@obtoai/agent-bridge 0.1.0-beta.2 → 0.1.0-beta.20

package/src/external-scanner.js

@@ -0,0 +1,546 @@
+'use strict';
+
+// Phase 6.1 — External Thread Discovery (scanner half).
+//
+// Scans the local filesystem for AI coding sessions started OUTSIDE the bridge
+// and returns a flat list of session records the daemon can POST to
+// /api/bridge/external/sync. The bridge UI then renders them alongside
+// bridge-owned threads — single pane of glass over all the user's AI work.
+//
+// Sources scanned:
+//   - Claude Code (CLI + VSCode Claude extension): both write JSONL session
+//     files at ~/.claude/projects/<encoded-projectdir>/<sessionId>.jsonl
+//   - Codex CLI: ~/.codex/sessions/YYYY/MM/DD/rollout-<ts>-<sessionId>.jsonl
+//   - opencode is NOT scanned — its SDK is server-bound, no shared JSONL store
+//   - Web tools (claude.ai chat, ChatGPT) are out of reach by design
+//
+// Privacy: we extract metadata + the LAST message preview only (1–2 lines,
+// capped at 200 chars). Full transcripts NEVER leave the user's machine.
+// The daemon POSTs the extracted records; the bridge stores them in the
+// agent_bridge_external_sessions Mongo collection.
+
+const fs = require('fs');
+const path = require('path');
+const os = require('os');
+
+const CLAUDE_DIR = path.join(os.homedir(), '.claude', 'projects');
+const CODEX_DIR = path.join(os.homedir(), '.codex', 'sessions');
+// Claude Desktop app's local-agent mode stores session JSONLs under
+// ~/Library/Application Support/Claude/local-agent-mode-sessions/<uuid>/<uuid>/local_<uuid>/.claude/projects/<encoded>/<sid>.jsonl
+// The file format is identical to Claude Code CLI's — same queue-operation
+// preamble, same user/assistant message shape. Just a different root.
+// Phase 6.2.5 — beta.19: surface these so the bridge sees every Claude
+// session on disk, not just the CLI/extension subset.
+const CLAUDE_DESKTOP_BASE = path.join(
+  os.homedir(),
+  'Library', 'Application Support', 'Claude', 'local-agent-mode-sessions',
+);
+
+const PREVIEW_MAX_CHARS = 200;
+const TITLE_MAX_CHARS = 80;
+const TAIL_READ_BYTES = 16384;       // fixed-budget tail (cheap; used for ai-title + lastMessage)
+const TAIL_STREAM_MAX_BYTES = 524288;// streaming tail cap — 512KB max read for recents (Phase 6.2.4)
+const TAIL_STREAM_CHUNK = 65536;     // 64KB chunk size for the backward stream
+const TAIL_STREAM_MIN_MESSAGES = 10; // stop reading once we have this many text-bearing lines
+const TITLE_MAX_LINES = 40;    // scan up to this many lines for a real first message
+const TITLE_MAX_BYTES = 65536; // hard ceiling per file even if MAX_LINES never reached
+
+// Read the tail of a (potentially large) JSONL file without slurping the
+// whole thing into memory. Returns a string (UTF-8) or '' on any failure.
+const readTail = (filePath, maxBytes = TAIL_READ_BYTES) => {
+  let fd = null;
+  try {
+    const stat = fs.statSync(filePath);
+    const size = stat.size;
+    const start = Math.max(0, size - maxBytes);
+    const len = size - start;
+    if (len <= 0) return '';
+    fd = fs.openSync(filePath, 'r');
+    const buf = Buffer.alloc(len);
+    fs.readSync(fd, buf, 0, len, start);
+    return buf.toString('utf8');
+  } catch (_) {
+    return '';
+  } finally {
+    if (fd !== null) { try { fs.closeSync(fd); } catch (_) {} }
+  }
+};
+
+// Phase 6.2.4 — streaming tail. For sessions where individual JSONL records
+// exceed the fixed 16KB budget (long assistant messages with lots of inline
+// code), the fixed tail returns 0 parseable lines. This reads backward in
+// 64KB chunks until we either have ≥TAIL_STREAM_MIN_MESSAGES text-bearing
+// user/assistant lines OR hit TAIL_STREAM_MAX_BYTES (512KB) — whichever
+// comes first. The first partial line at the buffer boundary is dropped
+// because it won't JSON.parse cleanly; that's fine, the next chunk will
+// pick it up complete.
+const readTailUntilMessages = (filePath) => {
+  let fd = null;
+  try {
+    const stat = fs.statSync(filePath);
+    if (stat.size === 0) return '';
+    fd = fs.openSync(filePath, 'r');
+    let buffer = '';
+    let pos = stat.size;
+    while (pos > 0 && buffer.length < TAIL_STREAM_MAX_BYTES) {
+      const readSize = Math.min(TAIL_STREAM_CHUNK, pos);
+      pos -= readSize;
+      const chunk = Buffer.alloc(readSize);
+      fs.readSync(fd, chunk, 0, readSize, pos);
+      buffer = chunk.toString('utf8') + buffer;
+      // Cheap completeness check: count parseable user/assistant lines with text.
+      const lines = buffer.split(/\r?\n/);
+      let count = 0;
+      for (const line of lines) {
+        if (!line.trim()) continue;
+        let obj;
+        try { obj = JSON.parse(line); } catch (_) { continue; }
+        const role = (obj && (obj.role || (obj.message && obj.message.role))) || null;
+        if (role !== 'user' && role !== 'assistant') continue;
+        const raw = (obj.message && obj.message.content) || obj.content;
+        if (!raw) continue;
+        if (typeof raw === 'string') {
+          if (raw.trim()) count++;
+        } else if (Array.isArray(raw)) {
+          for (const p of raw) {
+            if (p && (p.type === 'text' || typeof p.text === 'string') && String(p.text || '').trim()) {
+              count++; break;
+            }
+          }
+        }
+      }
+      if (count >= TAIL_STREAM_MIN_MESSAGES) break;
+    }
+    return buffer;
+  } catch (_) {
+    return '';
+  } finally {
+    if (fd !== null) { try { fs.closeSync(fd); } catch (_) {} }
+  }
+};
+
+// Pull the last N user/assistant *logical turns* from the tail, oldest-first.
+// Phase 6.2.3 — beta.17 fix: Claude Code streams a single assistant reply as
+// MULTIPLE JSONL lines (text → tool_use → tool_result → text → …), each its
+// own JSONL record. The earlier extractor took one line per turn and
+// fragmented responses into single-shard previews. Now we walk forward,
+// parse every text-bearing user/assistant line, and **coalesce consecutive
+// same-role lines into one logical turn** before slicing the last N. Result:
+// a real conversation, not a single mid-sentence excerpt.
+const RECENT_MESSAGE_BODY_MAX = 3000;
+const RECENT_TURN_COUNT = 10;
+
+const extractRecentMessages = (jsonlTail, n = RECENT_TURN_COUNT) => {
+  if (!jsonlTail) return [];
+  const lines = jsonlTail.split(/\r?\n/);
+  const parsed = [];
+  for (const lineRaw of lines) {
+    const line = lineRaw.trim();
+    if (!line) continue;
+    let obj;
+    try { obj = JSON.parse(line); } catch (_) { continue; }
+    let role = null, raw = null;
+    if (obj && obj.message && (obj.message.role || obj.type)) {
+      role = obj.message.role || (obj.type === 'user' ? 'user' : 'assistant');
+      raw = obj.message.content;
+    } else if (obj && obj.role && (obj.content || obj.text)) {
+      role = obj.role;
+      raw = obj.content != null ? obj.content : obj.text;
+    }
+    if (role !== 'user' && role !== 'assistant') continue;
+    if (raw == null) continue;
+    let text = '';
+    if (typeof raw === 'string') text = raw;
+    else if (Array.isArray(raw)) {
+      text = raw
+        .filter((p) => p && (p.type === 'text' || typeof p.text === 'string'))
+        .map((p) => String(p.text || ''))
+        .join('\n');
+    }
+    // Phase 6.2.6 — keep newlines so the bridge UI can render markdown
+    // (tables, lists, code blocks, paragraphs). Collapsing every \s+ to a
+    // single space was stripping all the formatting before storage. We
+    // still collapse runs of horizontal whitespace and excessive blank
+    // lines, but real line breaks survive.
+    text = text.replace(/[ \t]+/g, ' ').replace(/\n{3,}/g, '\n\n').trim();
+    if (!text) continue; // skip pure tool_use / tool_result lines (no text body)
+    parsed.push({ role, text, ts: obj.timestamp || null });
+  }
+
+  // Coalesce consecutive same-role lines (one logical turn split across JSONL
+  // records). This is the actual fix — without it, assistant turns with mid-
+  // reply tool calls fragment into the first text shard only. Joining with
+  // \n\n preserves paragraph boundaries between coalesced chunks so markdown
+  // renders correctly.
+  const coalesced = [];
+  for (const p of parsed) {
+    const last = coalesced[coalesced.length - 1];
+    if (last && last.role === p.role) {
+      last.text = (last.text + '\n\n' + p.text).replace(/\n{3,}/g, '\n\n').trim();
+      last.ts = p.ts || last.ts;
+    } else {
+      coalesced.push({ role: p.role, text: p.text, ts: p.ts });
+    }
+  }
+
+  // Filter user turns that are pure platform-injection noise. Assistant turns
+  // are never filtered — they're always real Claude/Codex output.
+  const filtered = coalesced.filter((m) => {
+    if (m.role !== 'user') return true;
+    return !isInjectionMessage(m.text);
+  });
+
+  const sliced = filtered.slice(-n);
+  return sliced.map((m) => ({
+    role: m.role,
+    body: m.text.length > RECENT_MESSAGE_BODY_MAX ? m.text.slice(0, RECENT_MESSAGE_BODY_MAX) : m.text,
+    ts: m.ts,
+  }));
+};
+
+// Claude Code writes an LLM-generated title as a `type: "ai-title"` JSONL
+// record near the end of each session file (this is the same title VSCode's
+// session list shows — "Analyze MongoDB MCP server architecture" style).
+// If we find one, it beats anything we could extract from the user's first
+// raw prompt. Scan the tail backwards to hit it fast.
+const extractAiTitleFromTail = (jsonlTail) => {
+  if (!jsonlTail) return '';
+  const lines = jsonlTail.split(/\r?\n/);
+  for (let i = lines.length - 1; i >= 0; i--) {
+    const line = lines[i];
+    // Cheap pre-filter so we only JSON.parse candidate lines.
+    if (line.indexOf('ai-title') === -1 && line.indexOf('aiTitle') === -1) continue;
+    try {
+      const obj = JSON.parse(line);
+      if (obj && obj.type === 'ai-title' && typeof obj.aiTitle === 'string') {
+        const t = obj.aiTitle.trim();
+        if (t) return t.length > TITLE_MAX_CHARS ? t.slice(0, TITLE_MAX_CHARS) : t;
+      }
+    } catch (_) { /* not a JSON line — keep walking */ }
+  }
+  return '';
+};
+
+
+// "user" messages BEFORE the human's first real prompt — system reminders,
+// IDE state, untrusted-metadata wrappers, command blocks, etc. These are
+// noise from a label perspective. Filter them so the first PLAIN user
+// prompt is what we use as the thread title.
+const isInjectionMessage = (text) => {
+  const s = String(text || '').trim();
+  if (!s) return true;
+  if (/^<[a-zA-Z!]/.test(s)) return true;                         // <system-reminder>, <ide_opened_file>, etc.
+  if (/^Sender \(untrusted metadata\)/i.test(s)) return true;
+  if (/^Conversation info \(untrusted metadata\)/i.test(s)) return true;
+  if (/^untrusted metadata/i.test(s)) return true;
+  if (/^Caveat:/i.test(s)) return true;
+  if (/^Claude was launched/i.test(s)) return true;
+  if (/^```json/i.test(s)) return true;
+  if (/^`<command-name>/.test(s)) return true;
+  if (/^This session is being continued/i.test(s)) return true;   // /resume preamble
+  return false;
+};
+
+// Walk a JSONL file line-by-line until we find a plain user message (skipping
+// metadata injections) or we've checked TITLE_MAX_LINES / TITLE_MAX_BYTES.
+// Streaming so big files don't get fully slurped. Returns a clean ≤80-char
+// title or '' if nothing meaningful was found in the bound.
+const extractTitleFromFile = (filePath) => {
+  let fd = null;
+  try {
+    fd = fs.openSync(filePath, 'r');
+    const CHUNK = 16 * 1024;
+    const buf = Buffer.alloc(CHUNK);
+    let carry = '';
+    let bytesRead = 0;
+    let linesChecked = 0;
+    while (bytesRead < TITLE_MAX_BYTES && linesChecked < TITLE_MAX_LINES) {
+      const n = fs.readSync(fd, buf, 0, CHUNK, bytesRead);
+      if (n === 0) break;
+      bytesRead += n;
+      carry += buf.toString('utf8', 0, n);
+      const lines = carry.split(/\r?\n/);
+      carry = lines.pop() || '';
+      for (const line of lines) {
+        if (linesChecked >= TITLE_MAX_LINES) break;
+        if (!line.trim()) continue;
+        linesChecked++;
+        let obj;
+        try { obj = JSON.parse(line); } catch (_) { continue; }
+        let role = null, raw = null;
+        if (obj && obj.message && (obj.message.role || obj.type)) {
+          role = obj.message.role || (obj.type === 'user' ? 'user' : 'assistant');
+          raw = obj.message.content;
+        } else if (obj && obj.role && (obj.content || obj.text)) {
+          role = obj.role;
+          raw = obj.content != null ? obj.content : obj.text;
+        }
+        if (role !== 'user' || raw == null) continue;
+        let text = '';
+        if (typeof raw === 'string') text = raw;
+        else if (Array.isArray(raw)) {
+          text = raw
+            .filter((p) => p && (p.type === 'text' || typeof p.text === 'string'))
+            .map((p) => String(p.text || ''))
+            .join('\n');
+        }
+        text = text.replace(/\s+/g, ' ').trim();
+        if (!text) continue;
+        if (isInjectionMessage(text)) continue;
+        return text.length > TITLE_MAX_CHARS ? text.slice(0, TITLE_MAX_CHARS) : text;
+      }
+    }
+    return '';
+  } catch (_) {
+    return '';
+  } finally {
+    if (fd !== null) { try { fs.closeSync(fd); } catch (_) {} }
+  }
+};
+
+// Walk a JSONL tail backwards, parse each non-empty line as JSON, return the
+// first one we can extract a message from. Tolerant of multiple shapes —
+// Claude and Codex write slightly different envelopes and the formats have
+// drifted across SDK versions.
+const extractLastMessage = (jsonlTail) => {
+  if (!jsonlTail) return null;
+  const lines = jsonlTail.split(/\r?\n/).filter((l) => l.trim().length > 0);
+  for (let i = lines.length - 1; i >= 0; i--) {
+    let obj;
+    try {
+      obj = JSON.parse(lines[i]);
+    } catch (_) {
+      continue;
+    }
+
+    // ── Claude Code session JSONL shapes ─────────────────────────────────
+    // Common:
+    //   { type: 'user',      message: { role: 'user',      content: '...' } }
+    //   { type: 'assistant', message: { role: 'assistant', content: [{type:'text', text:'...'}, ...] } }
+    if (obj && obj.message && (obj.message.role || obj.type)) {
+      const role = obj.message.role || (obj.type === 'user' ? 'user' : 'assistant');
+      let raw = obj.message.content;
+      let text = '';
+      if (typeof raw === 'string') text = raw;
+      else if (Array.isArray(raw)) {
+        text = raw
+          .filter((p) => p && (p.type === 'text' || typeof p.text === 'string'))
+          .map((p) => String(p.text || ''))
+          .join('\n');
+      }
+      text = text.trim();
+      if (text) return { author: role === 'user' ? 'user' : 'assistant', preview: text.slice(0, PREVIEW_MAX_CHARS) };
+    }
+
+    // ── Codex SDK rollout shapes ────────────────────────────────────────
+    //   { record_type: 'message', role: 'assistant', content: [{type:'text', text:'...'}] }
+    //   { type: 'message', role: 'user', content: '...' }
+    //   { event: 'output_text', text: '...', role: 'assistant' }
+    if (obj && (obj.role || obj.event) && (obj.content || obj.text)) {
+      const role = obj.role || (obj.event === 'input_text' ? 'user' : 'assistant');
+      let raw = obj.content != null ? obj.content : obj.text;
+      let text = '';
+      if (typeof raw === 'string') text = raw;
+      else if (Array.isArray(raw)) {
+        text = raw
+          .filter((p) => p && (p.type === 'text' || typeof p.text === 'string'))
+          .map((p) => String(p.text || ''))
+          .join('\n');
+      }
+      text = text.trim();
+      if (text) return { author: role === 'user' ? 'user' : 'assistant', preview: text.slice(0, PREVIEW_MAX_CHARS) };
+    }
+  }
+  return null;
+};
+
+// Decode Claude's project-dir filename encoding back to a path-like string.
+// Claude turns `/Users/divyansh/foo` → `-Users-divyansh-foo`. We can't
+// perfectly reverse it (project names with literal `-` are ambiguous), but
+// for display purposes leading-dash → leading-slash + dashes → slashes is
+// usually close enough. The BridgeExternal stores both the raw encoded
+// projectDir AND a decoded label; the view route picks the friendlier one.
+const decodeClaudeProjectDir = (encoded) => {
+  if (!encoded) return '';
+  if (encoded.startsWith('-')) {
+    return '/' + encoded.slice(1).replace(/-/g, '/');
+  }
+  return encoded.replace(/-/g, '/');
+};
+
+// Discover all Claude Desktop "local-agent-mode" project roots.
+// Structure is 3 levels deep before we hit `.claude/projects/`:
+//   <BASE>/<accountUuid>/<workspaceUuid>/local_<sessionUuid>/.claude/projects/
+// We collect every leaf `.claude/projects` dir, then walk each like we walk
+// the CLI's ~/.claude/projects.
+const findClaudeDesktopProjectRoots = () => {
+  const out = [];
+  let l1;
+  try { l1 = fs.readdirSync(CLAUDE_DESKTOP_BASE); } catch (_) { return out; }
+  for (const a of l1) {
+    const aPath = path.join(CLAUDE_DESKTOP_BASE, a);
+    let aStat; try { aStat = fs.statSync(aPath); } catch (_) { continue; }
+    if (!aStat.isDirectory()) continue;
+    let l2; try { l2 = fs.readdirSync(aPath); } catch (_) { continue; }
+    for (const b of l2) {
+      const bPath = path.join(aPath, b);
+      let bStat; try { bStat = fs.statSync(bPath); } catch (_) { continue; }
+      if (!bStat.isDirectory()) continue;
+      let l3; try { l3 = fs.readdirSync(bPath); } catch (_) { continue; }
+      for (const c of l3) {
+        if (!c.startsWith('local_')) continue;
+        const projects = path.join(bPath, c, '.claude', 'projects');
+        try { if (fs.statSync(projects).isDirectory()) out.push(projects); } catch (_) {}
+      }
+    }
+  }
+  return out;
+};
+
+// Walk a single Claude projects root (works for both ~/.claude/projects and
+// each of the Desktop app's local-agent-mode project roots — file format is
+// identical, only the path differs).
+const walkClaudeProjectsRoot = (root) => {
+  const out = [];
+  let topEntries;
+  try { topEntries = fs.readdirSync(root); } catch (_) { return out; }
+
+  for (const entry of topEntries) {
+    const projectPath = path.join(root, entry);
+    let projectStat;
+    try { projectStat = fs.statSync(projectPath); } catch (_) { continue; }
+    if (!projectStat.isDirectory()) continue;
+
+    let sessionFiles;
+    try { sessionFiles = fs.readdirSync(projectPath); } catch (_) { continue; }
+
+    for (const file of sessionFiles) {
+      if (!file.endsWith('.jsonl')) continue;
+      const sessionId = file.slice(0, -'.jsonl'.length);
+      const filePath = path.join(projectPath, file);
+      let stat;
+      try { stat = fs.statSync(filePath); } catch (_) { continue; }
+      const tail = readTail(filePath);
+      const lastMsg = extractLastMessage(tail);
+      let title = extractAiTitleFromTail(tail);
+      if (!title) title = extractTitleFromFile(filePath);
+      const recentMessages = extractRecentMessages(readTailUntilMessages(filePath));
+      out.push({
+        source: 'claude',
+        sessionId,
+        projectDir: entry,
+        projectName: decodeClaudeProjectDir(entry),
+        title: title,
+        recentMessages: recentMessages,
+        lastActivityAt: stat.mtimeMs,
+        lastMessagePreview: lastMsg ? lastMsg.preview : '',
+        lastMessageAuthor: lastMsg ? lastMsg.author : null,
+      });
+    }
+  }
+  return out;
+};
+
+// Scan all Claude session storage on this machine — CLI + VSCode extension
+// (~/.claude/projects) AND every Claude Desktop local-agent-mode subdir
+// (~/Library/Application Support/Claude/local-agent-mode-sessions/.../).
+const scanClaude = () => {
+  const roots = [CLAUDE_DIR].concat(findClaudeDesktopProjectRoots());
+  const out = [];
+  for (const root of roots) {
+    out.push(...walkClaudeProjectsRoot(root));
+  }
+  return out;
+};
+
+// Scan ~/.codex/sessions/YYYY/MM/DD/rollout-<ts>-<sessionId>.jsonl
+// The first JSONL line for a Codex rollout is a session-meta record that
+// contains the working directory; we read it once for projectDir.
+const scanCodex = () => {
+  const out = [];
+  let years;
+  try { years = fs.readdirSync(CODEX_DIR); } catch (_) { return out; }
+
+  for (const y of years) {
+    if (!/^\d{4}$/.test(y)) continue;
+    const yPath = path.join(CODEX_DIR, y);
+    let months;
+    try { months = fs.readdirSync(yPath); } catch (_) { continue; }
+    for (const m of months) {
+      if (!/^\d{2}$/.test(m)) continue;
+      const mPath = path.join(yPath, m);
+      let days;
+      try { days = fs.readdirSync(mPath); } catch (_) { continue; }
+      for (const d of days) {
+        if (!/^\d{2}$/.test(d)) continue;
+        const dPath = path.join(mPath, d);
+        let files;
+        try { files = fs.readdirSync(dPath); } catch (_) { continue; }
+        for (const f of files) {
+          if (!f.startsWith('rollout-') || !f.endsWith('.jsonl')) continue;
+          // session id is the last hex/uuid block before .jsonl
+          const sidMatch = f.match(/-([0-9a-f-]{8,})\.jsonl$/i);
+          if (!sidMatch) continue;
+          const sessionId = sidMatch[1];
+          const filePath = path.join(dPath, f);
+          let stat;
+          try { stat = fs.statSync(filePath); } catch (_) { continue; }
+
+          // Read the first KB to pull the session-meta's working directory.
+          let projectDir = '';
+          let fd = null;
+          try {
+            fd = fs.openSync(filePath, 'r');
+            const headBuf = Buffer.alloc(Math.min(2048, stat.size));
+            fs.readSync(fd, headBuf, 0, headBuf.length, 0);
+            const firstLine = headBuf.toString('utf8').split(/\r?\n/)[0] || '';
+            try {
+              const meta = JSON.parse(firstLine);
+              projectDir = String(
+                meta?.cwd ||
+                meta?.workingDirectory ||
+                meta?.working_directory ||
+                meta?.session_meta?.cwd ||
+                meta?.payload?.cwd ||
+                ''
+              );
+            } catch (_) { /* not a meta line — leave projectDir blank */ }
+          } catch (_) {} finally {
+            if (fd !== null) { try { fs.closeSync(fd); } catch (_) {} }
+          }
+
+          const tail = readTail(filePath);
+          const lastMsg = extractLastMessage(tail);
+          const title = extractTitleFromFile(filePath);
+          // Recent messages get the streaming tail (handles huge per-message
+      // assistant turns that overflow the 16KB fixed budget).
+      const recentMessages = extractRecentMessages(readTailUntilMessages(filePath));
+          out.push({
+            source: 'codex',
+            sessionId,
+            projectDir: projectDir || `${y}/${m}/${d}`,
+            projectName: projectDir || null,
+            title: title,
+            recentMessages: recentMessages,
+            lastActivityAt: stat.mtimeMs,
+            lastMessagePreview: lastMsg ? lastMsg.preview : '',
+            lastMessageAuthor: lastMsg ? lastMsg.author : null,
+          });
+        }
+      }
+    }
+  }
+  return out;
+};
+
+// Public entry: returns a flat list of every external session found.
+// Synchronous on purpose — the daemon calls this on a 30s timer and the
+// total IO is dominated by readdir + a single readSync per file. Async
+// would only complicate retry/cancel semantics with no real benefit.
+const scanAll = () => {
+  const claude = scanClaude();
+  const codex = scanCodex();
+  return claude.concat(codex);
+};
+
+module.exports = { scanAll, scanClaude, scanCodex, extractLastMessage, decodeClaudeProjectDir };