cchubber 0.1.0 → 0.3.0

package/src/readers/jsonl-reader.js

@@ -35,43 +35,27 @@ function readProjectsDir(dir, entries) {
 
     for (const hash of projectHashes) {
       const projectDir = join(dir, hash);
-      const jsonlFiles = readdirSync(projectDir).filter(f => f.endsWith('.jsonl'));
 
+      // Read top-level JSONL files (one per session)
+      const jsonlFiles = readdirSync(projectDir).filter(f => f.endsWith('.jsonl'));
       for (const file of jsonlFiles) {
-        const sessionId = basename(file, '.jsonl');
-        const filePath = join(projectDir, file);
-
-        try {
-          const raw = readFileSync(filePath, 'utf-8');
-          const lines = raw.split('\n').filter(l => l.trim());
-
-          for (const line of lines) {
-            try {
-              const record = JSON.parse(line);
-
-              // Only assistant messages have token usage
-              if (record.type !== 'assistant') continue;
-
-              const usage = record.message?.usage;
-              if (!usage) continue;
-
-              entries.push({
-                sessionId,
-                projectHash: hash,
-                timestamp: record.timestamp || '',
-                model: record.message?.model || 'unknown',
-                inputTokens: usage.input_tokens || 0,
-                outputTokens: usage.output_tokens || 0,
-                cacheCreationTokens: usage.cache_creation_input_tokens || 0,
-                cacheReadTokens: usage.cache_read_input_tokens || 0,
-                costUSD: record.costUSD || 0, // Pre-calculated by Claude Code
-              });
-            } catch {
-              // Skip malformed lines
+        readJsonlFile(join(projectDir, file), basename(file, '.jsonl'), hash, entries);
+      }
+
+      // Read subagent JSONL files (for Haiku/Sonnet model attribution)
+      // Dedup by message ID prevents double-counting
+      const subdirs = readdirSync(projectDir).filter(f => {
+        try { return statSync(join(projectDir, f)).isDirectory(); } catch { return false; }
+      });
+      for (const subdir of subdirs) {
+        const subagentDir = join(projectDir, subdir, 'subagents');
+        if (existsSync(subagentDir)) {
+          try {
+            const subFiles = readdirSync(subagentDir).filter(f => f.endsWith('.jsonl'));
+            for (const file of subFiles) {
+              readJsonlFile(join(subagentDir, file), basename(file, '.jsonl'), hash, entries);
             }
-          }
-        } catch {
-          // Skip unreadable files
+          } catch { /* skip */ }
         }
       }
     }
@@ -80,6 +64,51 @@ function readProjectsDir(dir, entries) {
   }
 }
 
+// Track seen message IDs to deduplicate (JSONL files contain dupes from session resume)
+const seenMessageIds = new Set();
+
+function readJsonlFile(filePath, sessionId, projectHash, entries) {
+  try {
+    const raw = readFileSync(filePath, 'utf-8');
+    const lines = raw.split('\n').filter(l => l.trim());
+
+    for (const line of lines) {
+      try {
+        const record = JSON.parse(line);
+
+        // Only assistant messages have token usage
+        if (record.type !== 'assistant') continue;
+
+        const usage = record.message?.usage;
+        if (!usage) continue;
+
+        // Deduplicate by message ID — JSONL files contain duplicates from session resume
+        const msgId = record.message?.id;
+        if (msgId) {
+          if (seenMessageIds.has(msgId)) continue;
+          seenMessageIds.add(msgId);
+        }
+
+        entries.push({
+          sessionId,
+          projectHash,
+          timestamp: record.timestamp || '',
+          model: record.message?.model || 'unknown',
+          inputTokens: usage.input_tokens || 0,
+          outputTokens: usage.output_tokens || 0,
+          cacheCreationTokens: usage.cache_creation_input_tokens || 0,
+          cacheReadTokens: usage.cache_read_input_tokens || 0,
+          costUSD: record.costUSD || 0,
+        });
+      } catch {
+        // Skip malformed lines
+      }
+    }
+  } catch {
+    // Skip unreadable files
+  }
+}
+
 /**
  * Aggregate JSONL entries into daily summaries.
  */
@@ -213,26 +242,17 @@ export function aggregateByProject(entries, claudeDir) {
     if (entry.timestamp < p.firstSeen) p.firstSeen = entry.timestamp;
   }
 
-  // Resolve project paths from .claude/projects/<hash>/.project_path if available
-  const projectsDir = join(claudeDir, 'projects');
+  // Decode project paths from directory names
+  // Claude Code encodes paths as: C--Users-asmir-Documents-Project-Name
+  // Decode: replace leading drive letter pattern, split on -, take last meaningful segments
   for (const proj of Object.values(byProject)) {
     proj.sessionCount = proj.sessions.size;
     delete proj.sessions;
 
-    // Try to read the project path file
-    try {
-      const pathFile = join(projectsDir, proj.hash, '.project_path');
-      if (existsSync(pathFile)) {
-        const raw = readFileSync(pathFile, 'utf-8').trim();
-        proj.path = raw;
-        // Use last directory segment as display name
-        proj.name = raw.split(/[/\\]/).filter(Boolean).pop() || proj.hash.slice(0, 8);
-      } else {
-        proj.name = proj.hash.slice(0, 8);
-      }
-    } catch {
-      proj.name = proj.hash.slice(0, 8);
-    }
+    // Decode the hash (which is the encoded path)
+    const decoded = decodeProjectHash(proj.hash);
+    proj.path = decoded.path;
+    proj.name = decoded.name;
   }
 
   return Object.values(byProject)
@@ -245,3 +265,50 @@ function cleanModelName(name) {
     .replace(/-\d{8}$/, '')
     .replace(/-20\d{6}$/, '');
 }
+
+/**
+ * Decode Claude Code's encoded project directory name into a readable path and name.
+ * Format: C--Users-asmir-Documents-Obsidian-Architect-OS-01-Projects-My-Project
+ * Becomes: C:/Users/asmir/Documents/.../My-Project → name: "My-Project"
+ */
+function decodeProjectHash(hash) {
+  if (!hash || hash === 'unknown') return { path: null, name: 'Unknown' };
+
+  // Replace the drive letter pattern: C-- → C:/
+  let decoded = hash.replace(/^([A-Z])--/, '$1:/');
+
+  // The rest uses - as separator, but some folder names have dashes too.
+  // Best heuristic: split on known path separators
+  // Common patterns: Users, Documents, Desktop, Projects, etc.
+  const pathSegments = decoded.split('-');
+
+  // Reconstruct a readable path
+  // The encoded format replaces / with - so we need to figure out boundaries
+  // Simple approach: reconstruct full path and extract last meaningful project name
+  const fullPath = decoded;
+
+  // Extract project name: take the last meaningful segments
+  // Skip common prefixes to find the project-specific part
+  const skipPrefixes = ['C:', 'Users', 'Documents', 'Desktop', 'Downloads', 'Obsidian', 'repos', 'projects', 'code', 'dev', 'src', 'home'];
+
+  let segments = hash.replace(/^[A-Z]--/, '').split('-');
+
+  // Find where the "interesting" name starts (after common path prefixes)
+  let nameStart = 0;
+  for (let i = 0; i < segments.length; i++) {
+    if (skipPrefixes.some(p => p.toLowerCase() === segments[i].toLowerCase())) {
+      nameStart = i + 1;
+    } else {
+      break;
+    }
+  }
+
+  // Take the last 2-3 meaningful segments as the project name
+  const nameSegments = segments.slice(Math.max(nameStart, segments.length - 3));
+  const name = nameSegments.join(' ') || hash.slice(0, 12);
+
+  // Reconstruct a shortened display path
+  const path = hash.replace(/^([A-Z])--/, '$1:/').replace(/-/g, '/');
+
+  return { path, name };
+}