ClawdDotNet/src/ClawdDotNet.Tools.WebFetch/WebFetchTool.cs

using System.Net;
using System.Text.Json;
using System.Text.RegularExpressions;
using System.Xml.Linq;
using ClawdDotNet.Core.Tools;
using Microsoft.Extensions.Logging;

namespace ClawdDotNet.Tools.WebFetch;

public sealed class WebFetchTool : IAgentTool
{
    public string Name => "WebFetch";

    public string Description => """
        Ruft statische Webseiten oder RSS-Feeds ab und extrahiert Text + Timestamps.
        Nur Domains aus der Whitelist erlaubt. Kein JavaScript-Rendering.
        Aktionen: fetch, rss
        """;

    public JsonElement InputSchema { get; } = JsonDocument.Parse("""
    {
      "type": "object",
      "required": ["action", "url"],
      "properties": {
        "action": {
          "type": "string",
          "enum": ["fetch", "rss"],
          "description": "fetch=HTML-Seite abrufen und zu Text konvertieren, rss=RSS/Atom-Feed parsen"
        },
        "url":     { "type": "string" },
        "selector":{ "type": "string",
                     "description": "optional: CSS-ähnlicher Hint welcher Teil relevant ist, z.B. 'table', 'article'" }
      }
    }
    """).RootElement.Clone();

    public async Task<ToolResult> ExecuteAsync(
        JsonElement input,
        AgentToolContext context,
        CancellationToken ct)
    {
        try
        {
            var url = input.GetProperty("url").GetString()!;
            var action = input.GetProperty("action").GetString()!;

            IReadOnlyDictionary<string, object?>? config = context.ToolConfig.TryGetValue("WebFetch", out var c) && c is JsonElement je 
                ? JsonSerializer.Deserialize<Dictionary<string, object?>>(je.GetRawText()) 
                : context.ToolConfig;

            if (config == null) return ToolResult.Fail("WebFetch Konfiguration fehlt.");

            var allowedDomains = config.GetValueOrDefault("allowedDomains") is JsonElement ad 
                ? ad.EnumerateArray().Select(x => x.GetString()!).ToList() 
                : (config.GetValueOrDefault("allowedDomains") as IEnumerable<string>)?.ToList() ?? new List<string>();

            // Adresse prüfen: Schema, privates Netz, Whitelist
            if (!UrlGuard.TryValidate(url, allowedDomains, out _, out var urlError))
                return ToolResult.Fail(urlError!);

            return action switch
            {
                "fetch" => await FetchPageAsync(url, allowedDomains, config, ct),
                "rss" => await FetchRssAsync(url, allowedDomains, ct),
                _ => ToolResult.Fail($"Unbekannte Aktion: {action}")
            };
        }
        catch (Exception ex)
        {
            context.Logger.LogError(ex, "Fehler in WebFetch");
            return ToolResult.Fail($"Fehler: {ex.Message}");
        }
    }

    private async Task<ToolResult> FetchPageAsync(
        string url, IReadOnlyCollection<string> allowedDomains,
        IReadOnlyDictionary<string, object?> config, CancellationToken ct)
    {
        using var http = CreateHttpClient(config);
        var fetchedAt = DateTime.UtcNow;

        var (response, redirectError) = await GetFollowingRedirectsAsync(http, url, allowedDomains, ct);
        if (redirectError is not null)
            return redirectError;

        using var _ = response!;

        if (!response.IsSuccessStatusCode)
            return ToolResult.Fail($"Seite nicht erreichbar: {url} → HTTP {(int)response.StatusCode} {response.ReasonPhrase}");

        DateTime? dataAsOf = response.Content.Headers.LastModified?.UtcDateTime 
                             ?? response.Headers.Date?.UtcDateTime;

        var html = await response.Content.ReadAsStringAsync(ct);
        var maxKb = GetConfigInt(config, "maxResponseKb", 512);
        if (html.Length > maxKb * 1024) html = html[..(maxKb * 1024)];

        var text = StripHtml(html);
        if (dataAsOf == null) dataAsOf = ExtractDateFromHtml(html);

        var result = new
        {
            fetchedAt = fetchedAt,
            dataAsOf = dataAsOf,
            source = url,
            data = new { text }
        };

        return ToolResult.Ok(JsonSerializer.Serialize(result, new JsonSerializerOptions { WriteIndented = true }));
    }

    private async Task<ToolResult> FetchRssAsync(
        string url, IReadOnlyCollection<string> allowedDomains, CancellationToken ct)
    {
        using var http = CreateHttpClient(new Dictionary<string, object?>());
        var fetchedAt = DateTime.UtcNow;

        var (response, redirectError) = await GetFollowingRedirectsAsync(http, url, allowedDomains, ct);
        if (redirectError is not null)
            return redirectError;

        using var _ = response!;

        if (!response.IsSuccessStatusCode)
            return ToolResult.Fail($"RSS-Feed nicht erreichbar: {url} → HTTP {(int)response.StatusCode} {response.ReasonPhrase}");

        var xml = await response.Content.ReadAsStringAsync(ct);
        var doc = XDocument.Parse(xml);

        var items = new List<object>();
        // Support RSS and Atom
        var ns = doc.Root?.Name.Namespace;

        if (doc.Root?.Name.LocalName == "rss")
        {
            foreach (var item in doc.Descendants("item").Take(20))
            {
                items.Add(new
                {
                    title = item.Element("title")?.Value,
                    link = item.Element("link")?.Value,
                    pubDate = item.Element("pubDate")?.Value,
                    description = StripHtml(item.Element("description")?.Value ?? "")
                });
            }
        }
        else // Atom
        {
            foreach (var entry in doc.Descendants((ns ?? XNamespace.None) + "entry").Take(20))
            {
                items.Add(new
                {
                    title = entry.Element(ns + "title")?.Value,
                    link = entry.Element(ns + "link")?.Attribute("href")?.Value,
                    pubDate = entry.Element(ns + "updated")?.Value ?? entry.Element(ns + "published")?.Value,
                    description = StripHtml(entry.Element(ns + "summary")?.Value ?? entry.Element(ns + "content")?.Value ?? "")
                });
            }
        }

        var result = new
        {
            fetchedAt = fetchedAt,
            dataAsOf = fetchedAt, // RSS current state
            source = url,
            data = new { items }
        };

        return ToolResult.Ok(JsonSerializer.Serialize(result, new JsonSerializerOptions { WriteIndented = true }));
    }

    private static string StripHtml(string html)
    {
        if (string.IsNullOrWhiteSpace(html)) return "";

        // Script und Style entfernen
        html = Regex.Replace(html, "<script.*?>.*?</script>", "", RegexOptions.Singleline | RegexOptions.IgnoreCase);
        html = Regex.Replace(html, "<style.*?>.*?</style>", "", RegexOptions.Singleline | RegexOptions.IgnoreCase);
        
        // Alle anderen Tags entfernen
        html = Regex.Replace(html, "<.*?>", " ", RegexOptions.Singleline);
        
        // Entities dekodieren
        html = WebUtility.HtmlDecode(html);
        
        // Whitespace säubern
        html = Regex.Replace(html, @"\s+", " ");
        html = html.Replace(" \n ", "\n").Replace(" \r ", "\r").Trim();
        
        return html;
    }

    private static DateTime? ExtractDateFromHtml(string html)
    {
        // Suche nach meta-Tags
        var patterns = new[]
        {
            "<meta.*?property=\"article:published_time\".*?content=\"(.*?)\"",
            "<meta.*?name=\"date\".*?content=\"(.*?)\"",
            "<time.*?datetime=\"(.*?)\""
        };

        foreach (var pattern in patterns)
        {
            var match = Regex.Match(html, pattern, RegexOptions.IgnoreCase);
            if (match.Success && DateTime.TryParse(match.Groups[1].Value, out var dt))
                return dt.ToUniversalTime();
        }

        return null;
    }

    private const int MaxRedirects = 5;

    /// <summary>
    /// Folgt Weiterleitungen selbst und prüft jedes Ziel erneut gegen Whitelist und
    /// Netzsperren. HttpClient würde sonst ungefragt bis ins lokale Netz folgen.
    /// </summary>
    private static async Task<(HttpResponseMessage? Response, ToolResult? Error)> GetFollowingRedirectsAsync(
        HttpClient http, string url, IReadOnlyCollection<string> allowedDomains, CancellationToken ct)
    {
        var current = url;

        for (var hop = 0; hop <= MaxRedirects; hop++)
        {
            if (!UrlGuard.TryValidate(current, allowedDomains, out var uri, out var error))
            {
                var context = hop == 0 ? "" : $" (Weiterleitung {hop} von {url})";
                return (null, ToolResult.Fail($"{error}{context}"));
            }

            var response = await http.GetAsync(uri, ct);

            if (!IsRedirect(response.StatusCode))
                return (response, null);

            var location = response.Headers.Location;
            response.Dispose();

            if (location is null)
                return (null, ToolResult.Fail($"Weiterleitung ohne Zieladresse bei {current}."));

            // Relative Weiterleitungen gegen die aktuelle Adresse auflösen.
            current = location.IsAbsoluteUri
                ? location.ToString()
                : new Uri(uri!, location).ToString();
        }

        return (null, ToolResult.Fail(
            $"Mehr als {MaxRedirects} Weiterleitungen ausgehend von {url} — abgebrochen."));
    }

    private static bool IsRedirect(HttpStatusCode status) => status is
        HttpStatusCode.MovedPermanently or
        HttpStatusCode.Found or
        HttpStatusCode.SeeOther or
        HttpStatusCode.TemporaryRedirect or
        HttpStatusCode.PermanentRedirect;

    private static HttpClient CreateHttpClient(IReadOnlyDictionary<string, object?> config)
    {
        // Weiterleitungen bewusst NICHT automatisch verfolgen — sie werden einzeln
        // aufgelöst und geprüft.
        var handler = new HttpClientHandler { AllowAutoRedirect = false };
        var client = new HttpClient(handler, disposeHandler: true);
        var timeout = GetConfigInt(config, "timeoutSeconds", 15);
        client.Timeout = TimeSpan.FromSeconds(timeout);
        client.DefaultRequestHeaders.UserAgent.ParseAdd(
            config.GetValueOrDefault("userAgent")?.ToString() ?? "ClawdDotNet-Agent/1.0");
        client.DefaultRequestHeaders.CacheControl = new System.Net.Http.Headers.CacheControlHeaderValue { NoCache = true, NoStore = true };
        return client;
    }

    /// <summary>
    /// Sicheres Lesen eines int-Werts aus der Config (funktioniert mit JsonElement und primitiven Typen).
    /// </summary>
    private static int GetConfigInt(IReadOnlyDictionary<string, object?> config, string key, int defaultValue)
    {
        var val = config.GetValueOrDefault(key);
        return val switch
        {
            null => defaultValue,
            int i => i,
            long l => (int)l,
            System.Text.Json.JsonElement je when je.ValueKind == System.Text.Json.JsonValueKind.Number => je.GetInt32(),
            _ => int.TryParse(val.ToString(), out var parsed) ? parsed : defaultValue
        };
    }
}