@adhdev/daemon-core 0.9.82-rc.49 → 0.9.82-rc.490

package/src/providers/native-history/antigravity-cli-transcript.ts

@@ -0,0 +1,1275 @@
+/**
+ * antigravity-cli-transcript — Daemon-side built-in native history adapter for antigravity-cli.
+ *
+ * Antigravity CLI stores conversations in multiple complementary locations:
+ *
+ *   1. ~/.gemini/antigravity-cli/history.jsonl
+ *      Lightweight index of user prompts, one row per turn:
+ *        { conversationId: string, display: string, workspace: string, timestamp: number }
+ *
+ *   2. ~/.gemini/antigravity-cli/brain/<uuid>/.system_generated/logs/transcript*.jsonl
+ *      Per-session full transcript, one row per agent step:
+ *        { source: 'USER_EXPLICIT'|'MODEL', type: string, content: string, status: 'DONE'|..., created_at: number }
+ *
+ *   3. ~/.gemini/antigravity-cli/conversations/<uuid>.pb
+ *      Legacy protobuf binary — schema not publicly documented. Adapter extracts
+ *      printable UTF-8 text runs as best-effort content (no proto library needed).
+ *
+ *   4. ~/.gemini/antigravity-cli/conversations/<uuid>.db  ← current format
+ *      Per-session SQLite database. Recent antigravity migrated conversation
+ *      storage from .pb (+ brain/*.jsonl) to a per-session SQLite db. The
+ *      schema is a trajectory of `steps`, NOT a simple messages(role,content)
+ *      table:
+ *        steps(idx INTEGER PK, step_type INTEGER, status INTEGER,
+ *              step_payload BLOB [protobuf], ...)
+ *      Each `step_payload` is a protobuf message. Empirically (introspected
+ *      from real stores):
+ *        - step_type 14 → a USER turn. The prompt text is the largest
+ *          contiguous UTF-8 run inside the payload (field 19 subtree).
+ *        - step_type 15 → a MODEL/assistant turn. The assistant's final
+ *          natural-language answer lives at payload field 20 → field 1
+ *          (identical to field 8). Field 20 → field 3 is the internal
+ *          reasoning summary and is intentionally NOT surfaced.
+ *        - other step types are tool calls / ephemeral system context.
+ *      We read the blobs with a tiny dependency-free protobuf field walker
+ *      (no proto schema / codegen needed) and map the two message step types.
+ *      Because the daemon does NOT read this db, native history previously
+ *      returned 0 rows for these sessions and read_chat fell back to the
+ *      pty parser (which only echoes the user's own input) — assistant
+ *      answers appeared lost even though they were on disk.
+ *
+ *      Schema-drift resilience: the exact field path (20 → 1/8 for the answer,
+ *      19 → 2/3 for the prompt) is empirically verified against real stores, but
+ *      antigravity may move it in a future build. So when the known path yields
+ *      no text, instead of silently dropping the turn we fall back to a UTF-8
+ *      printable-run scan of the payload (recoverMessageText) that recovers the
+ *      answer/prompt even if the field number drifted — while explicitly
+ *      EXCLUDING the reasoning subtree (20 → 3) so internal reasoning is never
+ *      surfaced as the answer. Only when even that finds nothing beyond
+ *      reasoning/metadata is the step dropped, and a content-free DEBUG
+ *      breadcrumb is logged so a real drift is greppable rather than invisible.
+ *
+ * This adapter provides:
+ *   - Full coverage from a per-session .db (current format) — preferred.
+ *   - Full coverage when a brain transcript exists (legacy authoritative source).
+ *   - Partial coverage (user prompts only) from history.jsonl as fallback.
+ *   - Best-effort raw-string extraction from .pb files when no other source exists.
+ *
+ * watchPath (from provider.v1.json):
+ *   ~/.gemini/antigravity-cli/history.jsonl
+ *   ~/.gemini/antigravity-cli/brain/{uuid}/.system_generated/logs/transcript*.jsonl
+ *   ~/.gemini/antigravity-cli/conversations/{uuid}.pb
+ *   ~/.gemini/antigravity-cli/conversations/{uuid}.db
+ *
+ * OSS code (AGPL-3.0). Must not import from packages/ (proprietary).
+ */
+
+import * as fs from 'fs';
+import * as path from 'path';
+import * as os from 'os';
+import { loadBetterSqlite3 } from '../../system/load-better-sqlite3.js';
+import { LOG } from '../../logging/logger.js';
+
+// ─── Types ─────────────────────────────────────────────────────────────────
+
+export type NativeHistoryRole = 'user' | 'assistant' | 'system';
+export type NativeHistoryKind = 'standard' | 'tool' | 'session_start';
+
+export interface NativeHistoryMessage {
+  ts: string;
+  receivedAt: number;
+  role: NativeHistoryRole;
+  content: string;
+  kind: NativeHistoryKind;
+  senderName?: string;
+  agent: 'antigravity-cli';
+  historySessionId: string;
+  workspace?: string;
+  /** Stable per-message identity (v2 contract). */
+  providerUnitKey?: string;
+}
+
+export interface NativeHistorySession {
+  messages: NativeHistoryMessage[];
+  providerSessionId: string;
+  source: 'provider-native';
+  sourcePath: string;
+  sourceMtimeMs: number;
+  /**
+   * 'full' — brain transcript was available and parsed.
+   * 'partial' — only history.jsonl user-prompts are available.
+   * 'best-effort' — .pb file bytes were used for raw text extraction.
+   */
+  nativeHistoryCoverage: 'full' | 'partial' | 'best-effort';
+  workspace?: string;
+  partialReason?: string;
+}
+
+export interface NativeHistorySessionMeta {
+  historySessionId: string;
+  sessionId: string;
+  sourcePath: string;
+  sourceMtimeMs: number;
+  messageCount: number;
+  firstMessageAt: number;
+  lastMessageAt: number;
+  sessionTitle?: string;
+  preview?: string;
+  workspace?: string;
+  agent: 'antigravity-cli';
+  source: 'provider-native';
+  nativeHistoryCoverage: 'full' | 'partial' | 'best-effort';
+  partialReason?: string;
+}
+
+// ─── Internal helpers ───────────────────────────────────────────────────────
+
+function extractTimestampValue(value: unknown): number {
+  if (typeof value === 'number' && Number.isFinite(value) && value > 0) return value;
+  if (typeof value === 'string') {
+    const numeric = Number(value);
+    if (Number.isFinite(numeric) && numeric > 0) return numeric;
+    const parsed = Date.parse(value);
+    if (Number.isFinite(parsed) && parsed > 0) return parsed;
+  }
+  return 0;
+}
+
+function statMtimeMs(filePath: string): number {
+  try { return fs.statSync(filePath).mtimeMs; } catch { return 0; }
+}
+
+function isUuidLike(value: string): boolean {
+  return /^[0-9a-f]{8}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{12}$/i.test(value);
+}
+
+function antigravityRoot(): string {
+  return path.join(os.homedir(), '.gemini', 'antigravity-cli');
+}
+
+function historyJsonlPath(): string {
+  return path.join(antigravityRoot(), 'history.jsonl');
+}
+
+function brainRoot(): string {
+  return path.join(antigravityRoot(), 'brain');
+}
+
+function conversationsRoot(): string {
+  return path.join(antigravityRoot(), 'conversations');
+}
+
+function resolvePathInside(root: string, ...segments: string[]): string | null {
+  const rootPath = path.resolve(root);
+  const targetPath = path.resolve(rootPath, ...segments);
+  if (targetPath !== rootPath && !targetPath.startsWith(rootPath + path.sep)) return null;
+  return targetPath;
+}
+
+/**
+ * Find the brain transcript file for a given session UUID.
+ * Prefers transcript_full.jsonl over transcript.jsonl when both exist.
+ */
+function findBrainTranscriptPath(sessionId: string): string | null {
+  if (!isUuidLike(sessionId)) return null;
+  const logsRoot = resolvePathInside(brainRoot(), sessionId, '.system_generated', 'logs');
+  if (!logsRoot || !fs.existsSync(logsRoot)) return null;
+
+  const candidates = (['transcript_full.jsonl', 'transcript.jsonl'] as const)
+    .map((file) => resolvePathInside(logsRoot, file))
+    .filter((p): p is string => p !== null && fs.existsSync(p));
+
+  if (candidates.length === 0) {
+    // Look for any transcript*.jsonl
+    let entries: fs.Dirent[] = [];
+    try { entries = fs.readdirSync(logsRoot, { withFileTypes: true }); } catch { return null; }
+    const transcriptFiles = entries
+      .filter((e) => e.isFile() && /^transcript.*\.jsonl$/.test(e.name))
+      .map((e) => path.join(logsRoot, e.name));
+    if (transcriptFiles.length === 0) return null;
+    transcriptFiles.sort((a, b) => statMtimeMs(b) - statMtimeMs(a));
+    return transcriptFiles[0];
+  }
+
+  candidates.sort((a, b) => statMtimeMs(b) - statMtimeMs(a));
+  return candidates[0];
+}
+
+/**
+ * List all session UUIDs that have a brain transcript directory.
+ */
+function listBrainSessions(): Array<{ sessionId: string; transcriptPath: string; sourceMtimeMs: number }> {
+  const root = brainRoot();
+  let entries: fs.Dirent[] = [];
+  try { entries = fs.readdirSync(root, { withFileTypes: true }); } catch { return []; }
+
+  const results: Array<{ sessionId: string; transcriptPath: string; sourceMtimeMs: number }> = [];
+  for (const entry of entries) {
+    if (!entry.isDirectory() || !isUuidLike(entry.name)) continue;
+    const transcriptPath = findBrainTranscriptPath(entry.name);
+    if (!transcriptPath) continue;
+    results.push({
+      sessionId: entry.name,
+      transcriptPath,
+      sourceMtimeMs: statMtimeMs(transcriptPath),
+    });
+  }
+  return results;
+}
+
+/**
+ * Strip USER_REQUEST XML wrapper from antigravity user prompts.
+ * Antigravity wraps structured user input in <USER_REQUEST>...</USER_REQUEST>.
+ */
+function extractUserRequestContent(content: string): string {
+  const raw = content.trim();
+  const match = raw.match(/<USER_REQUEST>\s*([\s\S]*?)\s*<\/USER_REQUEST>/i);
+  if (match) return match[1].trim();
+  return raw
+    .replace(/<ADDITIONAL_METADATA>[\s\S]*?<\/ADDITIONAL_METADATA>/gi, '')
+    .replace(/<USER_SETTINGS_CHANGE>[\s\S]*?<\/USER_SETTINGS_CHANGE>/gi, '')
+    .replace(/<\/?USER_REQUEST>/gi, '')
+    .trim();
+}
+
+/**
+ * Determine the NativeHistoryKind from an antigravity transcript row type.
+ */
+function antigravityRowKind(rowType: string): NativeHistoryKind {
+  if (rowType === 'PLANNER_RESPONSE') return 'standard';
+  if (rowType && rowType !== 'USER_INPUT') return 'tool';
+  return 'standard';
+}
+
+// ─── Brain transcript parser ─────────────────────────────────────────────────
+
+interface TranscriptRow {
+  source?: string;
+  type?: string;
+  content?: string;
+  status?: string;
+  created_at?: unknown;
+}
+
+/**
+ * Parse a brain transcript JSONL file into NativeHistoryMessages.
+ * Only rows with status === 'DONE' are processed.
+ * USER_EXPLICIT / USER_INPUT rows → user role.
+ * MODEL rows → assistant role.
+ */
+function parseBrainTranscript(
+  filePath: string,
+  sessionId: string,
+  workspace?: string,
+): NativeHistoryMessage[] | null {
+  let raw: string;
+  try { raw = fs.readFileSync(filePath, 'utf-8'); } catch { return null; }
+
+  const lines = raw.split('\n').filter(Boolean);
+  if (lines.length === 0) return null;
+
+  const records: NativeHistoryMessage[] = [];
+  const normalizedWorkspace = typeof workspace === 'string' ? workspace.trim() : '';
+  let fallbackTs = statMtimeMs(filePath) || Date.now();
+
+  for (const line of lines) {
+    let row: TranscriptRow | null = null;
+    try { row = JSON.parse(line) as TranscriptRow; } catch { continue; }
+    if (!row || typeof row !== 'object') continue;
+    if (row.status !== 'DONE') continue;
+
+    const rowSource = String(row.source ?? '').trim();
+    const rowType = String(row.type ?? '').trim();
+    const rawContent = typeof row.content === 'string' ? row.content : '';
+    const receivedAt = extractTimestampValue(row.created_at) || fallbackTs + records.length;
+
+    if (rowSource === 'USER_EXPLICIT' && rowType === 'USER_INPUT') {
+      const content = extractUserRequestContent(rawContent);
+      if (!content) continue;
+      const msg: NativeHistoryMessage = {
+        ts: new Date(receivedAt).toISOString(),
+        receivedAt,
+        role: 'user',
+        content,
+        kind: 'standard',
+        agent: 'antigravity-cli',
+        historySessionId: sessionId,
+      };
+      if (normalizedWorkspace) msg.workspace = normalizedWorkspace;
+      records.push(msg);
+    } else if (rowSource === 'MODEL') {
+      const content = rawContent.trim();
+      if (!content) continue;
+      const kind = antigravityRowKind(rowType);
+      const msg: NativeHistoryMessage = {
+        ts: new Date(receivedAt).toISOString(),
+        receivedAt,
+        role: 'assistant',
+        content,
+        kind,
+        agent: 'antigravity-cli',
+        historySessionId: sessionId,
+      };
+      if (kind === 'tool') msg.senderName = 'Tool';
+      if (normalizedWorkspace) msg.workspace = normalizedWorkspace;
+      records.push(msg);
+    }
+  }
+
+  return records.length > 0 ? records : null;
+}
+
+// ─── History JSONL parser ────────────────────────────────────────────────────
+
+interface HistoryRow {
+  conversationId: string;
+  display: string;
+  workspace: string;
+  receivedAt: number;
+}
+
+function readHistoryRows(): HistoryRow[] {
+  const sourcePath = historyJsonlPath();
+  let lines: string[] = [];
+  try { lines = fs.readFileSync(sourcePath, 'utf-8').split('\n').filter(Boolean); } catch { return []; }
+
+  const rows: HistoryRow[] = [];
+  for (const line of lines) {
+    let parsed: unknown = null;
+    try { parsed = JSON.parse(line); } catch { continue; }
+    if (!parsed || typeof parsed !== 'object') continue;
+
+    const row = parsed as Record<string, unknown>;
+    const conversationId = String(row.conversationId ?? '').trim();
+    const display = typeof row.display === 'string' ? row.display.trim() : '';
+    const workspace = typeof row.workspace === 'string' ? row.workspace.trim() : '';
+    const receivedAt = extractTimestampValue(row.timestamp);
+
+    if (!conversationId || !isUuidLike(conversationId) || !display || !receivedAt) continue;
+    rows.push({ conversationId, display, workspace, receivedAt });
+  }
+  return rows;
+}
+
+// ─── Protobuf best-effort text extraction ────────────────────────────────────
+
+const MIN_PRINTABLE_RUN = 8;
+
+/**
+ * Extract printable UTF-8 text runs from raw binary data (similar to the `strings` command).
+ * Runs of printable ASCII/UTF-8 characters of length >= MIN_PRINTABLE_RUN are collected.
+ * This is used as a best-effort fallback when the .pb schema is not available.
+ */
+function extractStringsFromBuffer(buf: Buffer): string[] {
+  const strings: string[] = [];
+  let current: number[] = [];
+
+  for (let i = 0; i < buf.length; i++) {
+    const byte = buf[i];
+    // Accept printable ASCII (32-126) and common whitespace (9=tab, 10=LF, 13=CR)
+    if ((byte >= 32 && byte <= 126) || byte === 9 || byte === 10 || byte === 13) {
+      current.push(byte);
+    } else {
+      if (current.length >= MIN_PRINTABLE_RUN) {
+        const str = Buffer.from(current).toString('utf-8').trim();
+        if (str) strings.push(str);
+      }
+      current = [];
+    }
+  }
+  // Flush remaining
+  if (current.length >= MIN_PRINTABLE_RUN) {
+    const str = Buffer.from(current).toString('utf-8').trim();
+    if (str) strings.push(str);
+  }
+
+  return strings;
+}
+
+/**
+ * Read a .pb conversation file and extract best-effort text content.
+ * Returns a single NativeHistoryMessage with the extracted content,
+ * or null if nothing readable was found.
+ */
+function parsePbFile(
+  filePath: string,
+  sessionId: string,
+): NativeHistoryMessage[] | null {
+  let buf: Buffer;
+  try { buf = fs.readFileSync(filePath); } catch { return null; }
+  if (buf.length === 0) return null;
+
+  const strings = extractStringsFromBuffer(buf);
+  // Filter out very short or likely-binary strings
+  const meaningful = strings.filter((s) => s.length >= MIN_PRINTABLE_RUN && /\w/.test(s));
+  if (meaningful.length === 0) return null;
+
+  const content = meaningful.join('\n');
+  const sourceMtimeMs = statMtimeMs(filePath);
+
+  return [
+    {
+      ts: new Date(sourceMtimeMs).toISOString(),
+      receivedAt: sourceMtimeMs,
+      role: 'assistant',
+      content,
+      kind: 'standard',
+      agent: 'antigravity-cli',
+      historySessionId: sessionId,
+    },
+  ];
+}
+
+// ─── SQLite (.db) conversation reader ────────────────────────────────────────
+//
+// Recent antigravity stores each conversation in a per-session SQLite db at
+// conversations/<uuid>.db. See the file header for the schema. We decode the
+// protobuf `step_payload` blobs with a minimal, dependency-free field walker —
+// we only need two leaf strings (user prompt / assistant answer), so a full
+// proto schema is unnecessary.
+
+/** Antigravity step_type values that map to a chat message. */
+const AGY_STEP_TYPE_USER = 14;
+const AGY_STEP_TYPE_MODEL = 15;
+
+interface ProtoField {
+  field: number;
+  wireType: number;
+  /** For wireType 2 (length-delimited): the raw bytes. */
+  bytes?: Buffer;
+  /** For wireType 0 (varint): the value. */
+  varint?: number;
+}
+
+/**
+ * Read a base-128 varint starting at `offset`. Returns [value, nextOffset].
+ * Values are read as JS numbers (safe: the fields we consume are small).
+ */
+function readVarint(buf: Buffer, offset: number): [number, number] {
+  let result = 0;
+  let shift = 0;
+  let i = offset;
+  while (i < buf.length) {
+    const byte = buf[i];
+    i += 1;
+    result += (byte & 0x7f) * Math.pow(2, shift);
+    if ((byte & 0x80) === 0) return [result, i];
+    shift += 7;
+    if (shift > 63) break; // malformed / oversized
+  }
+  return [result, i];
+}
+
+/**
+ * Decode the top-level fields of a protobuf message. Best-effort: stops on the
+ * first malformed byte rather than throwing, so a partially-corrupt blob still
+ * yields the fields decoded so far.
+ */
+function decodeProtoFields(buf: Buffer): ProtoField[] {
+  const fields: ProtoField[] = [];
+  let i = 0;
+  while (i < buf.length) {
+    const [key, afterKey] = readVarint(buf, i);
+    if (afterKey === i) break;
+    i = afterKey;
+    const field = Math.floor(key / 8);
+    const wireType = key & 7;
+    if (field <= 0) break;
+    if (wireType === 0) {
+      const [value, next] = readVarint(buf, i);
+      if (next === i) break;
+      i = next;
+      fields.push({ field, wireType, varint: value });
+    } else if (wireType === 2) {
+      const [len, afterLen] = readVarint(buf, i);
+      i = afterLen;
+      if (len < 0 || i + len > buf.length) break;
+      fields.push({ field, wireType, bytes: buf.subarray(i, i + len) });
+      i += len;
+    } else if (wireType === 5) {
+      i += 4;
+    } else if (wireType === 1) {
+      i += 8;
+    } else {
+      break; // wireType 3/4 (groups) — unused by antigravity payloads
+    }
+  }
+  return fields;
+}
+
+/** Return the bytes of the first length-delimited field with number `field`. */
+function firstLenField(buf: Buffer, field: number): Buffer | null {
+  for (const f of decodeProtoFields(buf)) {
+    if (f.field === field && f.wireType === 2 && f.bytes) return f.bytes;
+  }
+  return null;
+}
+
+/** Heuristic: is this buffer (mostly) printable UTF-8 text? */
+function looksLikeText(buf: Buffer): boolean {
+  if (buf.length === 0) return false;
+  let printable = 0;
+  for (let i = 0; i < buf.length; i++) {
+    const b = buf[i];
+    // ASCII printable + common whitespace, or any high byte (UTF-8 lead/cont).
+    if ((b >= 32 && b <= 126) || b === 9 || b === 10 || b === 13 || b >= 0x80) printable += 1;
+  }
+  return printable / buf.length >= 0.9;
+}
+
+/**
+ * Antigravity prefixes some assistant answers with a literal `MARKER_V1`
+ * sentinel followed by a blank line. Strip it so the user-visible bubble starts
+ * at the real text.
+ */
+function stripAnswerMarker(text: string): string {
+  return text.replace(/^\s*MARKER_V1\s*/, '');
+}
+
+/**
+ * Extract the assistant's final natural-language answer from a step_type 15
+ * payload: field 20 → field 1 (identical to field 8). Field 20 → field 3 is the
+ * private reasoning summary and is deliberately skipped. Returns '' if absent
+ * (e.g. a pure reasoning / tool-only model step, which carries no user-visible
+ * answer text).
+ */
+function extractModelAnswer(payload: Buffer): string {
+  const inner = firstLenField(payload, 20);
+  if (!inner) return '';
+  const answer = firstLenField(inner, 1) ?? firstLenField(inner, 8);
+  if (!answer || !looksLikeText(answer)) return '';
+  return stripAnswerMarker(answer.toString('utf-8')).trim();
+}
+
+/**
+ * Extract the user prompt from a step_type 14 payload: field 19 → field 2 (the
+ * clean prompt text; field 19 → field 3 wraps the same string with a leading
+ * newline and is used only as a fallback). The USER_REQUEST XML wrapper, when
+ * present, is unwrapped to match the brain-transcript reader's output.
+ */
+function extractUserPrompt(payload: Buffer): string {
+  const inner = firstLenField(payload, 19);
+  if (!inner) return '';
+  const raw = firstLenField(inner, 2) ?? firstLenField(inner, 3);
+  if (!raw || !looksLikeText(raw)) return '';
+  const text = raw.toString('utf-8').trim();
+  if (!text) return '';
+  return extractUserRequestContent(text);
+}
+
+/**
+ * The model step's private reasoning summary lives at field 20 → field 3. We
+ * surface it nowhere, but we DO need it: the schema-drift recovery below scans
+ * the raw payload for a plausible answer run, and the reasoning is itself a long
+ * natural-language run — so we extract it here purely to EXCLUDE it and avoid
+ * accidentally surfacing internal reasoning as the assistant answer.
+ */
+function extractModelReasoning(payload: Buffer): string {
+  const inner = firstLenField(payload, 20);
+  if (!inner) return '';
+  const reasoning = firstLenField(inner, 3);
+  if (!reasoning || !looksLikeText(reasoning)) return '';
+  return reasoning.toString('utf-8').trim();
+}
+
+/** Top-level protobuf field numbers present in a payload (for drift breadcrumbs). */
+function topLevelFieldNumbers(payload: Buffer): number[] {
+  return decodeProtoFields(payload).map((f) => f.field);
+}
+
+const MIN_RECOVERED_MESSAGE_CHARS = 12;
+
+/**
+ * Split a payload into UTF-8 text runs, schema-agnostically. Unlike
+ * extractStringsFromBuffer (ASCII-only, used for legacy .pb), this is UTF-8 aware
+ * so CJK / accented answers survive intact: we decode the whole blob as UTF-8
+ * (invalid byte sequences collapse to U+FFFD) and split on runs of C0/C1 control
+ * chars + the replacement char. Protobuf framing bytes (field tags, varint length
+ * prefixes) are almost always control or invalid-UTF-8, so each natural-language
+ * string field emerges as its own run while binary framing is discarded.
+ */
+function extractUtf8TextRuns(buf: Buffer): string[] {
+  if (buf.length === 0) return [];
+  const decoded = buf.toString('utf-8');
+  // Keep tab/newline/CR (0x09/0x0A/0x0D) inside runs — answers contain newlines.
+  // Everything else in C0 (incl. 0x1A, the field-3 tag that separates reasoning
+  // from the answer), DEL, and the replacement char are run separators.
+  const parts = decoded.split(/[\u0000-\u0008\u000B\u000C\u000E-\u001F\u007F\uFFFD]+/);
+  const runs: string[] = [];
+  for (const part of parts) {
+    const trimmed = part.trim();
+    if (trimmed.length >= MIN_PRINTABLE_RUN) runs.push(trimmed);
+  }
+  return runs;
+}
+
+/**
+ * Is this run plausibly a user-visible prose message, as opposed to the other
+ * text the payload also carries — internal reasoning (excluded separately),
+ * tool-call JSON arguments, code blobs, file paths, and uuid/session-id
+ * metadata? These filters were tuned against real antigravity stores so that a
+ * blind printable-run scan recovers a genuinely drifted answer while surfacing
+ * NONE of the tool-call / metadata runs that legitimately answer-less steps
+ * carry (verified: zero false recoveries across real conversation dbs).
+ */
+function isPlausibleMessageText(s: string): boolean {
+  if (s.length < MIN_RECOVERED_MESSAGE_CHARS) return false;
+  if (!/[A-Za-zÀ-￿]/.test(s)) return false; // must contain letters (incl. CJK)
+  if (/^(file:\/\/|[A-Za-z]:[\\/]|\/[A-Za-z0-9._-]+\/)/.test(s)) return false; // path/URI
+  // Prose is multi-word: real answers have several spaces; uuids / ids / tokens
+  // have none. This is the single strongest prose-vs-metadata discriminator.
+  if ((s.match(/ /g) ?? []).length < 2) return false;
+  // Reject structured tool-call args / JSON / code blobs. A model tool step
+  // carries its arguments as JSON (e.g. {"Query":...}, {"CommandLine":...});
+  // those must never be surfaced as an assistant answer.
+  if (/[[{]\s*"/.test(s)) return false;
+  const structural = (s.match(/[{}[\]":\\]/g) ?? []).length;
+  if (structural / s.length > 0.12) return false;
+  return true;
+}
+
+/**
+ * Schema-agnostic recovery of a message's text when the known field path yields
+ * nothing (a possible antigravity step_payload schema drift). Scans the payload
+ * for UTF-8 text runs and returns the longest plausible message run, EXCLUDING
+ * any run that matches one of `excludeTexts` (e.g. the reasoning subtree) so we
+ * never surface internal reasoning as the answer. Returns '' when nothing beyond
+ * reasoning/metadata is present — i.e. a legitimately answer-less step.
+ */
+function recoverMessageText(payload: Buffer, excludeTexts: string[]): string {
+  const exclusions = excludeTexts.map((t) => t.trim()).filter(Boolean);
+  let best = '';
+  for (const run of extractUtf8TextRuns(payload)) {
+    const candidate = stripAnswerMarker(run).trim();
+    if (!isPlausibleMessageText(candidate)) continue;
+    // Drop runs that are (or are contained in / contain) an excluded subtree.
+    if (exclusions.some((e) => e === candidate || e.includes(candidate) || candidate.includes(e))) {
+      continue;
+    }
+    if (candidate.length > best.length) best = candidate;
+  }
+  return best;
+}
+
+interface AgyDbStepRow {
+  idx: number;
+  step_type: number;
+  step_payload: Buffer | null;
+}
+
+/**
+ * Parse a per-session conversations/<uuid>.db (SQLite) into NativeHistoryMessages.
+ * Returns null when the db is unreadable, empty, or yields no chat messages.
+ */
+/**
+ * True when an error thrown by better-sqlite3 open/read is a transient
+ * SQLITE_BUSY / "database is locked" condition rather than a permanent one.
+ *
+ * On win32 antigravity holds a mandatory WAL write/checkpoint lock while it
+ * persists a step; a readonly open racing that lock throws SQLITE_BUSY. That
+ * is transient — the answer IS already on disk — so it must be retried, NOT
+ * collapsed to "no session" (which erases the just-written assistant answer on
+ * a chat_history re-query). macOS advisory locking + WAL reader-doesn't-block-
+ * writer masks this, which is why it is win32-specific.
+ */
+function isSqliteBusyError(err: unknown): boolean {
+  if (!err) return false;
+  const code = (err as any).code;
+  if (typeof code === 'string' && code.includes('SQLITE_BUSY')) return true;
+  const msg = err instanceof Error ? err.message : String(err);
+  return /SQLITE_BUSY|database is locked|database table is locked/i.test(msg);
+}
+
+const AGY_DB_BUSY_TIMEOUT_MS = 3000;
+const AGY_DB_MAX_ATTEMPTS = 4;
+const AGY_DB_RETRY_BACKOFF_MS = [50, 100, 150];
+
+function sleepBusy(ms: number): void {
+  // Synchronous busy-wait: parseConversationDb is a sync function called from a
+  // sync read path, and better-sqlite3 itself is synchronous. The waits are
+  // tiny (≤150ms) and only occur under genuine lock contention, so a short
+  // spin-sleep is acceptable and keeps the call site synchronous.
+  const end = Date.now() + ms;
+  while (Date.now() < end) { /* spin */ }
+}
+
+function parseConversationDb(
+  filePath: string,
+  sessionId: string,
+  workspace?: string,
+): NativeHistoryMessage[] | null {
+  let Database: any;
+  try {
+    Database = loadBetterSqlite3();
+  } catch (err) {
+    // better-sqlite3 binding genuinely unavailable (ABI mismatch / not built
+    // into this bundle). This is the only true "cannot read at all" case — a
+    // real load failure, distinct from transient lock contention below. Warn
+    // once at WARN so it is greppable; the reader degrades gracefully (returns
+    // null → dispatcher falls back to brain/.pb).
+    LOG.warn(
+      'NativeHistory',
+      `antigravity .db reader could not load better-sqlite3 for ${path.basename(filePath)}: ${err instanceof Error ? err.message : String(err)} (native binding unavailable — assistant answers in this .db will not surface)`,
+    );
+    return null;
+  }
+
+  let rows: AgyDbStepRow[] | null = null;
+  let lastBusyErr: unknown;
+
+  for (let attempt = 1; attempt <= AGY_DB_MAX_ATTEMPTS; attempt++) {
+    let db: any;
+    try {
+      db = new Database(filePath, { readonly: true, fileMustExist: true });
+      // Ask SQLite itself to wait (rather than failing fast) if the WAL
+      // lock is momentarily held by antigravity. Set as early as possible
+      // after open so the prepare/all below inherits the wait.
+      try { db.pragma(`busy_timeout = ${AGY_DB_BUSY_TIMEOUT_MS}`); } catch { /* ignore */ }
+      rows = db
+        .prepare(
+          `SELECT idx, step_type, step_payload
+             FROM steps
+            WHERE step_type IN (${AGY_STEP_TYPE_USER}, ${AGY_STEP_TYPE_MODEL})
+            ORDER BY idx ASC`,
+        )
+        .all() as AgyDbStepRow[];
+      break; // success
+    } catch (err) {
+      if (isSqliteBusyError(err)) {
+        // Transient WAL lock contention. Do NOT collapse to null on the first
+        // failure — the assistant answer is already persisted; treating a busy
+        // lock as "no session" is exactly what erased answers on re-query.
+        // Retry with a small backoff; only give up after attempts exhausted.
+        lastBusyErr = err;
+        if (attempt < AGY_DB_MAX_ATTEMPTS) {
+          sleepBusy(AGY_DB_RETRY_BACKOFF_MS[attempt - 1] ?? 150);
+          continue;
+        }
+        LOG.warn(
+          'NativeHistory',
+          `antigravity .db ${path.basename(filePath)} stayed locked (SQLITE_BUSY) after ${AGY_DB_MAX_ATTEMPTS} attempts: ${err instanceof Error ? err.message : String(err)} (WAL write/checkpoint lock contention — assistant answers may transiently not surface this read)`,
+        );
+        return null;
+      }
+      // `steps` table absent / unexpected schema, or a genuine open/parse
+      // failure that is not lock contention — a real (but recoverable) shape
+      // mismatch. Log at debug so a schema drift in a future antigravity
+      // release is diagnosable without spamming logs for every legacy db.
+      LOG.debug(
+        'NativeHistory',
+        `antigravity .db ${path.basename(filePath)} not readable: ${err instanceof Error ? err.message : String(err)}`,
+      );
+      return null;
+    } finally {
+      try { db?.close(); } catch { /* ignore */ }
+    }
+  }
+
+  void lastBusyErr; // referenced only for retry bookkeeping above
+
+  if (!Array.isArray(rows) || rows.length === 0) return null;
+
+  const normalizedWorkspace = typeof workspace === 'string' ? workspace.trim() : '';
+  const baseTs = statMtimeMs(filePath) || Date.now();
+  const messages: NativeHistoryMessage[] = [];
+
+  for (const row of rows) {
+    const payload = row.step_payload;
+    if (!payload || !Buffer.isBuffer(payload) || payload.length === 0) continue;
+
+    // Steps are ordered by idx; the db carries no per-step timestamp we can
+    // trust as ms, so synthesize a monotonically increasing receivedAt that
+    // preserves order (idx-derived) around the file mtime.
+    const receivedAt = baseTs + messages.length;
+
+    if (row.step_type === AGY_STEP_TYPE_USER) {
+      let content = extractUserPrompt(payload);
+      if (!content) {
+        // Primary field path (field 19 → 2/3) missed. Recover schema-agnostically
+        // rather than silently drop a user turn: scan the payload for the longest
+        // plausible prompt run (metadata/paths/tokens are filtered out). There is
+        // no reasoning subtree to exclude on the user side.
+        const recovered = extractUserRequestContent(recoverMessageText(payload, []));
+        if (recovered) {
+          content = recovered;
+          LOG.debug(
+            'NativeHistory',
+            `antigravity .db ${path.basename(filePath)} step ${row.idx} (type ${row.step_type}): user prompt absent at field 19; recovered ${content.length} chars via printable-run fallback — possible step_payload schema drift`,
+          );
+        } else {
+          LOG.debug(
+            'NativeHistory',
+            `antigravity .db ${path.basename(filePath)} step ${row.idx} (type ${row.step_type}) dropped: no user prompt text (payload ${payload.length}B, top-level fields [${topLevelFieldNumbers(payload).join(',')}])`,
+          );
+          continue;
+        }
+      }
+      const msg: NativeHistoryMessage = {
+        ts: new Date(receivedAt).toISOString(),
+        receivedAt,
+        role: 'user',
+        content,
+        kind: 'standard',
+        agent: 'antigravity-cli',
+        historySessionId: sessionId,
+      };
+      if (normalizedWorkspace) msg.workspace = normalizedWorkspace;
+      messages.push(msg);
+    } else if (row.step_type === AGY_STEP_TYPE_MODEL) {
+      let content = extractModelAnswer(payload);
+      if (!content) {
+        // The known answer path (field 20 → 1/8) yielded nothing. This is either
+        // (a) a legitimate reasoning-only / tool-planning step — the common case,
+        // which carries no user-visible answer — or (b) antigravity moved the
+        // answer to a different field/subtree (schema drift). Attempt a
+        // schema-agnostic recovery that EXCLUDES the reasoning subtree (field
+        // 20 → 3) so internal reasoning is never surfaced as the answer.
+        const reasoning = extractModelReasoning(payload);
+        const recovered = recoverMessageText(payload, reasoning ? [reasoning] : []);
+        if (recovered) {
+          content = recovered;
+          LOG.debug(
+            'NativeHistory',
+            `antigravity .db ${path.basename(filePath)} step ${row.idx} (type ${row.step_type}): answer absent at field 20; recovered ${content.length} chars via printable-run fallback — possible step_payload schema drift`,
+          );
+        } else {
+          // No answer at the primary path and nothing recoverable beyond
+          // reasoning/metadata → drop. Content-free breadcrumb so a genuine
+          // future drift (answer present but unreadable) is greppable, and the
+          // expected reasoning-only case is distinguishable via reasoningOnly.
+          LOG.debug(
+            'NativeHistory',
+            `antigravity .db ${path.basename(filePath)} step ${row.idx} (type ${row.step_type}) dropped: no answer text (payload ${payload.length}B, top-level fields [${topLevelFieldNumbers(payload).join(',')}], reasoningOnly=${reasoning ? 'yes' : 'no'})`,
+          );
+          continue;
+        }
+      }
+      const msg: NativeHistoryMessage = {
+        ts: new Date(receivedAt).toISOString(),
+        receivedAt,
+        role: 'assistant',
+        content,
+        kind: 'standard',
+        agent: 'antigravity-cli',
+        historySessionId: sessionId,
+      };
+      if (normalizedWorkspace) msg.workspace = normalizedWorkspace;
+      messages.push(msg);
+    }
+  }
+
+  return messages.length > 0 ? messages : null;
+}
+
+/**
+ * Assemble the history.jsonl user-prompt index for a single session id into a
+ * partial NativeHistorySession (user prompts only — history.jsonl never carries
+ * assistant answers). Returns null when the index has no rows for this session.
+ * Shared by the history.jsonl read case and the .db sibling fallback.
+ */
+function readHistoryJsonlSession(
+  resolvedSessionId: string,
+  workspace?: string,
+): NativeHistorySession | null {
+  if (!isUuidLike(resolvedSessionId)) return null;
+  const rows = readHistoryRows().filter((r) => r.conversationId === resolvedSessionId);
+  if (rows.length === 0) return null;
+
+  rows.sort((a, b) => a.receivedAt - b.receivedAt);
+  const firstWorkspace = workspace || rows.find((r) => r.workspace)?.workspace || '';
+
+  const messages: NativeHistoryMessage[] = [];
+  if (firstWorkspace) {
+    messages.push({
+      ts: new Date(rows[0].receivedAt).toISOString(),
+      receivedAt: rows[0].receivedAt,
+      role: 'system',
+      content: firstWorkspace,
+      kind: 'session_start',
+      agent: 'antigravity-cli',
+      historySessionId: resolvedSessionId,
+      workspace: firstWorkspace,
+    });
+  }
+
+  for (const row of rows) {
+    const msg: NativeHistoryMessage = {
+      ts: new Date(row.receivedAt).toISOString(),
+      receivedAt: row.receivedAt,
+      role: 'user',
+      content: row.display,
+      kind: 'standard',
+      agent: 'antigravity-cli',
+      historySessionId: resolvedSessionId,
+    };
+    if (row.workspace) msg.workspace = row.workspace;
+    messages.push(msg);
+  }
+
+  return {
+    messages,
+    providerSessionId: resolvedSessionId,
+    source: 'provider-native',
+    sourcePath: historyJsonlPath(),
+    sourceMtimeMs: statMtimeMs(historyJsonlPath()),
+    nativeHistoryCoverage: 'partial',
+    partialReason: 'antigravity_cli_history_jsonl_contains_user_prompts_only',
+  };
+}
+
+/**
+ * ANTIGRAVITY-COMPLETION-DASHBOARD-GAP: recover a bound session's transcript
+ * from the legacy sibling sources when its per-session .db read came back empty
+ * (transient WAL lock, assistant step not yet flushed, or a decode miss on a
+ * future step_payload schema). Tried in descending fidelity for the SAME
+ * session id — never cross-binds to another conversation:
+ *   1. brain transcript (full coverage, authoritative when present),
+ *   2. sibling conversations/<uuid>.pb (best-effort raw text),
+ *   3. history.jsonl (partial — user prompts only).
+ * Returns null when no sibling yields any message.
+ */
+function readAntigravitySiblingFallback(
+  sessionId: string,
+  workspace?: string,
+): NativeHistorySession | null {
+  if (!isUuidLike(sessionId)) return null;
+
+  // (1) brain transcript — full coverage when antigravity actually wrote it.
+  const brainPath = findBrainTranscriptPath(sessionId);
+  if (brainPath && statMtimeMs(brainPath) > 0) {
+    const brainMessages = parseBrainTranscript(brainPath, sessionId, workspace);
+    if (brainMessages && brainMessages.length > 0) {
+      return {
+        messages: brainMessages,
+        providerSessionId: sessionId,
+        source: 'provider-native',
+        sourcePath: brainPath,
+        sourceMtimeMs: statMtimeMs(brainPath),
+        nativeHistoryCoverage: 'full',
+        workspace,
+      };
+    }
+  }
+
+  // (2) sibling .pb — best-effort raw text extraction.
+  const pbPath = resolvePathInside(conversationsRoot(), `${sessionId}.pb`);
+  if (pbPath && fs.existsSync(pbPath)) {
+    const pbMessages = parsePbFile(pbPath, sessionId);
+    if (pbMessages && pbMessages.length > 0) {
+      return {
+        messages: pbMessages,
+        providerSessionId: sessionId,
+        source: 'provider-native',
+        sourcePath: pbPath,
+        sourceMtimeMs: statMtimeMs(pbPath),
+        nativeHistoryCoverage: 'best-effort',
+        partialReason: 'antigravity_cli_pb_raw_text_extraction',
+        workspace,
+      };
+    }
+  }
+
+  // (3) history.jsonl — partial (user prompts only). Last resort so the
+  //     dashboard at least shows the prompt when no answer source is readable.
+  return readHistoryJsonlSession(sessionId, workspace);
+}
+
+// ─── Public API ─────────────────────────────────────────────────────────────
+
+/**
+ * Read a single Antigravity CLI session by its source file path.
+ *
+ * `sessionPath` is the absolute path to one of:
+ *   - A brain transcript JSONL: ~/.gemini/antigravity-cli/brain/<uuid>/.system_generated/logs/transcript*.jsonl
+ *   - The shared history.jsonl: ~/.gemini/antigravity-cli/history.jsonl
+ *   - A conversation SQLite db: ~/.gemini/antigravity-cli/conversations/<uuid>.db
+ *   - A conversation protobuf: ~/.gemini/antigravity-cli/conversations/<uuid>.pb
+ *
+ * The session UUID is inferred from the directory name (brain), filename (pb), or
+ * must be provided explicitly via `sessionId` for history.jsonl paths.
+ *
+ * Returns `null` when the file is missing, empty, or yields no parseable messages.
+ */
+export function readSession(
+  sessionPath: string,
+  sessionId?: string,
+  workspace?: string,
+): NativeHistorySession | null {
+  if (!sessionPath || !path.isAbsolute(sessionPath)) return null;
+  if (!fs.existsSync(sessionPath)) return null;
+
+  const sourceMtimeMs = statMtimeMs(sessionPath);
+
+  // ── Case 1: brain transcript JSONL ──────────────────────────────────────
+  const brainRootPath = brainRoot();
+  if (sessionPath.startsWith(brainRootPath + path.sep) && sessionPath.endsWith('.jsonl')) {
+    // Extract UUID from the path: brain/<uuid>/.system_generated/...
+    const relative = sessionPath.slice(brainRootPath.length + 1);
+    const uuidFromPath = relative.split(path.sep)[0];
+    const resolvedSessionId = sessionId || (isUuidLike(uuidFromPath) ? uuidFromPath : '');
+    if (!resolvedSessionId) return null;
+
+    const messages = parseBrainTranscript(sessionPath, resolvedSessionId, workspace);
+    if (!messages || messages.length === 0) return null;
+
+    return {
+      messages,
+      providerSessionId: resolvedSessionId,
+      source: 'provider-native',
+      sourcePath: sessionPath,
+      sourceMtimeMs,
+      nativeHistoryCoverage: 'full',
+      workspace,
+    };
+  }
+
+  // ── Case 2a: .db conversation file (current per-session SQLite format) ───
+  if (sessionPath.endsWith('.db')) {
+    const dbSessionId = sessionId || path.basename(sessionPath, '.db');
+    if (!isUuidLike(dbSessionId)) return null;
+
+    const messages = parseConversationDb(sessionPath, dbSessionId, workspace);
+    if (messages && messages.length > 0) {
+      return {
+        messages,
+        providerSessionId: dbSessionId,
+        source: 'provider-native',
+        sourcePath: sessionPath,
+        sourceMtimeMs,
+        nativeHistoryCoverage: 'full',
+        workspace,
+      };
+    }
+
+    // ANTIGRAVITY-COMPLETION-DASHBOARD-GAP: the bound .db yielded nothing this
+    // read — transient (WAL lock / assistant step not yet flushed) or a decode
+    // miss on a future schema. The dispatcher binds STRAIGHT to <uuid>.db once
+    // the session id is known and never re-resolves to a sibling source, so a
+    // null here previously collapsed the whole read to native-unavailable even
+    // when the SAME session's assistant answer was recoverable from the legacy
+    // brain transcript / .pb / history.jsonl. Fall back across those sibling
+    // sources for the same session id before giving up. This is read-only,
+    // scoped to the bound session, and cannot cross-bind to another conversation.
+    const siblingFallback = readAntigravitySiblingFallback(dbSessionId, workspace);
+    if (siblingFallback) return siblingFallback;
+
+    return null;
+  }
+
+  // ── Case 2b: .pb conversation file (legacy protobuf) ─────────────────────
+  if (sessionPath.endsWith('.pb')) {
+    const pbSessionId = sessionId || path.basename(sessionPath, '.pb');
+    if (!isUuidLike(pbSessionId)) return null;
+
+    const messages = parsePbFile(sessionPath, pbSessionId);
+    if (!messages || messages.length === 0) return null;
+
+    return {
+      messages,
+      providerSessionId: pbSessionId,
+      source: 'provider-native',
+      sourcePath: sessionPath,
+      sourceMtimeMs,
+      nativeHistoryCoverage: 'best-effort',
+      partialReason: 'antigravity_cli_pb_raw_text_extraction',
+    };
+  }
+
+  // ── Case 3: history.jsonl (user prompts index) ──────────────────────────
+  if (path.basename(sessionPath) === 'history.jsonl') {
+    return readHistoryJsonlSession(sessionId || '', workspace);
+  }
+
+  return null;
+}
+
+/**
+ * List all Antigravity CLI sessions found across all watchPath locations.
+ *
+ * Priority:
+ *   1. Brain transcript sessions (full coverage) — preferred source.
+ *   2. History JSONL entries without a brain transcript (partial coverage).
+ *   3. .pb conversation files without any other source (best-effort).
+ *
+ * Returns summary metadata sorted by most recently updated first.
+ */
+export async function listSessions(_watchPath: string): Promise<NativeHistorySessionMeta[]> {
+  const results: NativeHistorySessionMeta[] = [];
+  const seen = new Set<string>();
+
+  // ── Step 1: brain transcripts (full coverage) ──────────────────────────
+  const historyRows = readHistoryRows();
+  const workspaceBySession = new Map<string, string>();
+  for (const row of historyRows) {
+    if (row.workspace && !workspaceBySession.has(row.conversationId)) {
+      workspaceBySession.set(row.conversationId, row.workspace);
+    }
+  }
+
+  for (const { sessionId, transcriptPath, sourceMtimeMs } of listBrainSessions()) {
+    const workspace = workspaceBySession.get(sessionId);
+    const messages = parseBrainTranscript(transcriptPath, sessionId, workspace);
+    if (!messages || messages.length === 0) continue;
+
+    const visible = messages.filter((m) => m.kind !== 'session_start');
+    if (visible.length === 0) continue;
+
+    const firstMsg = visible[0];
+    const lastMsg = visible[visible.length - 1];
+
+    results.push({
+      historySessionId: sessionId,
+      sessionId,
+      sourcePath: transcriptPath,
+      sourceMtimeMs,
+      messageCount: visible.length,
+      firstMessageAt: firstMsg.receivedAt || sourceMtimeMs,
+      lastMessageAt: lastMsg.receivedAt || sourceMtimeMs,
+      sessionTitle: lastMsg.content,
+      preview: lastMsg.content,
+      workspace,
+      agent: 'antigravity-cli',
+      source: 'provider-native',
+      nativeHistoryCoverage: 'full',
+    });
+    seen.add(sessionId);
+  }
+
+  // ── Step 2: history.jsonl entries without brain transcripts ────────────
+  const grouped = new Map<
+    string,
+    {
+      firstMessageAt: number;
+      lastMessageAt: number;
+      messageCount: number;
+      sessionTitle?: string;
+      preview?: string;
+      workspace?: string;
+    }
+  >();
+  for (const row of historyRows) {
+    const existing = grouped.get(row.conversationId) ?? {
+      firstMessageAt: row.receivedAt,
+      lastMessageAt: row.receivedAt,
+      messageCount: 0,
+    };
+    existing.messageCount += 1;
+    existing.firstMessageAt = Math.min(existing.firstMessageAt, row.receivedAt);
+    if (row.receivedAt >= existing.lastMessageAt) {
+      existing.lastMessageAt = row.receivedAt;
+      existing.sessionTitle = row.display;
+      existing.preview = row.display;
+      if (row.workspace) existing.workspace = row.workspace;
+    }
+    grouped.set(row.conversationId, existing);
+  }
+
+  const historySrc = historyJsonlPath();
+  const historyMtime = statMtimeMs(historySrc);
+
+  for (const [sessionId, info] of grouped.entries()) {
+    if (seen.has(sessionId)) continue;
+    results.push({
+      historySessionId: sessionId,
+      sessionId,
+      sourcePath: historySrc,
+      sourceMtimeMs: historyMtime,
+      messageCount: info.messageCount,
+      firstMessageAt: info.firstMessageAt,
+      lastMessageAt: info.lastMessageAt,
+      sessionTitle: info.sessionTitle,
+      preview: info.preview,
+      workspace: info.workspace,
+      agent: 'antigravity-cli',
+      source: 'provider-native',
+      nativeHistoryCoverage: 'partial',
+      partialReason: 'antigravity_cli_history_jsonl_contains_user_prompts_only',
+    });
+    seen.add(sessionId);
+  }
+
+  // ── Step 3: conversations/<uuid>.db and .pb without any other source ────
+  //
+  // Current antigravity writes per-session SQLite dbs; older sessions kept a
+  // .pb protobuf. Both can coexist in conversations/. Discover both, but when
+  // the same uuid has a .db, prefer it (full coverage) over the .pb (best
+  // effort). brain/history sources already in `seen` still win over either.
+  const convRoot = conversationsRoot();
+  if (fs.existsSync(convRoot)) {
+    let entries: fs.Dirent[] = [];
+    try { entries = fs.readdirSync(convRoot, { withFileTypes: true }); } catch { /* ignore */ }
+
+    // Group by uuid so a .db supersedes a sibling .pb of the same session.
+    const byUuid = new Map<string, { db?: string; pb?: string }>();
+    for (const entry of entries) {
+      if (!entry.isFile()) continue;
+      const dbMatch = /^([0-9a-f-]+)\.db$/i.exec(entry.name);
+      const pbMatch = /^([0-9a-f-]+)\.pb$/i.exec(entry.name);
+      if (dbMatch && isUuidLike(dbMatch[1])) {
+        const g = byUuid.get(dbMatch[1]) ?? {};
+        g.db = path.join(convRoot, entry.name);
+        byUuid.set(dbMatch[1], g);
+      } else if (pbMatch && isUuidLike(pbMatch[1])) {
+        const g = byUuid.get(pbMatch[1]) ?? {};
+        g.pb = path.join(convRoot, entry.name);
+        byUuid.set(pbMatch[1], g);
+      }
+    }
+
+    for (const [uuid, files] of byUuid.entries()) {
+      if (seen.has(uuid)) continue;
+
+      if (files.db) {
+        // Parse the db so listSessions reports accurate counts/preview and the
+        // session surfaces with full coverage (assistant answers included).
+        const workspace = workspaceBySession.get(uuid);
+        const messages = parseConversationDb(files.db, uuid, workspace);
+        const dbMtime = statMtimeMs(files.db);
+        if (messages && messages.length > 0) {
+          const lastMsg = messages[messages.length - 1];
+          const firstMsg = messages[0];
+          results.push({
+            historySessionId: uuid,
+            sessionId: uuid,
+            sourcePath: files.db,
+            sourceMtimeMs: dbMtime,
+            messageCount: messages.length,
+            firstMessageAt: firstMsg.receivedAt || dbMtime,
+            lastMessageAt: lastMsg.receivedAt || dbMtime,
+            sessionTitle: lastMsg.content,
+            preview: lastMsg.content,
+            workspace,
+            agent: 'antigravity-cli',
+            source: 'provider-native',
+            nativeHistoryCoverage: 'full',
+          });
+          seen.add(uuid);
+          continue;
+        }
+        // db unreadable/empty (e.g. sqlite binding unavailable) → fall through
+        // to the .pb best-effort entry below if one exists.
+      }
+
+      if (files.pb) {
+        const pbMtime = statMtimeMs(files.pb);
+        results.push({
+          historySessionId: uuid,
+          sessionId: uuid,
+          sourcePath: files.pb,
+          sourceMtimeMs: pbMtime,
+          messageCount: 0,
+          firstMessageAt: pbMtime,
+          lastMessageAt: pbMtime,
+          agent: 'antigravity-cli',
+          source: 'provider-native',
+          nativeHistoryCoverage: 'best-effort',
+          partialReason: 'antigravity_cli_pb_raw_text_extraction',
+        });
+        seen.add(uuid);
+      }
+    }
+  }
+
+  results.sort((a, b) => (b.lastMessageAt || 0) - (a.lastMessageAt || 0));
+  return results;
+}