@ascenda-one/history-import 0.1.14 → 0.1.15

package/README.md

@@ -30,8 +30,15 @@ command (same pattern as hooks pairing) and this CLI does the reading.
 4. **Metrics only by default.** Prompt/response text, thinking blocks and
    file contents never leave the machine. Content-level ingestion, if it ever
    ships, is a separate explicit opt-in — not this package's default path.
-5. **Aggregate before shipping.** Per-session / per-day events, not one event
-   per bubble — a single machine's stores hold tens of thousands of them.
+5. **Aggregate before shipping, unless a reader counts the rows.** Per-session
+   / per-day events, not one event per bubble — a single machine's stores hold
+   tens of thousands of them. The one deliberate exception is
+   `ai_tool_call_started`: the backend's work-demand rail derives
+   `toolCallCount` by counting rows of that type and reads no `toolCallCount`
+   key off metadata, so a session-level aggregate would ship, store, and be
+   counted by nothing. Per-call events also place the work in the right hour,
+   which a session spanning six of them cannot. Expect an order of magnitude
+   more events than a session-only import, and an `events.jsonl` to match.
 6. **Provenance is data.** Every event carries `historical_direct`,
    `historical_derived` or `historical_unparsed` — never the live
    `ai_work_telemetry` provenance — so no chart can pass history off as
@@ -46,6 +53,7 @@ command (same pattern as hooks pairing) and this CLI does the reading.
 | Staging/snapshot (copy-then-parse, WAL-aware, **torn down by the run that makes it**) | implemented |
 | `archive` (durable content-addressed copy, dedup, verify, restore, prune) | **implemented, verified on a real 4.1 GB store** |
 | **Claude Code extractor** (human-prompt/tool-result split, session folds incl. recursive subagent transcripts, after-hours, compaction, tool failures, context-window peak, human-corrected edits, correction cadence, gap-split active minutes, epoch marker) | **implemented, verified live** |
+| **Tool-call counting, all three stores** (`tool_use` items / `toolFormerData` / `toolInvocationSerialized`, one `ai_tool_call_started` per call) | **implemented; exercised end to end against all three stores on a developer machine** |
 | **Batch shipper** (`POST /v1/tool-events/batch`, salted hashes, stable importKey) | **implemented, verified live** |
 | **Cursor extractor** (composerHeaders + bubble aggregation via SQL-side `json_extract`, prompt text never parsed into the process, subagent-composer folding, epoch marker) | **implemented, verified live** |
 | **VS Code extractor** (Timeline-history Chat-Edit day×workspace aggregation, Copilot chatSessions folding, workspace identity via `workspace.json` longest-prefix match, epoch marker) | **implemented, verified live** |
@@ -119,7 +127,7 @@ ascenda-history-import archive         # the durable copy; --verify / --list / -
 ### Staging is scaffolding; `archive` is the copy
 
 `import` snapshots each store, extracts, and **deletes the snapshot in a
-`finally`** — success or failure — keeping only the ~10 MB `events.jsonl`.
+`finally`** — success or failure — keeping only `events.jsonl`.
 It also sweeps snapshots left by earlier runs. This is not tidiness: nineteen
 runs once left 254 GB on a 926 GB disk and took free space to 279 MB, and the
 first thing to notice was unrelated tooling failing with `ENOSPC`.

package/dist/cli.js

@@ -377,12 +377,155 @@ var require_payload = __commonJS({
   }
 });
 
+// ../packages/tool-contract/out/metricKeys.js
+var require_metricKeys = __commonJS({
+  "../packages/tool-contract/out/metricKeys.js"(exports) {
+    "use strict";
+    Object.defineProperty(exports, "__esModule", { value: true });
+    exports.METRIC_KEYS = void 0;
+    exports.backendMetricKeys = backendMetricKeys;
+    var CONTEXT_WINDOW_CANONICAL_ALIASES = [
+      "contextWindowPeakPct",
+      "context_window_peak_pct",
+      "contextWindowPct",
+      "context_window_pct"
+    ];
+    var CONTEXT_WINDOW_CURSOR_PERCENT_ALIASES = ["contextUsagePercent"];
+    exports.METRIC_KEYS = {
+      // ── Read by a backend reader ────────────────────────────────────────────
+      contextWindowPeakPct: {
+        readBy: ["backend", "handoff"],
+        backendAliases: CONTEXT_WINDOW_CANONICAL_ALIASES,
+        unit: "fraction of the context window (0\u20131; uncapped for >200k contexts)",
+        note: "Claude Code reports a true per-session peak. Cursor reports its composer's last known occupancy under the same key \u2014 the closest its store can answer, and not the same measurement."
+      },
+      contextUsagePercent: {
+        readBy: ["backend", "handoff"],
+        backendAliases: CONTEXT_WINDOW_CURSOR_PERCENT_ALIASES,
+        unit: "percent (0\u2013100)",
+        note: "Cursor's own column name. Superseded by contextWindowPeakPct on the wire; kept because the handoff reads it and imported rows carry it. Unit-explicit on the backend: always divided by 100, never put through the fraction-or-percent heuristic."
+      },
+      promptCount: { readBy: ["backend", "handoff"], backendAliases: ["promptCount", "prompt_count"] },
+      sessionMinutes: { readBy: ["backend"], backendAliases: ["sessionMinutes", "session_minutes"], unit: "minutes" },
+      durationBucket: { readBy: ["backend", "handoff"], backendAliases: ["durationBucket", "duration_bucket"] },
+      afterHoursPrompts: { readBy: ["backend", "handoff"], backendAliases: ["afterHoursPrompts", "after_hours_prompts"] },
+      inputTokens: { readBy: ["backend"], backendAliases: ["inputTokens", "input_tokens"], unit: "tokens" },
+      outputTokens: { readBy: ["backend"], backendAliases: ["outputTokens", "output_tokens"], unit: "tokens" },
+      cacheReadTokens: { readBy: ["backend"], backendAliases: ["cacheReadTokens", "cache_read_tokens"], unit: "tokens" },
+      queuedPrompts: { readBy: ["backend"], backendAliases: ["queuedPrompts", "queued_prompts"] },
+      linesChangedBucket: { readBy: ["backend"], backendAliases: ["linesChangedBucket", "lines_changed_bucket"] },
+      // ── Read by the local handoff only ──────────────────────────────────────
+      activeMinutes: { readBy: ["handoff"], unit: "minutes" },
+      afterHoursRequests: { readBy: ["handoff"] },
+      approximateLintErrorsCount: { readBy: ["handoff"] },
+      canceledCount: { readBy: ["handoff"] },
+      chatEditCount: { readBy: ["handoff"] },
+      compactionCount: {
+        readBy: ["handoff"],
+        note: "The backend counts context_compression_* event rows, not this key. Both are emitted; this one is for the handoff."
+      },
+      contextWindowPeakTokens: {
+        readBy: ["handoff"],
+        unit: "tokens",
+        note: "The measured quantity, with no assumed denominator. Prefer this to the ratio for any within-person baseline."
+      },
+      date: { readBy: ["handoff"] },
+      errorCount: { readBy: ["handoff"] },
+      filesChangedCount: { readBy: ["handoff"] },
+      humanChangesCount: { readBy: ["handoff"] },
+      linesAdded: { readBy: ["handoff"] },
+      linesRemoved: { readBy: ["handoff"] },
+      primaryModel: { readBy: ["handoff"] },
+      requestCount: { readBy: ["handoff"] },
+      sessionStartedAt: { readBy: ["handoff"] },
+      subagentComposers: { readBy: ["handoff"] },
+      subagentToolCallCount: {
+        readBy: ["handoff"],
+        note: "Claude Code only: calls made inside subagent transcripts, kept out of toolCallCount so the main-loop count matches what the wire events carry."
+      },
+      subagentTranscripts: { readBy: ["handoff"] },
+      toolCallCount: {
+        readBy: ["handoff"],
+        note: "The backend counts ai_tool_call_started event rows, not this key \u2014 same split as compactionCount. Deduplicated on the tool-call id; see each extractor for what one call means in its store."
+      },
+      toolCallsUndated: {
+        readBy: ["handoff"],
+        note: "Cursor only: calls whose every record carries an empty createdAt. Counted in toolCallCount, but no event exists for them \u2014 the wire total is smaller than this session count by exactly this number."
+      },
+      toolFailureCount: { readBy: ["handoff"] },
+      totalEntryCount: { readBy: ["handoff"] },
+      userModifiedEditCount: {
+        readBy: ["handoff"],
+        note: "Null, never 0 \u2014 Claude Code never sets userModified true, so 0 would assert 'no AI edit was ever corrected by hand'."
+      },
+      // ── Diagnostic: read by nobody, on purpose ──────────────────────────────
+      abandonedPromptCount: { readBy: ["diagnostic"] },
+      // Epoch-marker metrics. The marker is local-only and never reaches the wire
+      // (see EXTRACTION_EPOCH_KIND), but it travels as a NormalizedHistoricalEvent
+      // and so is keyed by the same vocabulary.
+      windowOldest: { readBy: ["handoff"], note: "Oldest event the extraction saw \u2014 the marker's left edge." },
+      windowNewest: { readBy: ["handoff"], note: "Newest event the extraction saw \u2014 the marker's right edge." },
+      projectsWithNoReadableTranscript: { readBy: ["diagnostic"] },
+      unparsedComposerHeaders: { readBy: ["diagnostic"] },
+      unknownComposerHeaderTypes: { readBy: ["diagnostic"] },
+      orphanedBubbles: { readBy: ["diagnostic"] },
+      orphanedSubagentBubbles: { readBy: ["diagnostic"] },
+      sessionsWithoutTimeline: { readBy: ["diagnostic"] },
+      emptyComposers: { readBy: ["diagnostic"] },
+      apiErrorCount: { readBy: ["diagnostic"] },
+      assistantTurns: { readBy: ["diagnostic"] },
+      compactionAutoCount: { readBy: ["diagnostic"] },
+      compactionManualCount: { readBy: ["diagnostic"] },
+      editDayCount: { readBy: ["diagnostic"] },
+      emptyChatSessions: { readBy: ["diagnostic"] },
+      gitBranch: { readBy: ["diagnostic"] },
+      linesChanged: { readBy: ["diagnostic"] },
+      malformedChatSessionLines: { readBy: ["diagnostic"] },
+      malformedHistoryEntries: { readBy: ["diagnostic"] },
+      mode: { readBy: ["diagnostic"] },
+      modelCount: { readBy: ["diagnostic"] },
+      modelSwitchCount: { readBy: ["diagnostic"] },
+      rapidRepromptCount: { readBy: ["diagnostic"] },
+      schemaUnreadable: { readBy: ["diagnostic"] },
+      sessionCount: { readBy: ["diagnostic"] },
+      sessionsFromBubbleTimeline: { readBy: ["diagnostic"] },
+      sessionsFromCheckpointTimeline: { readBy: ["diagnostic"] },
+      sessionsFromRecencyTimeline: { readBy: ["diagnostic"] },
+      subagentAssistantTurns: { readBy: ["diagnostic"] },
+      subagentPrompts: { readBy: ["diagnostic"] },
+      subagentTokensTotal: { readBy: ["diagnostic"] },
+      toolName: {
+        readBy: ["diagnostic"],
+        note: "Set per ai_tool_call_started event by #43's extractors and shipped in wire metadata, but no reader resolves it server-side yet (the backend's ToolName column is MCP audit, not telemetry). Registered after the fact: #38 and #43 merged past each other, and the union caught it on the next compile \u2014 metaLines all over again."
+      },
+      toolResultCount: { readBy: ["diagnostic"] },
+      toolResultErrorCount: { readBy: ["diagnostic"] },
+      unknownBubbles: { readBy: ["diagnostic"] },
+      unknownLines: { readBy: ["diagnostic"] },
+      metaLines: {
+        readBy: ["diagnostic"],
+        note: 'Recognised-but-skipped transcript machinery (file-history-snapshot, queued-command, \u2026). Split out of unknownLines by #43 so that number keeps meaning "a type nobody has looked at". Registered here after the fact: #41 and #43 merged past each other, and the union caught it on the next compile \u2014 which is this module doing its job.'
+      },
+      unparsedBubbles: { readBy: ["diagnostic"] },
+      unparsedChatSessionFiles: { readBy: ["diagnostic"] },
+      unparsedHistoryFiles: { readBy: ["diagnostic"] },
+      unparsedLines: { readBy: ["diagnostic"] },
+      unreadableChatSessionFiles: { readBy: ["diagnostic"] },
+      unreadableHistoryFiles: { readBy: ["diagnostic"] },
+      unrecognisedChatSessionFiles: { readBy: ["diagnostic"] }
+    };
+    function backendMetricKeys() {
+      return Object.entries(exports.METRIC_KEYS).filter(([, spec]) => spec.readBy.includes("backend")).map(([key, spec]) => [key, spec.backendAliases ?? [key]]);
+    }
+  }
+});
+
 // ../packages/tool-contract/out/index.js
 var require_out = __commonJS({
   "../packages/tool-contract/out/index.js"(exports) {
     "use strict";
     Object.defineProperty(exports, "__esModule", { value: true });
-    exports.ASCENDA_SEMANTIC_PROVENANCE = exports.ASCENDA_HISTORICAL_CONSENT_SCOPE = exports.ASCENDA_COLLABORATION_CONSENT_SCOPE = exports.ASCENDA_SEMANTIC_CONSENT_SCOPE = exports.ASCENDA_PROVENANCE = exports.ASCENDA_CONSENT_SCOPE = exports.EVENT_WORKLOAD_CATEGORY = exports.COLLABORATION_EVENT_TYPES = exports.SEMANTIC_WORK_SIGNAL_EVENT_TYPES = void 0;
+    exports.backendMetricKeys = exports.METRIC_KEYS = exports.ASCENDA_SEMANTIC_PROVENANCE = exports.ASCENDA_HISTORICAL_CONSENT_SCOPE = exports.ASCENDA_COLLABORATION_CONSENT_SCOPE = exports.ASCENDA_SEMANTIC_CONSENT_SCOPE = exports.ASCENDA_PROVENANCE = exports.ASCENDA_CONSENT_SCOPE = exports.EVENT_WORKLOAD_CATEGORY = exports.COLLABORATION_EVENT_TYPES = exports.SEMANTIC_WORK_SIGNAL_EVENT_TYPES = void 0;
     exports.SEMANTIC_WORK_SIGNAL_EVENT_TYPES = [
       "approach_churn_detected",
       "goal_drift_detected",
@@ -441,6 +584,13 @@ var require_out = __commonJS({
     exports.ASCENDA_COLLABORATION_CONSENT_SCOPE = "workflow_telemetry";
     exports.ASCENDA_HISTORICAL_CONSENT_SCOPE = "historical_import";
     exports.ASCENDA_SEMANTIC_PROVENANCE = "semantic_work_signals";
+    var metricKeys_1 = require_metricKeys();
+    Object.defineProperty(exports, "METRIC_KEYS", { enumerable: true, get: function() {
+      return metricKeys_1.METRIC_KEYS;
+    } });
+    Object.defineProperty(exports, "backendMetricKeys", { enumerable: true, get: function() {
+      return metricKeys_1.backendMetricKeys;
+    } });
   }
 });
 
@@ -675,6 +825,7 @@ var require_tokenStore = __commonJS({
       };
     }();
     Object.defineProperty(exports, "__esModule", { value: true });
+    exports.ascendaHome = ascendaHome;
     exports.defaultTokenFilePath = defaultTokenFilePath2;
     exports.persistEventWriteToken = persistEventWriteToken;
     exports.readTokenFile = readTokenFile2;
@@ -682,8 +833,11 @@ var require_tokenStore = __commonJS({
     var fs10 = __importStar(__require("fs"));
     var os3 = __importStar(__require("os"));
     var path10 = __importStar(__require("path"));
+    function ascendaHome() {
+      return process.env.ASCENDA_HOME ?? path10.join(os3.homedir(), ".ascenda");
+    }
     function defaultTokenFilePath2(toolInstallationId) {
-      return path10.join(os3.homedir(), ".ascenda", "tokens", sanitizeFilePart(toolInstallationId));
+      return path10.join(ascendaHome(), "tokens", sanitizeFilePart(toolInstallationId));
     }
     function persistEventWriteToken(tokenFilePath, token) {
       const dir = path10.dirname(tokenFilePath);
@@ -1723,8 +1877,8 @@ var require_out2 = __commonJS({
   "../packages/tool-kit/out/index.js"(exports) {
     "use strict";
     Object.defineProperty(exports, "__esModule", { value: true });
-    exports.workContextRegistryFilePath = exports.readWorkContextRegistry = exports.recordWorkContextAlias = exports.recordWorkContext = exports.deriveWorkContext = exports.hashWithMachineSalt = exports.readOrCreateMachineSalt = exports.machineSaltFilePath = exports.markFailureNotified = exports.shouldAnnounceFailure = exports.recordSendOutcome = exports.readCollectorState = exports.defaultStateFilePath = exports.readTokenFile = exports.persistEventWriteToken = exports.defaultTokenFilePath = exports.recordTurnStart = exports.consumeTurnDurationMs = exports.resolveContextHashes = exports.loadCliAgentConfig = exports.deliverHookEvents = exports.DEFAULT_API_BASE_URL = exports.resolveEventLogPath = exports.expandUserPath = exports.appendEventLog = exports.EVENT_LOG_ENV_VAR = exports.buildEventPayload = exports.AscendaSemanticEventError = exports.AscendaEventSender = exports.looksLikeCorrection = exports.outcomeForHook = exports.inferOutcome = exports.getNestedNumber = exports.getNestedString = exports.getNested = exports.getNumber = exports.getString = exports.localHourAt = exports.utcOffsetMinutesAt = exports.BUSINESS_DAY = exports.isOutsideBusinessHours = exports.isAfterHours = exports.bucketDurationMs = exports.bucketLinesChanged = exports.invitesDebrief = exports.classifyWorkMilestone = exports.isReworkGitAction = exports.classifyGitAction = exports.isVerificationCommand = exports.classifyCommand = void 0;
-    exports.isRetryableStatus = exports.parseIngestResponse = exports.postToolEventsBatch = exports.postToolEvent = exports.renewToolToken = exports.getPairingStatus = exports.createPairingSession = exports.AscendaApiError = exports.liveBusSocketCandidates = exports.liveBusSocketPath = exports.bucketPromptSize = exports.emitLiveSignal = void 0;
+    exports.readWorkContextRegistry = exports.recordWorkContextAlias = exports.recordWorkContext = exports.deriveWorkContext = exports.hashWithMachineSalt = exports.readOrCreateMachineSalt = exports.machineSaltFilePath = exports.markFailureNotified = exports.shouldAnnounceFailure = exports.recordSendOutcome = exports.readCollectorState = exports.defaultStateFilePath = exports.readTokenFile = exports.persistEventWriteToken = exports.defaultTokenFilePath = exports.ascendaHome = exports.recordTurnStart = exports.consumeTurnDurationMs = exports.resolveContextHashes = exports.loadCliAgentConfig = exports.deliverHookEvents = exports.DEFAULT_API_BASE_URL = exports.resolveEventLogPath = exports.expandUserPath = exports.appendEventLog = exports.EVENT_LOG_ENV_VAR = exports.buildEventPayload = exports.AscendaSemanticEventError = exports.AscendaEventSender = exports.looksLikeCorrection = exports.outcomeForHook = exports.inferOutcome = exports.getNestedNumber = exports.getNestedString = exports.getNested = exports.getNumber = exports.getString = exports.localHourAt = exports.utcOffsetMinutesAt = exports.BUSINESS_DAY = exports.isOutsideBusinessHours = exports.isAfterHours = exports.bucketDurationMs = exports.bucketLinesChanged = exports.invitesDebrief = exports.classifyWorkMilestone = exports.isReworkGitAction = exports.classifyGitAction = exports.isVerificationCommand = exports.classifyCommand = void 0;
+    exports.isRetryableStatus = exports.parseIngestResponse = exports.postToolEventsBatch = exports.postToolEvent = exports.renewToolToken = exports.getPairingStatus = exports.createPairingSession = exports.AscendaApiError = exports.liveBusSocketCandidates = exports.liveBusSocketPath = exports.bucketPromptSize = exports.emitLiveSignal = exports.workContextRegistryFilePath = void 0;
     var commandClassifier_1 = require_commandClassifier();
     Object.defineProperty(exports, "classifyCommand", { enumerable: true, get: function() {
       return commandClassifier_1.classifyCommand;
@@ -1838,6 +1992,9 @@ var require_out2 = __commonJS({
       return turnState_1.recordTurnStart;
     } });
     var tokenStore_1 = require_tokenStore();
+    Object.defineProperty(exports, "ascendaHome", { enumerable: true, get: function() {
+      return tokenStore_1.ascendaHome;
+    } });
     Object.defineProperty(exports, "defaultTokenFilePath", { enumerable: true, get: function() {
       return tokenStore_1.defaultTokenFilePath;
     } });
@@ -2368,6 +2525,12 @@ var HISTORICAL_PROVENANCE = {
 var HISTORICAL_CONSENT_SCOPE = import_tool_contract.ASCENDA_HISTORICAL_CONSENT_SCOPE;
 var EXTRACTION_EPOCH_KIND = "extraction_epoch";
 
+// src/toolName.ts
+function sanitizeToolName(toolName) {
+  if (!toolName) return "unknown";
+  return toolName.replace(/[^a-zA-Z0-9_-]/g, "").slice(0, 40) || "unknown";
+}
+
 // src/daySlice.ts
 var LOCAL_TIMEZONE = Intl.DateTimeFormat().resolvedOptions().timeZone ?? null;
 function localDayKey(t) {
@@ -2484,6 +2647,19 @@ function isToolFailureLine(record) {
     (item) => typeof item === "object" && item !== null && item.type === "tool_result" && item.is_error === true
   );
 }
+function toolUseNamesOf(record) {
+  const message = record.message;
+  const content = message?.content;
+  if (!Array.isArray(content)) return [];
+  const names = [];
+  for (const item of content) {
+    if (typeof item !== "object" || item === null) continue;
+    const entry = item;
+    if (entry.type !== "tool_use") continue;
+    names.push(sanitizeToolName(typeof entry.name === "string" ? entry.name : void 0));
+  }
+  return names;
+}
 function countChangedLines(structuredPatch) {
   let count = 0;
   for (const hunk of structuredPatch) {
@@ -2531,10 +2707,13 @@ function newFold(sessionId, projectSlug) {
     userModifiedEditCount: 0,
     linesChangedTotal: 0,
     abandonedPromptCount: 0,
+    toolCallCount: 0,
     subagentTranscripts: 0,
     subagentPrompts: 0,
     subagentAssistantTurns: 0,
     subagentTokensTotal: 0,
+    subagentToolCallCount: 0,
+    toolCalls: [],
     timelinePoints: [],
     humanPromptTimestamps: []
   };
@@ -2638,6 +2817,16 @@ async function foldLinesInto(fold, filePath, opts) {
         case "assistant": {
           const record = JSON.parse(line);
           const usage = record.message?.usage;
+          const toolNames = toolUseNamesOf(record);
+          if (toolNames.length > 0) {
+            if (opts.isSidechain) fold.subagentToolCallCount += toolNames.length;
+            else fold.toolCallCount += toolNames.length;
+            if (sniffed.occurredAt) {
+              for (const name of toolNames) {
+                fold.toolCalls.push({ at: sniffed.occurredAt, name });
+              }
+            }
+          }
           if (opts.isSidechain) {
             fold.subagentAssistantTurns += 1;
             if (usage) {
@@ -2796,11 +2985,25 @@ async function* extractClaudeCode(snapshotDir, extractionId) {
           extractionId
         };
       }
+      for (const call of fold.toolCalls) {
+        yield {
+          occurredAt: call.at,
+          store: "claude_code",
+          sourceVersion: fold.sourceVersion,
+          sessionRef: fold.sessionId,
+          repoRef: fold.cwd ?? fold.projectSlug,
+          eventKind: "ai_tool_call_started",
+          metrics: { toolName: call.name },
+          provenance: HISTORICAL_PROVENANCE.direct,
+          extractionId
+        };
+      }
       const durationMs = sessionDurationMs(fold);
       const toolFailureCount = fold.toolResultErrorCount + fold.apiErrorCount;
       const sessionMetrics = {
         promptCount: fold.humanPrompts,
         assistantTurns: fold.assistantTurns,
+        toolCallCount: fold.toolCallCount,
         toolResultCount: fold.toolResults,
         queuedPrompts: fold.queuedPrompts,
         inputTokens: fold.inputTokens,
@@ -2830,7 +3033,8 @@ async function* extractClaudeCode(snapshotDir, extractionId) {
         subagentTranscripts: fold.subagentTranscripts,
         subagentPrompts: fold.subagentPrompts,
         subagentAssistantTurns: fold.subagentAssistantTurns,
-        subagentTokensTotal: fold.subagentTokensTotal
+        subagentTokensTotal: fold.subagentTokensTotal,
+        subagentToolCallCount: fold.subagentToolCallCount
       };
       if (durationMs !== null) sessionMetrics.sessionMinutes = Math.round(durationMs / 6e4);
       sessionMetrics.sessionStartedAt = fold.firstTs;
@@ -2996,7 +3200,9 @@ SELECT
   json_extract(value,'$.tokenCount.outputTokens') AS outputTokens,
   coalesce(json_array_length(value,'$.humanChanges'),0) AS humanChangesCount,
   coalesce(json_array_length(value,'$.approximateLintErrors'),0) AS lintErrorsCount,
-  json_extract(value,'$.modelInfo.modelName') AS modelName
+  json_extract(value,'$.modelInfo.modelName') AS modelName,
+  json_extract(value,'$.toolFormerData.toolCallId') AS toolCallId,
+  json_extract(value,'$.toolFormerData.name') AS toolName
 FROM cursorDiskKV WHERE key LIKE 'bubbleId:%';
 `;
 function repoInfoOf(row) {
@@ -3057,9 +3263,22 @@ function sniffCursorBubbleRow(row) {
     humanChangesCount: asNumber2(row.humanChangesCount),
     lintErrorsCount: asNumber2(row.lintErrorsCount),
     modelName: row.modelName,
-    sourceVersion: String(row.v)
+    sourceVersion: String(row.v),
+    toolCallId: row.toolCallId,
+    toolName: row.toolName
   };
 }
+function foldToolCall(fold, sniffed) {
+  const id = sniffed.toolCallId;
+  if (!id) return;
+  const at = sniffed.createdAt && sniffed.createdAt.length > 0 ? sniffed.createdAt : null;
+  const known = fold.toolCallFirstSeen.get(id);
+  if (known === void 0) fold.toolCallFirstSeen.set(id, at);
+  else if (at !== null && (known === null || at < known)) fold.toolCallFirstSeen.set(id, at);
+  if (sniffed.toolName && !fold.toolCallNames.has(id)) {
+    fold.toolCallNames.set(id, sanitizeToolName(sniffed.toolName));
+  }
+}
 function newFold2(row, sniffed) {
   return {
     composerId: row.composerId,
@@ -3088,6 +3307,8 @@ function newFold2(row, sniffed) {
     models: /* @__PURE__ */ new Map(),
     bubbleVersions: /* @__PURE__ */ new Map(),
     humanPromptTimestamps: [],
+    toolCallFirstSeen: /* @__PURE__ */ new Map(),
+    toolCallNames: /* @__PURE__ */ new Map(),
     subagentComposerIds: /* @__PURE__ */ new Set(),
     subagentPromptCount: 0,
     subagentAssistantTurns: 0,
@@ -3211,6 +3432,7 @@ async function* extractCursor(snapshotDir, extractionId) {
           ownFold.lastBubbleMs = ms;
         }
       }
+      foldToolCall(ownFold, sniffed);
       if (sniffed.kind === "user") {
         ownFold.humanPrompts += 1;
         if (sniffed.createdAt && (0, import_tool_kit2.isAfterHours)(new Date(sniffed.createdAt))) {
@@ -3235,6 +3457,7 @@ async function* extractCursor(snapshotDir, extractionId) {
     }
     parentFold.subagentComposerIds.add(sniffed.composerId);
     parentFold.subagentTokensTotal += sniffed.inputTokens + sniffed.outputTokens;
+    foldToolCall(parentFold, sniffed);
     if (sniffed.kind === "user") {
       parentFold.subagentPromptCount += 1;
     } else {
@@ -3281,8 +3504,30 @@ async function* extractCursor(snapshotDir, extractionId) {
         extractionId
       };
     }
+    let toolCallsUndated = 0;
+    for (const [id, at] of [...fold.toolCallFirstSeen].sort((a, b) => a[0].localeCompare(b[0]))) {
+      if (at === null) {
+        toolCallsUndated += 1;
+        continue;
+      }
+      yield {
+        occurredAt: at,
+        store: "cursor",
+        sourceVersion,
+        sessionRef: fold.composerId,
+        repoRef: fold.repoRef,
+        eventKind: "ai_tool_call_started",
+        metrics: { toolName: fold.toolCallNames.get(id) ?? "unknown" },
+        provenance: HISTORICAL_PROVENANCE.direct,
+        extractionId
+      };
+    }
     const durationMs = sessionDurationMs2(fold);
     const sessionMetrics = {
+      toolCallCount: fold.toolCallFirstSeen.size,
+      // The gap between the count and the events above, stated rather than
+      // left to be inferred from a shortfall.
+      toolCallsUndated,
       promptCount: fold.humanPrompts,
       assistantTurns: fold.assistantTurns,
       inputTokens: fold.inputTokens,
@@ -3307,6 +3552,7 @@ async function* extractCursor(snapshotDir, extractionId) {
     if (fold.mode) sessionMetrics.mode = fold.mode;
     if (fold.contextUsagePercent !== null) {
       sessionMetrics.contextUsagePercent = Math.round(fold.contextUsagePercent * 100) / 100;
+      sessionMetrics.contextWindowPeakPct = Math.round(fold.contextUsagePercent * 100) / 1e4;
     }
     const primaryModel = pickTop(fold.models);
     if (primaryModel) sessionMetrics.primaryModel = primaryModel;
@@ -3495,6 +3741,22 @@ async function foldEditDays(historyDir, index) {
   }
   return result;
 }
+function toolInvocationsOf(request) {
+  const response = request.response;
+  if (!Array.isArray(response)) return [];
+  const calls = [];
+  for (const part of response) {
+    if (typeof part !== "object" || part === null) continue;
+    const entry = part;
+    if (entry.kind !== "toolInvocationSerialized") continue;
+    if (typeof entry.toolCallId !== "string" || entry.toolCallId.length === 0) continue;
+    calls.push({
+      id: entry.toolCallId,
+      name: sanitizeToolName(typeof entry.toolId === "string" ? entry.toolId : void 0)
+    });
+  }
+  return calls;
+}
 var JSONL_KIND_HEADER = 0;
 var JSONL_KIND_SET = 1;
 var JSONL_KIND_SPLICE = 2;
@@ -3656,7 +3918,8 @@ async function foldChatSessions(workspaceStorageDir) {
         canceledCount: 0,
         errorCount: 0,
         models: /* @__PURE__ */ new Map(),
-        requestTimestamps: []
+        requestTimestamps: [],
+        toolCalls: /* @__PURE__ */ new Map()
       };
       for (const r of requests) {
         if (typeof r !== "object" || r === null) continue;
@@ -3675,6 +3938,9 @@ async function foldChatSessions(workspaceStorageDir) {
         if (typeof req.modelId === "string") {
           fold.models.set(req.modelId, (fold.models.get(req.modelId) ?? 0) + 1);
         }
+        for (const call of toolInvocationsOf(req)) {
+          if (!fold.toolCalls.has(call.id)) fold.toolCalls.set(call.id, { at: iso, name: call.name });
+        }
       }
       if (fold.requestCount > 0) result.sessions.push(fold);
     }
@@ -3728,8 +3994,22 @@ async function* extractVsCode(source, extractionId) {
         extractionId
       };
     }
+    for (const call of fold.toolCalls.values()) {
+      yield {
+        occurredAt: call.at,
+        store: "vscode",
+        sourceVersion: String(KNOWN_CHAT_SESSION_VERSION),
+        sessionRef: fold.sessionId,
+        repoRef: fold.workspaceRoot,
+        eventKind: "ai_tool_call_started",
+        metrics: { toolName: call.name },
+        provenance: HISTORICAL_PROVENANCE.derived,
+        extractionId
+      };
+    }
     const durationMs = Date.parse(fold.lastTs) - Date.parse(fold.firstTs);
     const sessionMetrics = {
+      toolCallCount: fold.toolCalls.size,
       requestCount: fold.requestCount,
       afterHoursRequests: fold.afterHoursRequests,
       canceledCount: fold.canceledCount,
@@ -4155,7 +4435,7 @@ function importKeyOf(event, ordinal) {
     [event.store, event.sessionRef ?? "", event.eventKind, event.occurredAt, String(ordinal)].join(
       "|"
     )
-  ).digest("hex").slice(0, 16);
+  ).digest("hex");
 }
 var workContextMemo = /* @__PURE__ */ new Map();
 function workContextOf(repoRef) {
@@ -4318,7 +4598,7 @@ import * as fs8 from "node:fs/promises";
 import * as os2 from "node:os";
 import * as path8 from "node:path";
 var APP_BUNDLE_ID = "one.ascenda.ascendaMissionControl";
-var HANDOFF_SCHEMA = 2;
+var HANDOFF_SCHEMA = 3;
 function handoffDir(home = os2.homedir()) {
   return path8.join(home, "Library", "Containers", APP_BUNDLE_ID, "Data", ".ascenda", "history-import");
 }
@@ -4366,6 +4646,8 @@ function buildHandoff(events, extractionId, generatedAt) {
       // counting it in case upstream starts populating it, and the handoff
       // reports the count only when it is non-zero.
       userModifiedEditCount: Number(event.metrics.userModifiedEditCount ?? 0) > 0 ? Number(event.metrics.userModifiedEditCount) : null,
+      toolCallCount: Number(event.metrics.toolCallCount ?? 0),
+      subagentToolCallCount: Number(event.metrics.subagentToolCallCount ?? 0),
       subagentTranscripts: Number(event.metrics.subagentTranscripts ?? 0),
       days: event.dayBreakdown ?? [],
       provenance: event.provenance
@@ -4411,6 +4693,8 @@ function buildCursorHandoff(events, extractionId, generatedAt) {
       contextUsagePercent: typeof event.metrics.contextUsagePercent === "number" ? event.metrics.contextUsagePercent : null,
       humanChangesCount: Number(event.metrics.humanChangesCount ?? 0),
       approximateLintErrorsCount: Number(event.metrics.approximateLintErrorsCount ?? 0),
+      toolCallCount: Number(event.metrics.toolCallCount ?? 0),
+      toolCallsUndated: Number(event.metrics.toolCallsUndated ?? 0),
       subagentComposers: Number(event.metrics.subagentComposers ?? 0),
       days: event.dayBreakdown ?? [],
       provenance: event.provenance
@@ -4463,6 +4747,7 @@ function buildVsCodeHandoff(events, extractionId, generatedAt) {
       primaryModel: typeof event.metrics.primaryModel === "string" ? event.metrics.primaryModel : null,
       canceledCount: Number(event.metrics.canceledCount ?? 0),
       errorCount: Number(event.metrics.errorCount ?? 0),
+      toolCallCount: Number(event.metrics.toolCallCount ?? 0),
       days: event.dayBreakdown ?? [],
       provenance: event.provenance
     });
@@ -4524,8 +4809,6 @@ async function safeScan(store, rootPath, scan) {
   }
 }
 var READ_FAILURE_METRICS = [
-  "unparsedFiles",
-  "unreadableFiles",
   "unparsedHistoryFiles",
   "unreadableHistoryFiles",
   "malformedHistoryEntries",
@@ -4534,7 +4817,10 @@ var READ_FAILURE_METRICS = [
   "unrecognisedChatSessionFiles",
   "malformedChatSessionLines",
   "unparsedLines",
-  "unparsedTranscripts"
+  // Projects, not files — the total is a count of things this run could not
+  // read, and the warning wording ("file(s)/record(s)") is looser than the
+  // set it sums.
+  "projectsWithNoReadableTranscript"
 ];
 function readFailuresOf(events) {
   let total = 0;
@@ -4547,6 +4833,9 @@ function readFailuresOf(events) {
   }
   return total;
 }
+function appendAll(target, source) {
+  for (const item of source) target.push(item);
+}
 async function runSource(outcomes, store, inventory, allEvents, extract, buildStoreHandoff, scanError = null) {
   if (scanError !== null) {
     process.stdout.write(`${inventory.store} FAILED to scan: ${scanError}
@@ -4568,7 +4857,7 @@ async function runSource(outcomes, store, inventory, allEvents, extract, buildSt
       kindCounts[event.eventKind] = (kindCounts[event.eventKind] ?? 0) + 1;
     });
   } catch (error) {
-    allEvents.push(...events);
+    appendAll(allEvents, events);
     outcomes.push({
       store,
       status: "failed",
@@ -4581,7 +4870,7 @@ async function runSource(outcomes, store, inventory, allEvents, extract, buildSt
 `);
     return;
   }
-  allEvents.push(...events);
+  appendAll(allEvents, events);
   process.stdout.write(`${store} extracted: ${events.length.toLocaleString("en-US")} events
 `);
   for (const [kind, n] of Object.entries(kindCounts)) {