1package tools
2
3import (
4 "context"
5 _ "embed"
6 "fmt"
7 "io"
8 "net/http"
9 "strings"
10 "time"
11 "unicode/utf8"
12
13 "charm.land/fantasy"
14 md "github.com/JohannesKaufmann/html-to-markdown"
15 "github.com/PuerkitoBio/goquery"
16 "github.com/charmbracelet/crush/internal/permission"
17)
18
19const FetchToolName = "fetch"
20
21//go:embed fetch.md
22var fetchDescription []byte
23
24func NewFetchTool(permissions permission.Service, workingDir string, client *http.Client) fantasy.AgentTool {
25 if client == nil {
26 client = &http.Client{
27 Timeout: 30 * time.Second,
28 Transport: &http.Transport{
29 MaxIdleConns: 100,
30 MaxIdleConnsPerHost: 10,
31 IdleConnTimeout: 90 * time.Second,
32 },
33 }
34 }
35
36 return fantasy.NewParallelAgentTool(
37 FetchToolName,
38 string(fetchDescription),
39 func(ctx context.Context, params FetchParams, call fantasy.ToolCall) (fantasy.ToolResponse, error) {
40 if params.URL == "" {
41 return fantasy.NewTextErrorResponse("URL parameter is required"), nil
42 }
43
44 format := strings.ToLower(params.Format)
45 if format != "text" && format != "markdown" && format != "html" {
46 return fantasy.NewTextErrorResponse("Format must be one of: text, markdown, html"), nil
47 }
48
49 if !strings.HasPrefix(params.URL, "http://") && !strings.HasPrefix(params.URL, "https://") {
50 return fantasy.NewTextErrorResponse("URL must start with http:// or https://"), nil
51 }
52
53 sessionID := GetSessionFromContext(ctx)
54 if sessionID == "" {
55 return fantasy.ToolResponse{}, fmt.Errorf("session ID is required for creating a new file")
56 }
57
58 p, err := permissions.Request(ctx,
59 permission.CreatePermissionRequest{
60 SessionID: sessionID,
61 Path: workingDir,
62 ToolCallID: call.ID,
63 ToolName: FetchToolName,
64 Action: "fetch",
65 Description: fmt.Sprintf("Fetch content from URL: %s", params.URL),
66 Params: FetchPermissionsParams(params),
67 },
68 )
69 if err != nil {
70 return fantasy.ToolResponse{}, err
71 }
72 if !p {
73 return fantasy.ToolResponse{}, permission.ErrorPermissionDenied
74 }
75
76 // Handle timeout with context
77 requestCtx := ctx
78 if params.Timeout > 0 {
79 maxTimeout := 120 // 2 minutes
80 if params.Timeout > maxTimeout {
81 params.Timeout = maxTimeout
82 }
83 var cancel context.CancelFunc
84 requestCtx, cancel = context.WithTimeout(ctx, time.Duration(params.Timeout)*time.Second)
85 defer cancel()
86 }
87
88 req, err := http.NewRequestWithContext(requestCtx, "GET", params.URL, nil)
89 if err != nil {
90 return fantasy.ToolResponse{}, fmt.Errorf("failed to create request: %w", err)
91 }
92
93 req.Header.Set("User-Agent", "crush/1.0")
94
95 resp, err := client.Do(req)
96 if err != nil {
97 return fantasy.ToolResponse{}, fmt.Errorf("failed to fetch URL: %w", err)
98 }
99 defer resp.Body.Close()
100
101 if resp.StatusCode != http.StatusOK {
102 return fantasy.NewTextErrorResponse(fmt.Sprintf("Request failed with status code: %d", resp.StatusCode)), nil
103 }
104
105 maxSize := int64(5 * 1024 * 1024) // 5MB
106 body, err := io.ReadAll(io.LimitReader(resp.Body, maxSize))
107 if err != nil {
108 return fantasy.NewTextErrorResponse("Failed to read response body: " + err.Error()), nil
109 }
110
111 content := string(body)
112
113 isValidUt8 := utf8.ValidString(content)
114 if !isValidUt8 {
115 return fantasy.NewTextErrorResponse("Response content is not valid UTF-8"), nil
116 }
117 contentType := resp.Header.Get("Content-Type")
118
119 switch format {
120 case "text":
121 if strings.Contains(contentType, "text/html") {
122 text, err := extractTextFromHTML(content)
123 if err != nil {
124 return fantasy.NewTextErrorResponse("Failed to extract text from HTML: " + err.Error()), nil
125 }
126 content = text
127 }
128
129 case "markdown":
130 if strings.Contains(contentType, "text/html") {
131 markdown, err := convertHTMLToMarkdown(content)
132 if err != nil {
133 return fantasy.NewTextErrorResponse("Failed to convert HTML to Markdown: " + err.Error()), nil
134 }
135 content = markdown
136 }
137
138 content = "```\n" + content + "\n```"
139
140 case "html":
141 // return only the body of the HTML document
142 if strings.Contains(contentType, "text/html") {
143 doc, err := goquery.NewDocumentFromReader(strings.NewReader(content))
144 if err != nil {
145 return fantasy.NewTextErrorResponse("Failed to parse HTML: " + err.Error()), nil
146 }
147 body, err := doc.Find("body").Html()
148 if err != nil {
149 return fantasy.NewTextErrorResponse("Failed to extract body from HTML: " + err.Error()), nil
150 }
151 if body == "" {
152 return fantasy.NewTextErrorResponse("No body content found in HTML"), nil
153 }
154 content = "<html>\n<body>\n" + body + "\n</body>\n</html>"
155 }
156 }
157 // calculate byte size of content
158 contentSize := int64(len(content))
159 if contentSize > MaxReadSize {
160 content = content[:MaxReadSize]
161 content += fmt.Sprintf("\n\n[Content truncated to %d bytes]", MaxReadSize)
162 }
163
164 return fantasy.NewTextResponse(content), nil
165 })
166}
167
168func extractTextFromHTML(html string) (string, error) {
169 doc, err := goquery.NewDocumentFromReader(strings.NewReader(html))
170 if err != nil {
171 return "", err
172 }
173
174 text := doc.Find("body").Text()
175 text = strings.Join(strings.Fields(text), " ")
176
177 return text, nil
178}
179
180func convertHTMLToMarkdown(html string) (string, error) {
181 converter := md.NewConverter("", true, nil)
182
183 markdown, err := converter.ConvertString(html)
184 if err != nil {
185 return "", err
186 }
187
188 return markdown, nil
189}