266 lines
8.3 KiB
Go
266 lines
8.3 KiB
Go
|
package handler
|
||
|
|
|
||
|
|
import (
|
||
|
|
"crypto/sha256"
|
||
|
|
"encoding/json"
|
||
|
|
"fmt"
|
||
|
|
"net/http"
|
||
|
|
"strings"
|
||
|
|
"time"
|
||
|
|
|
||
|
|
"github.com/seifghazi/claude-code-monitor/internal/model"
|
||
|
|
)
|
||
|
|
|
||
|
|
// SanitizeHeaders removes sensitive headers before logging/storage
|
||
|
|
func SanitizeHeaders(headers http.Header) http.Header {
|
||
|
|
sanitized := make(http.Header)
|
||
|
|
|
||
|
|
sensitiveHeaders := []string{
|
||
|
|
"x-api-key",
|
||
|
|
"api-key",
|
||
|
|
"authorization",
|
||
|
|
"anthropic-api-key",
|
||
|
|
"openai-api-key",
|
||
|
|
"bearer",
|
||
|
|
}
|
||
|
|
|
||
|
|
for key, values := range headers {
|
||
|
|
lowerKey := strings.ToLower(key)
|
||
|
|
isSensitive := false
|
||
|
|
|
||
|
|
for _, sensitive := range sensitiveHeaders {
|
||
|
|
if strings.Contains(lowerKey, sensitive) {
|
||
|
|
isSensitive = true
|
||
|
|
break
|
||
|
|
}
|
||
|
|
}
|
||
|
|
|
||
|
|
if isSensitive {
|
||
|
|
sanitized[key] = []string{"[REDACTED]"}
|
||
|
|
} else {
|
||
|
|
sanitized[key] = values
|
||
|
|
}
|
||
|
|
}
|
||
|
|
|
||
|
|
return sanitized
|
||
|
|
}
|
||
|
|
|
||
|
|
// ConversationDiffAnalyzer analyzes conversation flows to identify new vs repeated content
|
||
|
|
type ConversationDiffAnalyzer struct{}
|
||
|
|
|
||
|
|
// NewConversationDiffAnalyzer creates a new conversation diff analyzer
|
||
|
|
func NewConversationDiffAnalyzer() *ConversationDiffAnalyzer {
|
||
|
|
return &ConversationDiffAnalyzer{}
|
||
|
|
}
|
||
|
|
|
||
|
|
// ConversationFlowData represents the flow analysis of a conversation
|
||
|
|
type ConversationFlowData struct {
|
||
|
|
TotalMessages int `json:"totalMessages"`
|
||
|
|
NewMessages []int `json:"newMessages"` // Indices of new messages
|
||
|
|
DuplicateMessages []int `json:"duplicateMessages"` // Indices of duplicate messages
|
||
|
|
MessageHashes []string `json:"messageHashes"` // Content hashes for deduplication
|
||
|
|
ConversationHash string `json:"conversationHash"` // Hash of entire conversation
|
||
|
|
PreviousHash string `json:"previousHash"` // Hash of previous conversation state
|
||
|
|
Changes []ConversationChange `json:"changes"` // Detailed changes
|
||
|
|
FlowMetadata map[string]interface{} `json:"flowMetadata"` // Additional metadata
|
||
|
|
}
|
||
|
|
|
||
|
|
// ConversationChange represents a specific change in the conversation
|
||
|
|
type ConversationChange struct {
|
||
|
|
Type string `json:"type"` // "added", "modified", "context"
|
||
|
|
MessageIdx int `json:"messageIdx"` // Index of the message
|
||
|
|
Role string `json:"role"` // Role of the message
|
||
|
|
ContentHash string `json:"contentHash"` // Hash of the content
|
||
|
|
Preview string `json:"preview"` // Short preview of content
|
||
|
|
Timestamp string `json:"timestamp"` // When this change was detected
|
||
|
|
}
|
||
|
|
|
||
|
|
// AnalyzeConversationFlow analyzes a conversation to identify what's new vs repeated
|
||
|
|
func (c *ConversationDiffAnalyzer) AnalyzeConversationFlow(messages []model.AnthropicMessage, previousConversation []model.AnthropicMessage) *ConversationFlowData {
|
||
|
|
totalMessages := len(messages)
|
||
|
|
|
||
|
|
// Create hashes for current conversation
|
||
|
|
currentHashes := make([]string, totalMessages)
|
||
|
|
for i, msg := range messages {
|
||
|
|
currentHashes[i] = c.hashMessage(msg)
|
||
|
|
}
|
||
|
|
|
||
|
|
// Create hashes for previous conversation (if any)
|
||
|
|
var previousHashes []string
|
||
|
|
if previousConversation != nil {
|
||
|
|
previousHashes = make([]string, len(previousConversation))
|
||
|
|
for i, msg := range previousConversation {
|
||
|
|
previousHashes[i] = c.hashMessage(msg)
|
||
|
|
}
|
||
|
|
}
|
||
|
|
|
||
|
|
// Identify new vs duplicate messages
|
||
|
|
newMessages := []int{}
|
||
|
|
duplicateMessages := []int{}
|
||
|
|
changes := []ConversationChange{}
|
||
|
|
|
||
|
|
// Simple approach: messages that appear after the previous conversation length are new
|
||
|
|
previousLength := len(previousHashes)
|
||
|
|
|
||
|
|
for i, msg := range messages {
|
||
|
|
isNew := i >= previousLength
|
||
|
|
|
||
|
|
// More sophisticated check: compare hashes
|
||
|
|
if !isNew && i < len(previousHashes) {
|
||
|
|
isNew = currentHashes[i] != previousHashes[i]
|
||
|
|
}
|
||
|
|
|
||
|
|
if isNew {
|
||
|
|
newMessages = append(newMessages, i)
|
||
|
|
changes = append(changes, ConversationChange{
|
||
|
|
Type: "added",
|
||
|
|
MessageIdx: i,
|
||
|
|
Role: msg.Role,
|
||
|
|
ContentHash: currentHashes[i],
|
||
|
|
Preview: c.getMessagePreview(msg),
|
||
|
|
Timestamp: fmt.Sprintf("%d", time.Now().Unix()),
|
||
|
|
})
|
||
|
|
} else {
|
||
|
|
duplicateMessages = append(duplicateMessages, i)
|
||
|
|
changes = append(changes, ConversationChange{
|
||
|
|
Type: "context",
|
||
|
|
MessageIdx: i,
|
||
|
|
Role: msg.Role,
|
||
|
|
ContentHash: currentHashes[i],
|
||
|
|
Preview: c.getMessagePreview(msg),
|
||
|
|
Timestamp: fmt.Sprintf("%d", time.Now().Unix()),
|
||
|
|
})
|
||
|
|
}
|
||
|
|
}
|
||
|
|
|
||
|
|
// If no previous conversation, consider a reasonable threshold of "new" vs "context"
|
||
|
|
if previousConversation == nil && totalMessages > 1 {
|
||
|
|
// Heuristic: last 30% of messages are "new", rest is context
|
||
|
|
newThreshold := max(1, int(float64(totalMessages)*0.3))
|
||
|
|
contextEnd := totalMessages - newThreshold
|
||
|
|
|
||
|
|
newMessages = []int{}
|
||
|
|
duplicateMessages = []int{}
|
||
|
|
changes = []ConversationChange{}
|
||
|
|
|
||
|
|
for i := 0; i < totalMessages; i++ {
|
||
|
|
if i >= contextEnd {
|
||
|
|
newMessages = append(newMessages, i)
|
||
|
|
changes = append(changes, ConversationChange{
|
||
|
|
Type: "added",
|
||
|
|
MessageIdx: i,
|
||
|
|
Role: messages[i].Role,
|
||
|
|
ContentHash: currentHashes[i],
|
||
|
|
Preview: c.getMessagePreview(messages[i]),
|
||
|
|
Timestamp: fmt.Sprintf("%d", time.Now().Unix()),
|
||
|
|
})
|
||
|
|
} else {
|
||
|
|
duplicateMessages = append(duplicateMessages, i)
|
||
|
|
changes = append(changes, ConversationChange{
|
||
|
|
Type: "context",
|
||
|
|
MessageIdx: i,
|
||
|
|
Role: messages[i].Role,
|
||
|
|
ContentHash: currentHashes[i],
|
||
|
|
Preview: c.getMessagePreview(messages[i]),
|
||
|
|
Timestamp: fmt.Sprintf("%d", time.Now().Unix()),
|
||
|
|
})
|
||
|
|
}
|
||
|
|
}
|
||
|
|
}
|
||
|
|
|
||
|
|
// Generate conversation hashes
|
||
|
|
conversationHash := c.hashConversation(messages)
|
||
|
|
previousHash := ""
|
||
|
|
if previousConversation != nil {
|
||
|
|
previousHash = c.hashConversation(previousConversation)
|
||
|
|
}
|
||
|
|
|
||
|
|
return &ConversationFlowData{
|
||
|
|
TotalMessages: totalMessages,
|
||
|
|
NewMessages: newMessages,
|
||
|
|
DuplicateMessages: duplicateMessages,
|
||
|
|
MessageHashes: currentHashes,
|
||
|
|
ConversationHash: conversationHash,
|
||
|
|
PreviousHash: previousHash,
|
||
|
|
Changes: changes,
|
||
|
|
FlowMetadata: map[string]interface{}{
|
||
|
|
"newCount": len(newMessages),
|
||
|
|
"duplicateCount": len(duplicateMessages),
|
||
|
|
"analyzeTime": time.Now().Format(time.RFC3339),
|
||
|
|
},
|
||
|
|
}
|
||
|
|
}
|
||
|
|
|
||
|
|
// hashMessage creates a hash of a message for deduplication
|
||
|
|
func (c *ConversationDiffAnalyzer) hashMessage(msg model.AnthropicMessage) string {
|
||
|
|
// Create a stable representation of the message
|
||
|
|
content := c.normalizeMessageContent(msg.Content)
|
||
|
|
data := fmt.Sprintf("%s|%s", msg.Role, content)
|
||
|
|
|
||
|
|
hash := sha256.Sum256([]byte(data))
|
||
|
|
return fmt.Sprintf("%x", hash[:8]) // Use first 8 bytes for shorter hash
|
||
|
|
}
|
||
|
|
|
||
|
|
// hashConversation creates a hash of the entire conversation
|
||
|
|
func (c *ConversationDiffAnalyzer) hashConversation(messages []model.AnthropicMessage) string {
|
||
|
|
var parts []string
|
||
|
|
for _, msg := range messages {
|
||
|
|
parts = append(parts, c.hashMessage(msg))
|
||
|
|
}
|
||
|
|
|
||
|
|
conversationData := strings.Join(parts, "|")
|
||
|
|
hash := sha256.Sum256([]byte(conversationData))
|
||
|
|
return fmt.Sprintf("%x", hash[:16]) // Use first 16 bytes for conversation hash
|
||
|
|
}
|
||
|
|
|
||
|
|
// normalizeMessageContent converts message content to a normalized string
|
||
|
|
func (c *ConversationDiffAnalyzer) normalizeMessageContent(content interface{}) string {
|
||
|
|
switch v := content.(type) {
|
||
|
|
case string:
|
||
|
|
return strings.TrimSpace(v)
|
||
|
|
case []interface{}:
|
||
|
|
var parts []string
|
||
|
|
for _, item := range v {
|
||
|
|
if block, ok := item.(map[string]interface{}); ok {
|
||
|
|
if text, hasText := block["text"].(string); hasText {
|
||
|
|
parts = append(parts, strings.TrimSpace(text))
|
||
|
|
} else if blockType, hasType := block["type"].(string); hasType {
|
||
|
|
// Handle different content types (tool_use, etc.)
|
||
|
|
switch blockType {
|
||
|
|
case "tool_use":
|
||
|
|
if name, hasName := block["name"].(string); hasName {
|
||
|
|
parts = append(parts, fmt.Sprintf("TOOL:%s", name))
|
||
|
|
}
|
||
|
|
case "tool_result":
|
||
|
|
parts = append(parts, "TOOL_RESULT")
|
||
|
|
default:
|
||
|
|
parts = append(parts, fmt.Sprintf("CONTENT:%s", blockType))
|
||
|
|
}
|
||
|
|
}
|
||
|
|
}
|
||
|
|
}
|
||
|
|
return strings.Join(parts, " ")
|
||
|
|
default:
|
||
|
|
// Convert to JSON and back for normalization
|
||
|
|
jsonBytes, _ := json.Marshal(content)
|
||
|
|
return string(jsonBytes)
|
||
|
|
}
|
||
|
|
}
|
||
|
|
|
||
|
|
// getMessagePreview creates a short preview of a message
|
||
|
|
func (c *ConversationDiffAnalyzer) getMessagePreview(msg model.AnthropicMessage) string {
|
||
|
|
content := c.normalizeMessageContent(msg.Content)
|
||
|
|
if len(content) > 100 {
|
||
|
|
return content[:100] + "..."
|
||
|
|
}
|
||
|
|
return content
|
||
|
|
}
|
||
|
|
|
||
|
|
// max returns the maximum of two integers
|
||
|
|
func max(a, b int) int {
|
||
|
|
if a > b {
|
||
|
|
return a
|
||
|
|
}
|
||
|
|
return b
|
||
|
|
}
|