openlatch-client 0.1.6

/// `openlatch doctor` command handler.
///
/// Runs diagnostic checks and reports results.
/// All path references use `config::openlatch_dir()` per PLAT-02.
///
/// `--fix`, `--restore`, and `--rescue` dispatch to sibling modules
/// (`doctor_fix`, `doctor_restore`, `doctor_rescue`). The shared
/// `run_all_checks` helper returns a structured `DoctorReport` so those
/// modules can re-run diagnostics for before/after deltas without
/// duplicating ~600 LOC of probe logic.
use crate::cli::commands::lifecycle;
use crate::cli::commands::{doctor_fix, doctor_rescue, doctor_restore};
use crate::cli::output::{OutputConfig, OutputFormat};
use crate::cli::DoctorArgs;
use crate::config;
use crate::error::OlError;
use crate::hooks;

/// A single doctor check result.
#[derive(Clone)]
pub(crate) struct DoctorCheck {
    pub pass: bool,
    pub message: String,
}

impl DoctorCheck {
    pub(crate) fn pass(message: impl Into<String>) -> Self {
        Self {
            pass: true,
            message: message.into(),
        }
    }

    pub(crate) fn fail(message: impl Into<String>) -> Self {
        Self {
            pass: false,
            message: message.into(),
        }
    }
}

/// Aggregate diagnostic snapshot returned by [`run_all_checks`].
///
/// Carries enough state for `--fix` and `--rescue` flows to inspect
/// daemon liveness and the originally-configured port without re-probing.
#[allow(dead_code)] // fields consumed by doctor_fix / doctor_rescue (steps 4–7)
pub(crate) struct DoctorReport {
    pub checks: Vec<DoctorCheck>,
    pub issues: Vec<String>,
    pub daemon_alive: bool,
    pub daemon_uptime_secs: Option<u64>,
    pub port: u16,
}

#[allow(dead_code)] // methods consumed by doctor_fix (step 4) for before/after deltas
impl DoctorReport {
    /// Convenience: `true` when every check passed.
    pub(crate) fn all_pass(&self) -> bool {
        self.checks.iter().all(|c| c.pass)
    }

    /// Number of failing checks.
    pub(crate) fn fail_count(&self) -> usize {
        self.checks.iter().filter(|c| !c.pass).count()
    }
}

/// Run the `openlatch doctor` command.
///
/// Default invocation (no flags): runs diagnostic checks and reports results.
/// `--fix`, `--restore`, `--rescue` dispatch to the matching sibling module.
/// Combined `--fix --rescue` and `--restore --rescue` are allowed —
/// rescue runs first to snapshot pre-fix state.
///
/// # Errors
///
/// Returns an error only if a fix/restore/rescue helper itself fails —
/// individual diagnostic check failures are reported, not returned.
pub fn run_doctor(args: &DoctorArgs, output: &OutputConfig) -> Result<(), OlError> {
    // Hidden self-test: deliberately panic to exercise the Sentry pipeline.
    // Used once per release during smoke validation — see brainstorm
    // Decision 14. Panics after crash-report init has already completed
    // at the top of main(), so the panic hook is live.
    if args.trigger_panic {
        panic!("openlatch crash-report validation panic");
    }

    // Combined-flag composition: rescue first to capture pre-fix state.
    if args.rescue && args.fix {
        doctor_rescue::run(args, output, /* fix_applied_after = */ true)?;
        return doctor_fix::run(args, output);
    }
    if args.rescue && args.restore {
        doctor_rescue::run(args, output, false)?;
        return doctor_restore::run(args, output);
    }
    if args.fix {
        return doctor_fix::run(args, output);
    }
    if args.restore {
        return doctor_restore::run(args, output);
    }
    if args.rescue {
        return doctor_rescue::run(args, output, false);
    }

    crate::cli::header::print(output, &["doctor"]);
    if output.format == OutputFormat::Human && !output.quiet {
        eprintln!();
    }

    let report = run_all_checks(output)?;
    print_diagnostic_results(&report, output);
    Ok(())
}

/// Run every diagnostic check and assemble a [`DoctorReport`].
///
/// Pure read-only — never mutates state. Reusable by `doctor_fix` (for
/// before/after deltas) and `doctor_rescue` (for the bundled health
/// snapshot).
pub(crate) fn run_all_checks(_output: &OutputConfig) -> Result<DoctorReport, OlError> {
    let cfg = config::Config::load(None, None, false)?;
    let ol_dir = config::openlatch_dir();

    let mut issues: Vec<String> = Vec::new();
    let mut checks: Vec<DoctorCheck> = Vec::new();

    // Check 1: Agent detection
    match hooks::detect_agent() {
        Ok(agent) => {
            let label = match &agent {
                hooks::DetectedAgent::ClaudeCode { claude_dir, .. } => {
                    format!("Claude Code ({})", claude_dir.display())
                }
            };
            checks.push(DoctorCheck::pass(format!("Agent detected: {label}")));
        }
        Err(e) => {
            let msg = format!("Agent not found: {} ({})", e.message, e.code);
            checks.push(DoctorCheck::fail(msg.clone()));
            issues.push("No AI agent detected. Install Claude Code to use OpenLatch.".to_string());
        }
    }

    // Check 2: Config file exists and is parseable (PLAT-02: OS-aware path)
    let config_path = ol_dir.join("config.toml");
    if config_path.exists() {
        match config::Config::load(None, None, false) {
            Ok(_) => {
                checks.push(DoctorCheck::pass(format!(
                    "Config file: {}",
                    config_path.display()
                )));
            }
            Err(e) => {
                checks.push(DoctorCheck::fail(format!(
                    "Config file invalid: {} ({})",
                    e.message, e.code
                )));
                issues.push(format!(
                    "Config file '{}' has errors. Delete and re-run 'openlatch init'.",
                    config_path.display()
                ));
            }
        }
    } else {
        checks.push(DoctorCheck::fail(format!(
            "Config file missing: {}",
            config_path.display()
        )));
        issues.push("Config file missing. Run 'openlatch init' to create it.".to_string());
    }

    // Check 2b: Crash report consent state (panic reporting to Sentry).
    // Diagnostic only — never a fail condition. Reports the resolved state so
    // users can verify their opt-out (`SENTRY_DISABLED=1` or
    // `[crashreport] enabled = false`) actually landed.
    #[cfg(feature = "crash-report")]
    {
        let resolved = crate::crash_report::current_state(&ol_dir);
        let label = match resolved.decided_by {
            crate::crash_report::consent::DecidedBy::SentryDisabledEnv => {
                "off (SENTRY_DISABLED env)"
            }
            crate::crash_report::consent::DecidedBy::NoBakedDsn => "off (no DSN baked)",
            crate::crash_report::consent::DecidedBy::ConfigFile => {
                if resolved.enabled() {
                    "on (config.toml)"
                } else {
                    "off (config.toml)"
                }
            }
            crate::crash_report::consent::DecidedBy::DefaultEnabled => "on (default)",
        };
        checks.push(DoctorCheck::pass(format!("Crash reporting: {label}")));
    }
    #[cfg(not(feature = "crash-report"))]
    {
        checks.push(DoctorCheck::pass(
            "Crash reporting: not compiled in".to_string(),
        ));
    }

    // Check 3: Auth token exists and non-empty (PLAT-02: OS-aware path)
    let token_path = ol_dir.join("daemon.token");
    let mut daemon_token: Option<String> = None;
    if token_path.exists() {
        match std::fs::read_to_string(&token_path) {
            Ok(content) if !content.trim().is_empty() => {
                daemon_token = Some(content.trim().to_string());
                checks.push(DoctorCheck::pass(format!(
                    "Auth token: {} (valid)",
                    token_path.display()
                )));
            }
            Ok(_) => {
                checks.push(DoctorCheck::fail(format!(
                    "Auth token empty: {}",
                    token_path.display()
                )));
                issues.push("Auth token is empty. Run 'openlatch init' to regenerate.".to_string());
            }
            Err(e) => {
                checks.push(DoctorCheck::fail(format!(
                    "Auth token unreadable: {} — {e}",
                    token_path.display()
                )));
                issues.push(format!(
                    "Cannot read token file '{}'. Check file permissions.",
                    token_path.display()
                ));
            }
        }
    } else {
        checks.push(DoctorCheck::fail(format!(
            "Auth token missing: {}",
            token_path.display()
        )));
        issues.push("Auth token missing. Run 'openlatch init' to generate one.".to_string());
    }

    // Check 4: Daemon reachable via /health
    let pid = lifecycle::read_pid_file();
    let daemon_alive = pid.map(lifecycle::is_process_alive).unwrap_or(false);
    if daemon_alive {
        let url = format!("http://127.0.0.1:{}/health", cfg.port);
        let reachable = reqwest::blocking::get(&url)
            .map(|r| r.status().is_success())
            .unwrap_or(false);
        if reachable {
            checks.push(DoctorCheck::pass(format!(
                "Daemon: running on port {} (PID {})",
                cfg.port,
                pid.unwrap()
            )));
        } else {
            checks.push(DoctorCheck::fail(format!(
                "Daemon: process alive (PID {}) but /health unreachable on port {}",
                pid.unwrap(),
                cfg.port
            )));
            issues.push(format!(
                "Daemon process exists but /health on port {} doesn't respond. Try 'openlatch restart'.",
                cfg.port
            ));
        }
    } else {
        checks.push(DoctorCheck::fail(format!(
            "Daemon: not running (port {})",
            cfg.port
        )));
        issues.push("Daemon is not running. Run 'openlatch start' to start it.".to_string());
    }

    // Check 4b: Cloud reachability.
    // When the daemon is running we query its own /metrics endpoint for the
    // cloud_status it actually observes — this catches misconfigured URLs or
    // dropped events that the doctor process itself would never see (because it
    // reads its own env, which may differ from the env the daemon was started with).
    // When the daemon is down we fall back to a direct cloud ping from this process.
    let mut daemon_uptime_secs: Option<u64> = None;
    if daemon_alive {
        let metrics_url = format!("http://127.0.0.1:{}/metrics", cfg.port);
        match reqwest::blocking::get(&metrics_url).and_then(|r| r.json::<serde_json::Value>()) {
            Ok(m) => {
                daemon_uptime_secs = m.get("uptime_secs").and_then(|v| v.as_u64());
                let status = m
                    .get("cloud_status")
                    .and_then(|v| v.as_str())
                    .unwrap_or("unknown");
                let drops = m
                    .get("cloud_drop_count")
                    .and_then(|v| v.as_u64())
                    .unwrap_or(0);
                let api_url = m
                    .get("cloud_api_url")
                    .and_then(|v| v.as_str())
                    .unwrap_or("");
                match status {
                    "connected" => {
                        checks.push(DoctorCheck::pass(format!("Cloud: connected ({api_url})")));
                    }
                    "network_error" => {
                        let msg =
                            format!("Cloud: network errors — {drops} event(s) dropped ({api_url})");
                        checks.push(DoctorCheck::fail(msg.clone()));
                        issues.push(format!(
                            "Daemon cannot reach cloud at {api_url}. Check OPENLATCH_API_URL and network connectivity."
                        ));
                    }
                    "auth_error" => {
                        let msg = format!("Cloud: auth error — credential rejected ({api_url})");
                        checks.push(DoctorCheck::fail(msg));
                        issues.push(
                            "Run 'openlatch auth login' to refresh the cloud credential."
                                .to_string(),
                        );
                    }
                    "no_credential" => {
                        let msg = format!("Cloud: no credential — forwarding paused ({api_url})");
                        checks.push(DoctorCheck::fail(msg));
                        issues.push(
                            "Run 'openlatch auth login' to store an API key, or set \
                             [cloud] enabled = false in config.toml."
                                .to_string(),
                        );
                    }
                    "not_configured" => {
                        checks.push(DoctorCheck::pass(
                            "Cloud: disabled in daemon config".to_string(),
                        ));
                    }
                    other => {
                        checks.push(DoctorCheck::fail(format!(
                            "Cloud: unknown status '{other}'"
                        )));
                    }
                }
            }
            Err(_) => {
                checks.push(DoctorCheck::fail(
                    "Cloud: could not read daemon /metrics — daemon may be unhealthy".to_string(),
                ));
            }
        }
    } else if cfg.cloud.enabled {
        match tokio::runtime::Runtime::new() {
            Ok(rt) => {
                let (pass, message) = rt.block_on(check_cloud(&cfg.cloud.api_url));
                if pass {
                    checks.push(DoctorCheck::pass(message));
                } else {
                    checks.push(DoctorCheck::fail(message.clone()));
                    if message.contains("not authenticated") {
                        issues.push("Run 'openlatch auth login' to enable cloud sync.".to_string());
                    } else {
                        issues.push(format!(
                            "Cloud unreachable or credential rejected at {}. Check network connectivity or re-run 'openlatch auth login'.",
                            cfg.cloud.api_url
                        ));
                    }
                }
            }
            Err(_) => {
                checks.push(DoctorCheck::fail(
                    "Cloud: check skipped (runtime init failed)".to_string(),
                ));
            }
        }
    } else {
        checks.push(DoctorCheck::pass("Cloud: disabled in config".to_string()));
    }

    // Check 5: Hooks installed in settings.json
    if let Ok(agent) = hooks::detect_agent() {
        let settings_path = match &agent {
            hooks::DetectedAgent::ClaudeCode { settings_path, .. } => settings_path.clone(),
        };

        if settings_path.exists() {
            match std::fs::read_to_string(&settings_path) {
                Ok(content) => {
                    // Check for _openlatch marker entries
                    let has_pre_tool_use =
                        content.contains("PreToolUse") && content.contains("_openlatch");
                    let has_user_prompt =
                        content.contains("UserPromptSubmit") && content.contains("_openlatch");
                    let has_stop = content.contains("Stop") && content.contains("_openlatch");

                    let mut missing: Vec<&str> = Vec::new();
                    if !has_pre_tool_use {
                        missing.push("PreToolUse");
                    }
                    if !has_user_prompt {
                        missing.push("UserPromptSubmit");
                    }
                    if !has_stop {
                        missing.push("Stop");
                    }

                    if missing.is_empty() {
                        checks.push(DoctorCheck::pass(format!(
                            "Hooks: all entries present in {}",
                            settings_path.display()
                        )));
                    } else {
                        for hook in &missing {
                            checks.push(DoctorCheck::fail(format!(
                                "Hooks: {hook} missing from {}",
                                settings_path.display()
                            )));
                        }
                        issues.push(format!(
                            "Missing hooks: {}. Run 'openlatch init' to fix hook installation.",
                            missing.join(", ")
                        ));
                    }
                }
                Err(e) => {
                    checks.push(DoctorCheck::fail(format!(
                        "Hooks: cannot read {} — {e}",
                        settings_path.display()
                    )));
                    issues.push(format!(
                        "Cannot read settings file '{}'. Check permissions.",
                        settings_path.display()
                    ));
                }
            }

            // Check 6: Hook command sanity — binary path exists + matches
            // resolve_hook_binary_path(), and settings.json OPENLATCH_TOKEN
            // matches daemon.token on disk. Catches stale post-reinstall state
            // where the daemon loaded a fresh token but settings.json still
            // carries the old one, or a drifted binary path after `cargo install`.
            check_hook_binding(
                &settings_path,
                daemon_token.as_deref(),
                &mut checks,
                &mut issues,
            );
        } else {
            checks.push(DoctorCheck::fail(format!(
                "Hooks: settings.json not found at {}",
                settings_path.display()
            )));
            issues.push(
                "settings.json not found. Run 'openlatch init' to install hooks.".to_string(),
            );
        }
    }

    // Check 6b: Supervision state. Persistence is a security property — a
    // wrongly-advertised supervisor is worse than no supervisor, so we treat the
    // (Active, not-installed) drift as a fail condition that --fix can heal.
    check_supervision_state(&cfg, &mut checks, &mut issues);

    // Check 7: Fallback log activity while daemon has been alive. The hook
    // binary only writes to fallback.jsonl when the daemon POST fails — so
    // any entry newer than the daemon's start time proves the hook can't
    // reach its own daemon (token mismatch, wrong port, IPv6-only loopback).
    check_fallback_activity(
        &ol_dir,
        daemon_alive,
        daemon_uptime_secs,
        &mut checks,
        &mut issues,
    );

    Ok(DoctorReport {
        checks,
        issues,
        daemon_alive,
        daemon_uptime_secs,
        port: cfg.port,
    })
}

/// Render a [`DoctorReport`] using the configured output format.
///
/// JSON mode emits `{checks, issues, issue_count}` for parity with the
/// pre-refactor schema. Human mode prints checkmark/cross lines plus a
/// summary footer.
pub(crate) fn print_diagnostic_results(report: &DoctorReport, output: &OutputConfig) {
    if output.format == OutputFormat::Json {
        let checks_json: Vec<serde_json::Value> = report
            .checks
            .iter()
            .map(|c| {
                serde_json::json!({
                    "pass": c.pass,
                    "message": c.message,
                })
            })
            .collect();
        output.print_json(&serde_json::json!({
            "checks": checks_json,
            "issues": report.issues,
            "issue_count": report.issues.len(),
        }));
    } else if !output.quiet {
        for check in &report.checks {
            if check.pass {
                let mark = crate::cli::color::checkmark(output.color);
                eprintln!("{mark} {}", check.message);
            } else {
                let cross = crate::cli::color::cross(output.color);
                eprintln!("{cross} {}", check.message);
            }
        }

        eprintln!();
        if report.issues.is_empty() {
            eprintln!("All checks passed.");
        } else {
            let n = report.issues.len();
            eprintln!(
                "{n} issue{} found. Run 'openlatch init' to fix hook installation.",
                if n == 1 { "" } else { "s" }
            );
        }
    }
}

/// Probe the cloud by loading the stored API key and calling
/// `GET /api/v1/users/me` via the same validator used by `openlatch auth status`.
/// Returns (pass, human-readable message).
async fn check_cloud(api_url: &str) -> (bool, String) {
    use secrecy::ExposeSecret;

    let store = crate::core::auth::KeyringCredentialStore::new();
    let file_store = crate::cli::commands::auth::make_file_store();

    let key = match crate::core::auth::retrieve_credential(
        &store as &dyn crate::core::auth::CredentialStore,
        &file_store as &dyn crate::core::auth::CredentialStore,
    ) {
        Ok(k) => k,
        Err(_) => return (false, "Cloud: not authenticated".to_string()),
    };

    let key_str = key.expose_secret().to_string();
    let result = crate::cli::commands::auth::validate_online_full(&key_str, api_url).await;

    if result.online {
        let org = if result.org_name.is_empty() {
            String::new()
        } else {
            format!(" — org {}", result.org_name)
        };
        (true, format!("Cloud: reachable ({api_url}){org}"))
    } else {
        (
            false,
            format!("Cloud: unreachable or credential rejected ({api_url})"),
        )
    }
}

/// Verify the hook command in settings.json still points at a usable binary
/// and that the bearer token the hook subprocess will receive matches the
/// one the daemon has loaded.
fn check_hook_binding(
    settings_path: &std::path::Path,
    daemon_token: Option<&str>,
    checks: &mut Vec<DoctorCheck>,
    issues: &mut Vec<String>,
) {
    let raw = match std::fs::read_to_string(settings_path) {
        Ok(s) => s,
        // Read-error is already reported by Check 5; don't double-count.
        Err(_) => return,
    };
    let parsed = match crate::hooks::jsonc::parse_settings_value(&raw) {
        Ok(v) => v,
        Err(e) => {
            checks.push(DoctorCheck::fail(format!(
                "Hook config: cannot parse settings.json ({})",
                e.code
            )));
            return;
        }
    };

    let expected_bin = crate::hooks::resolve_hook_binary_path();
    let commands = openlatch_hook_commands(&parsed);
    if commands.is_empty() {
        // No _openlatch entries — Check 5 already flagged this path.
        return;
    }

    let mut bin_missing: Vec<String> = Vec::new();
    let mut bin_drifted: Vec<String> = Vec::new();
    for cmd in &commands {
        let Some(bin) = extract_quoted_binary(cmd) else {
            continue;
        };
        let bin_path = std::path::PathBuf::from(&bin);
        if !bin_path.exists() {
            bin_missing.push(bin);
        } else if bin_path != expected_bin {
            bin_drifted.push(bin);
        }
    }

    if bin_missing.is_empty() && bin_drifted.is_empty() {
        checks.push(DoctorCheck::pass(format!(
            "Hook binary: {} (referenced by settings.json)",
            expected_bin.display()
        )));
    }
    if !bin_missing.is_empty() {
        checks.push(DoctorCheck::fail(format!(
            "Hook binary missing: {}",
            bin_missing.join(", ")
        )));
        issues.push(
            "Hook command points at a binary that does not exist. Run 'openlatch init' to rewrite the hook command."
                .to_string(),
        );
    }
    if !bin_drifted.is_empty() {
        checks.push(DoctorCheck::fail(format!(
            "Hook binary drift: settings.json uses {}, current install is {}",
            bin_drifted.join(", "),
            expected_bin.display()
        )));
        issues.push(
            "Hook command points at a stale binary. Run 'openlatch init' to re-link settings.json to the current install."
                .to_string(),
        );
    }

    // Token cross-check — settings.json env.OPENLATCH_TOKEN vs daemon.token.
    if let Some(token) = daemon_token {
        let settings_token = parsed
            .get("env")
            .and_then(|e| e.get("OPENLATCH_TOKEN"))
            .and_then(|v| v.as_str());
        match settings_token {
            Some(t) if t == token => {
                checks.push(DoctorCheck::pass(
                    "Hook token: settings.json env matches daemon.token".to_string(),
                ));
            }
            Some(_) => {
                checks.push(DoctorCheck::fail(
                    "Hook token: settings.json OPENLATCH_TOKEN does not match daemon.token"
                        .to_string(),
                ));
                issues.push(
                    "Hook subprocess will be rejected with 401. Run 'openlatch init' to re-sync the token."
                        .to_string(),
                );
            }
            None => {
                checks.push(DoctorCheck::fail(
                    "Hook token: OPENLATCH_TOKEN not set in settings.json env".to_string(),
                ));
                issues.push(
                    "Hook subprocess will not receive OPENLATCH_TOKEN. Run 'openlatch init' to install it."
                        .to_string(),
                );
            }
        }
    }
}

fn openlatch_hook_commands(settings: &serde_json::Value) -> Vec<String> {
    let Some(hooks) = settings.get("hooks").and_then(|v| v.as_object()) else {
        return Vec::new();
    };
    let mut out: Vec<String> = Vec::new();
    for entries in hooks.values().filter_map(|v| v.as_array()) {
        for entry in entries {
            if entry.get("_openlatch").and_then(|v| v.as_bool()) != Some(true) {
                continue;
            }
            let Some(inner) = entry.get("hooks").and_then(|v| v.as_array()) else {
                continue;
            };
            for h in inner {
                if let Some(cmd) = h.get("command").and_then(|v| v.as_str()) {
                    out.push(cmd.to_string());
                }
            }
        }
    }
    out
}

/// The hook command we write always quotes the binary path as its first
/// token (so spaces in Windows paths survive); everything between the first
/// pair of double quotes is the path.
fn extract_quoted_binary(command: &str) -> Option<String> {
    let start = command.find('"')? + 1;
    let end = command[start..].find('"')? + start;
    if start == end {
        return None;
    }
    Some(command[start..end].to_string())
}

/// Probe the OS-native supervisor and compare its actual state against
/// `config.supervision`. The comparison catches drift in both directions:
/// the user manually deleted the plist/task/unit, OR the config was never
/// updated but the supervisor is still alive.
fn check_supervision_state(
    cfg: &config::Config,
    checks: &mut Vec<DoctorCheck>,
    issues: &mut Vec<String>,
) {
    use crate::supervision::{select_supervisor, SupervisionMode, SupervisorKind};
    let configured_backend = match cfg.supervision.backend {
        SupervisorKind::Launchd => "launchd",
        SupervisorKind::Systemd => "systemd",
        SupervisorKind::TaskScheduler => "task_scheduler",
        SupervisorKind::None => "none",
    };

    match (&cfg.supervision.mode, select_supervisor()) {
        (SupervisionMode::Active, Some(sup)) => match sup.status() {
            Ok(s) if s.installed && s.running => {
                checks.push(DoctorCheck::pass(format!(
                    "Supervision: active ({configured_backend}, {})",
                    s.description
                )));
            }
            Ok(s) if s.installed => {
                checks.push(DoctorCheck::fail(format!(
                    "Supervision: installed but supervisor not running ({configured_backend})"
                )));
                issues.push(
                    "Supervisor is registered but not running. Check OS logs or run \
                     'openlatch supervision install' to rebuild the artifact."
                        .to_string(),
                );
            }
            Ok(_) => {
                checks.push(DoctorCheck::fail(format!(
                    "Supervision: config says active but OS artifact is missing ({configured_backend})"
                )));
                issues.push(
                    "Supervisor drifted — OS artifact deleted while config still expects it. \
                     Run 'openlatch supervision install' (or 'openlatch doctor --fix')."
                        .to_string(),
                );
            }
            Err(e) => {
                checks.push(DoctorCheck::fail(format!(
                    "Supervision: cannot query supervisor ({}) — {}",
                    e.code, e.message
                )));
            }
        },
        (SupervisionMode::Deferred, _) => {
            let reason = cfg
                .supervision
                .disabled_reason
                .as_deref()
                .unwrap_or("unknown");
            checks.push(DoctorCheck::fail(format!(
                "Supervision: deferred — {reason}"
            )));
            issues.push(
                "Supervision is deferred. Run 'openlatch supervision install' once the \
                 underlying issue is resolved (headless session, missing systemd, etc)."
                    .to_string(),
            );
        }
        (SupervisionMode::Disabled, _) => {
            let reason = cfg
                .supervision
                .disabled_reason
                .as_deref()
                .unwrap_or("user_opt_out");
            checks.push(DoctorCheck::pass(format!(
                "Supervision: disabled ({reason})"
            )));
        }
        (SupervisionMode::Active, None) => {
            checks.push(DoctorCheck::fail(
                "Supervision: config says active but no supervisor is available on this OS"
                    .to_string(),
            ));
            issues.push(
                "Supervisor unsupported on this OS. Run 'openlatch supervision disable' to \
                 clear the stale state."
                    .to_string(),
            );
        }
    }
}

/// Flag fallback.jsonl as a problem when any of its bytes were written after
/// the currently-running daemon started. That's the on-disk trace of a hook
/// subprocess failing to reach its own daemon.
fn check_fallback_activity(
    ol_dir: &std::path::Path,
    daemon_alive: bool,
    daemon_uptime_secs: Option<u64>,
    checks: &mut Vec<DoctorCheck>,
    issues: &mut Vec<String>,
) {
    if !daemon_alive {
        // Without a live daemon there's no "since daemon start" reference
        // point, and any fallback activity is expected (daemon is down).
        return;
    }
    let Some(uptime) = daemon_uptime_secs else {
        return;
    };
    let fallback_path = ol_dir.join("logs").join("fallback.jsonl");
    if !fallback_path.exists() {
        checks.push(DoctorCheck::pass(
            "Hook fallback log: no offline events recorded".to_string(),
        ));
        return;
    }

    let meta = match std::fs::metadata(&fallback_path) {
        Ok(m) => m,
        Err(e) => {
            checks.push(DoctorCheck::fail(format!(
                "Hook fallback log: cannot stat {} — {e}",
                fallback_path.display()
            )));
            return;
        }
    };
    let mtime = match meta.modified() {
        Ok(t) => t,
        Err(_) => return,
    };
    let daemon_start = std::time::SystemTime::now()
        .checked_sub(std::time::Duration::from_secs(uptime))
        .unwrap_or(std::time::UNIX_EPOCH);

    if mtime > daemon_start {
        let age_secs = mtime
            .duration_since(daemon_start)
            .map(|d| d.as_secs())
            .unwrap_or(0);
        checks.push(DoctorCheck::fail(format!(
            "Hook fallback log: {} written {age_secs}s after daemon start — hook subprocess is not reaching the daemon",
            fallback_path.display()
        )));
        issues.push(
            "Hook binary is writing to fallback.jsonl while the daemon is up. \
             Likely causes: token mismatch, wrong port, or stale settings.json. \
             Run 'openlatch doctor' above for Hook token/binary checks, then 'openlatch init' to repair."
                .to_string(),
        );
    } else {
        checks.push(DoctorCheck::pass(format!(
            "Hook fallback log: quiet since daemon start ({})",
            fallback_path.display()
        )));
    }
}