();
    if !offer_apertus_onboarding {
        println!(
            "  {DETAIL} Existing local model framework/model cache detected; skipping automatic SGLang + Apertus recommendation."
        );
    }

    // 2. LLM provider
    let providers = if offer_apertus_onboarding {
        vec![
            "SGLang (local, recommended for Apertus)",
            "vLLM (local)",
            "Docker Model Runner (local)",
            "Ollama (local)",
            "OpenAI",
            "Anthropic",
            "Google AI",
            "Moonshot",
            "OpenRouter",
            "llama-cpp (local)",
        ]
    } else {
        vec![
            "Ollama (local)",
            "SGLang (local)",
            "vLLM (local)",
            "Docker Model Runner (local)",
            "OpenAI",
            "Anthropic",
            "Google AI",
            "Moonshot",
            "OpenRouter",
            "llama-cpp (local)",
        ]
    };
    let provider_idx = Select::new()
        .with_prompt("  Select LLM provider")
        .items(&providers)
        .default(if offer_apertus_onboarding { 0 } else { 4 })
        .interact()?;

    let (provider_prefix, needs_api_key) = match (offer_apertus_onboarding, provider_idx) {
        (true, 0) => ("sglang", false),
        (true, 1) => ("vllm", false),
        (true, 2) => ("docker-model-runner", false),
        (true, 3) => ("ollama", false),
        (true, 4) => ("openai", true),
        (true, 5) => ("anthropic", true),
        (true, 6) => ("google", true),
        (true, 7) => ("moonshot", true),
        (true, 8) => ("openrouter", true),
        (true, 9) => ("llama-cpp", false),
        (false, 0) => ("ollama", false),
        (false, 1) => ("sglang", false),
        (false, 2) => ("vllm", false),
        (false, 3) => ("docker-model-runner", false),
        (false, 4) => ("openai", true),
        (false, 5) => ("anthropic", true),
        (false, 6) => ("google", true),
        (false, 7) => ("moonshot", true),
        (false, 8) => ("openrouter", true),
        (false, 9) => ("llama-cpp", false),
        _ => ("openai", true),
    };

    // 3. API key
    let api_key = if needs_api_key {
        let key: String = Input::new()
            .with_prompt("  API key (or press Enter to set later)")
            .allow_empty(true)
            .interact_text()?;
        if key.is_empty() { None } else { Some(key) }
    } else {
        None
    };

    // 4. Model selection
    let ram_gb = detect_system_ram_gb();
    let model = match provider_prefix {
        "sglang" | "vllm" | "docker-model-runner" | "ollama" => {
            if offer_apertus_onboarding {
                if let Some(ram) = ram_gb {
                    println!("  {DETAIL} Detected system RAM: {ram} GB");
                } else {
                    println!(
                        "  {WARN} Could not detect system RAM. Only 8B Apertus is recommended by default."
                    );
                }

                match provider_prefix {
                    "sglang" if which_binary("sglang").is_none() => {
                        println!("  {WARN} sglang binary not found.");
                        let install_now = Confirm::new()
                            .with_prompt("  Install SGLang now via pip? (recommended for Apertus)")
                            .default(true)
                            .interact()?;
                        if install_now {
                            let py_bin = which_binary("python3")
                                .or_else(|| which_binary("python"))
                                .unwrap_or_else(|| "python3".into());
                            let status = std::process::Command::new(py_bin)
                                .args(["-m", "pip", "install", "--user", "sglang[all]"])
                                .status();
                            if status.as_ref().map(|s| s.success()).unwrap_or(false) {
                                println!("  {OK} SGLang install completed.");
                            } else {
                                println!(
                                    "  {WARN} SGLang install failed. You can install it later and keep this model selection."
                                );
                            }
                        }
                    }
                    "vllm" if which_binary("vllm").is_none() => {
                        println!("  {WARN} vllm command not found.");
                        let install_now = Confirm::new()
                            .with_prompt("  Install vLLM now via pip?")
                            .default(false)
                            .interact()?;
                        if install_now {
                            let py_bin = which_binary("python3")
                                .or_else(|| which_binary("python"))
                                .unwrap_or_else(|| "python3".into());
                            let status = std::process::Command::new(py_bin)
                                .args(["-m", "pip", "install", "--user", "vllm"])
                                .status();
                            if status.as_ref().map(|s| s.success()).unwrap_or(false) {
                                println!("  {OK} vLLM install completed.");
                            } else {
                                println!(
                                    "  {WARN} vLLM install failed. You can install it later and keep this model selection."
                                );
                            }
                        }
                    }
                    "docker-model-runner" if which_binary("docker").is_none() => {
                        println!("  {WARN} Docker not found. Docker Model Runner requires Docker.");
                    }
                    "ollama" if which_binary("ollama").is_none() => {
                        println!(
                            "  {WARN} Ollama not found. Install from https://ollama.ai to run local models."
                        );
                    }
                    _ => {}
                }

                let model_options = aperture_options_for_provider(provider_prefix, ram_gb);
                let model_idx = Select::new()
                    .with_prompt("  Select Apertus model")
                    .items(&model_options)
                    .default(0)
                    .interact()?;
                model_options[model_idx].clone()
            } else {
                let default_model = match provider_prefix {
                    "ollama" => "ollama/qwen3:8b",
                    "sglang" => "sglang/default",
                    "vllm" => "vllm/default",
                    "docker-model-runner" => "docker-model-runner/default",
                    _ => "ollama/qwen3:8b",
                };
                Input::new()
                    .with_prompt("  Model")
                    .default(default_model.into())
                    .interact_text()?
            }
        }
        _ => {
            let default_model = match provider_prefix {
                "openai" => "openai/gpt-4o",
                "anthropic" => "anthropic/claude-sonnet-4-20250514",
                "google" => "google/gemini-3.1-pro-preview",
                "moonshot" => "moonshot/kimi-k2.5",
                "openrouter" => "openrouter/google/gemini-3.1-pro-preview",
                "llama-cpp" => "llama-cpp/default",
                _ => "sglang/apertus-8b-instruct:latest",
            };
            Input::new()
                .with_prompt("  Model")
                .default(default_model.into())
                .interact_text()?
        }
    };

    // 5. Server port
    let port: String = Input::new()
        .with_prompt("  Server port")
        .default("18789".into())
        .interact_text()?;
    let port_num: u16 = port.parse().unwrap_or(18789);

    // 6. Channels
    let enable_telegram = Confirm::new()
        .with_prompt("  Enable Telegram channel?")
        .default(false)
        .interact()?;

    let (telegram_token, telegram_chat_ids) = if enable_telegram {
        let token: String = Input::new()
            .with_prompt("  Telegram bot token")
            .interact_text()?;
        println!("  Tip: to find your Telegram chat ID, message @userinfobot on Telegram.");
        let chat_ids_raw: String = Input::new()
            .with_prompt("  Allowed Telegram chat IDs (comma-separated, or empty to allow all)")
            .default(String::new())
            .allow_empty(true)
            .interact_text()?;
        let chat_ids: Vec<String> = chat_ids_raw
            .split(',')
            .map(|s| s.trim().to_string())
            .filter(|s| !s.is_empty())
            .collect();
        (Some(token), chat_ids)
    } else {
        (None, Vec::new())
    };

    let enable_discord = Confirm::new()
        .with_prompt("  Enable Discord channel?")
        .default(false)
        .interact()?;

    let discord_token = if enable_discord {
        let token: String = Input::new()
            .with_prompt("  Discord bot token")
            .interact_text()?;
        Some(token)
    } else {
        None
    };

    // 7. Workspace directory
    let home = roboticus_core::home_dir();
    let home = home.to_string_lossy();
    let default_workspace = format!("{home}/.roboticus/workspace");
    let workspace: String = Input::new()
        .with_prompt("  Workspace directory")
        .default(default_workspace)
        .interact_text()?;

    // 8. Database path
    let default_db = format!("{home}/.roboticus/state.db");
    let db_path: String = Input::new()
        .with_prompt("  Database path")
        .default(default_db)
        .interact_text()?;

    // Generate config
    let mut config = String::new();
    config.push_str("# Roboticus Configuration (generated by onboard wizard)\n\n");
    config.push_str("[agent]\n");
    config.push_str(&format!("name = \"{agent_name}\"\n"));
    config.push_str(&format!(
        "id = \"{}\"\n",
        agent_name.to_lowercase().replace(' ', "-")
    ));
    config.push_str(&format!("workspace = \"{}\"\n", toml_safe_path(&workspace)));
    config.push_str("log_level = \"info\"\n\n");

    let server_api_key = roboticus_core::config_utils::generate_server_api_key();
    config.push_str("[server]\n");
    config.push_str(&format!("port = {port_num}\n"));
    config.push_str("bind = \"localhost\"\n");
    config.push_str(&format!("api_key = \"{server_api_key}\"\n\n"));

    config.push_str("[database]\n");
    config.push_str(&format!("path = \"{}\"\n\n", toml_safe_path(&db_path)));

    config.push_str("[models]\n");
    config.push_str(&format!("primary = \"{model}\"\n"));
    config.push_str("fallbacks = []\n\n");

    config.push_str("[models.routing]\n");
    config.push_str("mode = \"metascore\"\n");
    config.push_str("confidence_threshold = 0.9\n");
    config.push_str("local_first = true\n\n");

    config.push_str(
        "# Bundled provider defaults (sglang, vllm, docker-model-runner, ollama, openai, anthropic, google, openrouter)\n",
    );
    config.push_str("# are auto-merged. Override or add new providers here.\n");
    if api_key.is_some() {
        config.push_str(&format!(
            "# Set the API key via env: {}_API_KEY\n\n",
            provider_prefix.to_uppercase()
        ));
    } else {
        config.push('\n');
    }

    config.push_str("[memory]\n");
    config.push_str("working_budget_pct = 30.0\n");
    config.push_str("episodic_budget_pct = 25.0\n");
    config.push_str("semantic_budget_pct = 20.0\n");
    config.push_str("procedural_budget_pct = 15.0\n");
    config.push_str("relationship_budget_pct = 10.0\n\n");

    config.push_str("[treasury]\n");
    config.push_str("per_payment_cap = 100.0\n");
    config.push_str("hourly_transfer_limit = 500.0\n");
    config.push_str("daily_transfer_limit = 2000.0\n");
    config.push_str("minimum_reserve = 5.0\n");
    config.push_str("daily_inference_budget = 50.0\n\n");

    if let Some(ref token) = telegram_token {
        config.push_str("[channels.telegram]\n");
        config.push_str(&format!("token = \"{token}\"\n"));
        if telegram_chat_ids.is_empty() {
            config.push_str("# allowed_chat_ids = []  # Tip: message @userinfobot on Telegram to find your chat ID\n\n");
        } else {
            config.push_str(&format!(
                "allowed_chat_ids = [{}]\n\n",
                telegram_chat_ids
                    .iter()
                    .map(|id| id.to_string())
                    .collect::<Vec<_>>()
                    .join(", ")
            ));
        }
    }

    if let Some(ref token) = discord_token {
        config.push_str("[channels.discord]\n");
        config.push_str(&format!("token = \"{token}\"\n\n"));
    }

    config.push_str("[skills]\n");
    config.push_str(&format!(
        "skills_dir = \"{}\"\n\n",
        toml_safe_path(&format!("{home}/.roboticus/skills"))
    ));

    config.push_str("[a2a]\n");
    config.push_str("enabled = true\n");

    // Write config
    let config_path = "roboticus.toml";
    let is_first_install = !std::path::Path::new(config_path).exists();
    if !is_first_install {
        let overwrite = Confirm::new()
            .with_prompt("  roboticus.toml already exists. Overwrite?")
            .default(false)
            .interact()?;
        if !overwrite {
            println!("\n  Aborted. Existing config preserved.\n");
            return Ok(());
        }
    }

    std::fs::write(config_path, &config)?;
    println!("\n  {OK} Configuration written to {config_path}");
    println!(
        "  {DETAIL} A random {BOLD}server.api_key{RESET} was generated for REST / MCP clients (see [server] in the file)."
    );

    // Create workspace dir
    let ws_path = std::path::Path::new(&workspace);
    if !ws_path.exists() {
        std::fs::create_dir_all(ws_path)?;
        println!("  {OK} Created workspace: {workspace}");
    }

    // Create skills dir with starter skills
    let skills_path = format!("{home}/.roboticus/skills");
    let sp = std::path::Path::new(&skills_path);
    if !sp.exists() {
        std::fs::create_dir_all(sp)?;
    }
    let skills_written = write_starter_skills(sp)?;
    if skills_written > 0 {
        println!("  {ACTION} Created {skills_written} starter skills");
    } else {
        println!("  {OK} Skills directory ready");
    }

    // Personality setup
    println!("\n  {BOLD}Personality Setup{RESET}\n");
    let personality_options = vec![
        "Keep Roboticus (recommended default)",
        "Quick setup (5 questions)",
        "Full interview (guided conversation with your agent)",
    ];
    let personality_idx = Select::new()
        .with_prompt("  How would you like to configure your agent's personality?")
        .items(&personality_options)
        .default(0)
        .interact()?;

    match personality_idx {
        0 => {
            roboticus_core::personality::write_defaults(ws_path)?;
            println!("  {OK} Roboticus personality loaded (OS.toml + FIRMWARE.toml)");
        }
        1 => {
            run_quick_personality_setup(ws_path)?;
        }
        2 => {
            let basic_name: String = Input::new()
                .with_prompt("  Agent name")
                .default(agent_name.clone())
                .interact_text()?;
            let domains = vec!["general", "developer", "business", "creative", "research"];
            let domain_idx = Select::new()
                .with_prompt("  Primary domain")
                .items(&domains)
                .default(0)
                .interact()?;

            let starter_os = roboticus_core::personality::generate_os_toml(
                &basic_name,
                "balanced",
                "suggest",
                domains[domain_idx],
            );
            std::fs::write(ws_path.join("OS.toml"), &starter_os)?;
            roboticus_core::personality::write_defaults(ws_path)?;

            println!();
            println!("  {OK} Starter personality written.");
            println!("  {DETAIL} Start your agent:  {BOLD}roboticus serve{RESET}");
            println!("  {DETAIL} Then send it:      {BOLD}/interview{RESET}");
            println!("  {DETAIL} The agent will walk you through a deep personality interview.");
        }
        _ => {}
    }

    // On first install, explicitly ask whether to run the interview flow.
    if is_first_install && personality_idx != 2 {
        let do_interview = Confirm::new()
            .with_prompt("  Run the guided personality interview now? (recommended)")
            .default(true)
            .interact()?;
        if do_interview {
            println!();
            println!("  {DETAIL} Start your agent:  {BOLD}roboticus serve{RESET}");
            println!("  {DETAIL} Then send it:      {BOLD}/interview{RESET}");
            println!("  {DETAIL} The agent will walk you through a deep personality interview.");
        }
    }

    println!();
    println!("  {OK} Setup complete! Run {BOLD}roboticus serve{RESET} to start.");
    println!();

    Ok(())
}

#[cfg(test)]
mod tests {
    use super::*;
    use crate::test_support::EnvGuard;

    #[test]
    fn write_starter_skills_is_idempotent() {
        let dir = tempfile::tempdir().unwrap();
        let first = write_starter_skills(dir.path()).unwrap();
        assert_eq!(first, STARTER_SKILLS.len());

        let second = write_starter_skills(dir.path()).unwrap();
        assert_eq!(second, 0);
    }

    #[test]
    fn aperture_options_include_70b_only_for_high_ram() {
        let low = aperture_options_for_provider("sglang", Some(32));
        assert!(low.iter().any(|m| m.ends_with(APERTUS_8B_SUFFIX)));
        assert!(!low.iter().any(|m| m.ends_with(APERTUS_70B_SUFFIX)));

        let high = aperture_options_for_provider("sglang", Some(128));
        assert!(high.iter().any(|m| m.ends_with(APERTUS_8B_SUFFIX)));
        assert!(high.iter().any(|m| m.ends_with(APERTUS_70B_SUFFIX)));
    }

    #[test]
    fn starter_skills_contain_expected_frontmatter() {
        for (name, content) in STARTER_SKILLS {
            assert!(name.ends_with(".md"));
            assert!(content.starts_with("---"));
            assert!(content.contains("name:"));
            assert!(content.contains("description:"));
            assert!(content.contains("triggers:"));
        }
    }

    #[serial_test::serial]
    #[test]
    fn has_hf_model_cache_detects_models_directory() {
        let dir = tempfile::tempdir().unwrap();
        let hf_home = dir.path().join(".cache").join("huggingface");
        let hub = hf_home.join("hub");
        std::fs::create_dir_all(&hub).unwrap();
        std::fs::create_dir_all(hub.join("models--test--foo")).unwrap();
        let _guard = EnvGuard::set("HF_HOME", hf_home.to_str().unwrap());
        assert!(has_hf_model_cache());
    }

    #[serial_test::serial]
    #[test]
    fn has_hf_model_cache_false_when_unset_or_empty() {
        let dir = tempfile::tempdir().unwrap();
        let hf_home = dir.path().join("empty_hf");
        std::fs::create_dir_all(&hf_home).unwrap();
        let _guard = EnvGuard::set("HF_HOME", hf_home.to_str().unwrap());
        assert!(!has_hf_model_cache());
    }

    #[test]
    fn has_existing_local_model_stack_false_with_no_tools_or_cache() {
        let empty_path = tempfile::tempdir().unwrap();
        let path_var = std::ffi::OsString::from(empty_path.path().to_str().unwrap());
        assert!(!has_framework_in_path(&path_var));
        // HF cache with no models-– dir returns false.
        let hf_home = tempfile::tempdir().unwrap();
        assert!(!hf_home.path().join("hub").exists());
    }

    #[test]
    fn detect_system_ram_gb_returns_reasonable_value_when_present() {
        let ram = detect_system_ram_gb();
        if let Some(v) = ram {
            assert!(v > 0);
        }
    }
}