perchai-cli 2.4.26 → 2.4.29

package/README.md

@@ -19,6 +19,7 @@ Run `perch`, then use these commands inside the terminal chat:
 
 - `/help` — show commands.
 - `/status` — show cwd, auth, mode, persona, permission, and thread.
+- `/usage` — show tokens and estimated cost for this session.
 - `/cwd [dir]` — show or change the working directory.
 - `/permission [default|auto_review|take_the_wheel|plan]` — show or change permission mode.
 - `/permissions [default|auto_review|take_the_wheel|plan]` — alias for `/permission`.

package/dist/perch.mjs

@@ -75566,6 +75566,7 @@ var init_payroll = __esm({
 // lib/perchBusinessTools/index.ts
 var init_perchBusinessTools = __esm({
   "lib/perchBusinessTools/index.ts"() {
+    "use strict";
     init_generateAPAuditPacket();
     init_inventoryFolder();
     init_loadBusinessTables();
@@ -75917,6 +75918,7 @@ function isTurnAbortedError(error) {
 var TURN_STOPPED_BY_USER_MESSAGE;
 var init_turnAbort = __esm({
   "features/perchTerminal/runtime/turnAbort.ts"() {
+    "use strict";
     TURN_STOPPED_BY_USER_MESSAGE = "Turn stopped by user.";
   }
 });
@@ -76921,6 +76923,8 @@ function buildTranscriptSegments(state) {
   const terminalSegmentIdx = /* @__PURE__ */ new Map();
   const workerRunIdx = /* @__PURE__ */ new Map();
   const flockRunIdx = /* @__PURE__ */ new Map();
+  const flockIdByWorkerId = /* @__PURE__ */ new Map();
+  const flockWorkerIdByToolCallId = /* @__PURE__ */ new Map();
   const diagnosticIdxByKey = /* @__PURE__ */ new Map();
   const capabilityIdx = /* @__PURE__ */ new Map();
   const liveCardIdx = /* @__PURE__ */ new Map();
@@ -76944,6 +76948,32 @@ function buildTranscriptSegments(state) {
     if (runId) sandboxRunIdx.set(runId, idx);
     if (toolCallId) sandboxRunIdx.set(toolCallId, idx);
   }
+  function updateFlockWorkerByWorkerId(workerId, updater) {
+    if (!workerId) return false;
+    const flockId = flockIdByWorkerId.get(workerId);
+    if (!flockId) return false;
+    const idx = flockRunIdx.get(flockId);
+    if (idx === void 0) return true;
+    const seg = segments[idx];
+    if (seg?.kind !== "flock_run") return true;
+    segments[idx] = {
+      ...seg,
+      workers: seg.workers.map(
+        (worker) => worker.workerId === workerId ? updater(worker) : worker
+      )
+    };
+    return true;
+  }
+  function isFlockOwnedToolEvent(event) {
+    return Boolean(
+      event.workerId && flockIdByWorkerId.has(event.workerId) || event.toolCallId && flockWorkerIdByToolCallId.has(event.toolCallId)
+    );
+  }
+  function isQuietFlockDiagnostic(event) {
+    if (!event.workerId || !flockIdByWorkerId.has(event.workerId)) return false;
+    if (event.code === "tool_call_budget_exhausted") return false;
+    return event.code === "source_read_budget_reached" || event.code === "source_analysis_max_iteration_closing_attempted" || event.code === "source_analysis_max_iteration_synthesized" || event.code === "source_analysis_max_iteration_closing_fallback" || event.code === "source_analysis_max_iteration_incomplete" || /source-read budget|source-analysis|read-only source/i.test(event.message);
+  }
   function flushReasoning() {
     if (!pendingReasoning) return;
     segments.push({ kind: "reasoning", text: pendingReasoning, seq: seq++, timestamp: pendingReasoningTs });
@@ -77037,6 +77067,15 @@ function buildTranscriptSegments(state) {
           ev.toolCallId ? [ev.toolCallId] : [],
           ev.ts
         );
+        if (ev.workerId && flockIdByWorkerId.has(ev.workerId)) {
+          if (ev.toolCallId) flockWorkerIdByToolCallId.set(ev.toolCallId, ev.workerId);
+          updateFlockWorkerByWorkerId(ev.workerId, (worker) => ({
+            ...worker,
+            toolCalls: (worker.toolCalls ?? 0) + 1,
+            lastToolName: ev.toolName
+          }));
+          break;
+        }
         if (ev.workerId) {
           const idx = segments.length;
           segments.push({
@@ -77091,6 +77130,7 @@ function buildTranscriptSegments(state) {
       }
       case "tool_call_completed": {
         flushReasoning();
+        if (isFlockOwnedToolEvent(ev)) break;
         const id = ev.toolCallId ?? ev.toolName;
         const idx = toolCallIdx.get(id);
         if (idx !== void 0) {
@@ -77107,6 +77147,7 @@ function buildTranscriptSegments(state) {
       }
       case "tool_result": {
         flushReasoning();
+        if (isFlockOwnedToolEvent(ev)) break;
         const id = ev.toolCallId ?? ev.toolName;
         if (ev.toolName === "run_sandbox_code") {
           updateSandboxRun(ev.toolCallId, null, (seg) => ({ ...seg, resultText: ev.text }));
@@ -77135,6 +77176,7 @@ function buildTranscriptSegments(state) {
       case "tool_output_delta": {
         flushReasoning();
         if (!ev.text) break;
+        if (isFlockOwnedToolEvent(ev)) break;
         const id = ev.toolCallId ?? ev.toolName;
         if (ev.toolName === "run_sandbox_code") {
           updateSandboxRun(ev.toolCallId, null, (seg) => ({ ...seg, resultText: `${seg.resultText ?? ""}${ev.text}` }));
@@ -77543,6 +77585,7 @@ function buildTranscriptSegments(state) {
       case "diagnostic": {
         flushReasoning();
         if (!ev.message.trim()) break;
+        if (isQuietFlockDiagnostic(ev)) break;
         if (ev.code === "tool_call_budget_exhausted") {
           let attributed = false;
           if (ev.workerId) {
@@ -77694,6 +77737,7 @@ function buildTranscriptSegments(state) {
       }
       case "tool_call_failed": {
         flushReasoning();
+        if (isFlockOwnedToolEvent(ev)) break;
         const id = ev.toolCallId ?? ev.toolName;
         const idx = toolCallIdx.get(id);
         if (idx !== void 0) {
@@ -77808,6 +77852,12 @@ function buildTranscriptSegments(state) {
       }
       case "worker_run_started": {
         flushReasoning();
+        if (updateFlockWorkerByWorkerId(ev.workerId, (worker) => ({
+          ...worker,
+          status: worker.status === "queued" ? "running" : worker.status
+        }))) {
+          break;
+        }
         const idx = segments.length;
         segments.push({
           kind: "worker_run",
@@ -77824,6 +77874,12 @@ function buildTranscriptSegments(state) {
       }
       case "worker_run_completed": {
         flushReasoning();
+        if (updateFlockWorkerByWorkerId(ev.workerId, (worker) => ({
+          ...worker,
+          status: ev.ok ? "done" : "failed"
+        }))) {
+          break;
+        }
         const runIdx = workerRunIdx.get(ev.workerId);
         if (runIdx !== void 0) {
           const seg = segments[runIdx];
@@ -77876,6 +77932,11 @@ function buildTranscriptSegments(state) {
         if (idx === void 0) break;
         const seg = segments[idx];
         if (seg?.kind !== "flock_run") break;
+        if (ev.accepted) {
+          for (const worker of ev.workers) {
+            flockIdByWorkerId.set(worker.workerId, ev.flockId);
+          }
+        }
         segments[idx] = ev.accepted ? {
           ...seg,
           status: "running",
@@ -77897,7 +77958,11 @@ function buildTranscriptSegments(state) {
         if (seg?.kind !== "flock_run") break;
         const known = seg.workers.some((worker) => worker.flockWorkerId === ev.flockWorkerId);
         const workers = known ? seg.workers.map(
-          (worker) => worker.flockWorkerId === ev.flockWorkerId ? { ...worker, status: ev.status } : worker
+          (worker) => worker.flockWorkerId === ev.flockWorkerId ? {
+            ...worker,
+            status: ev.status,
+            note: ev.detail ?? worker.note
+          } : worker
         ) : [
           ...seg.workers,
           {
@@ -77905,7 +77970,8 @@ function buildTranscriptSegments(state) {
             workerId: ev.workerId,
             displayName: ev.displayName,
             nickname: ev.nickname,
-            status: ev.status
+            status: ev.status,
+            note: ev.detail
           }
         ];
         segments[idx] = { ...seg, workers };
@@ -78308,9 +78374,13 @@ function segmentToCompact(seg) {
         summary: seg.summary?.slice(0, 300),
         flockWorkers: seg.workers.map((worker) => ({
           flockWorkerId: worker.flockWorkerId,
+          workerId: worker.workerId,
           displayName: worker.displayName,
           nickname: worker.nickname,
-          status: worker.status
+          status: worker.status,
+          note: worker.note,
+          toolCalls: worker.toolCalls,
+          lastToolName: worker.lastToolName
         }))
       };
     case "worker_tool_call":
@@ -86622,6 +86692,7 @@ function truncateHistoryLine(value, max2) {
 }
 var init_operatorTruth = __esm({
   "features/perchTerminal/runtime/operatorTruth.ts"() {
+    "use strict";
   }
 });
 
@@ -91587,7 +91658,6 @@ function listFinancialPlaybooks() {
 var AP_AUDIT_PACKET_DEF, PLAYBOOKS;
 var init_registry2 = __esm({
   "features/perchTerminal/runtime/financialPlaybooks/registry.ts"() {
-    "use strict";
     init_managedWorkflowRegistry2();
     init_toolNames();
     AP_AUDIT_PACKET_DEF = {
@@ -220709,11 +220779,11 @@ function flockTaskGate(rawTask) {
   if (!task) {
     return { ok: false, task, reason: "No task given. Usage: /flock <task>." };
   }
-  if (/\/flock\b/i.test(task)) {
+  if (/(^|[\s([{])\/flock(?:\s|$)/i.test(task)) {
     return {
       ok: false,
       task,
-      reason: "A flock cannot gather another flock \u2014 recursive /flock is not allowed."
+      reason: "Subagents cannot start another /flock run from inside a /flock request."
     };
   }
   const wordCount = task.split(/\s+/).length;
@@ -221475,15 +221545,30 @@ function buildSpawnContext(input, flockId, signal, emit, worker) {
   };
 }
 function buildFlockSummary(plan, outcomes, status, toolCallsUsed, wallTimeHit) {
-  const header = status === "completed" ? `Flock finished: ${plan.summary}.` : status === "partial" ? `Flock finished with partial results (${plan.summary}).` : status === "cancelled" ? `Flock stopped early${wallTimeHit ? " \u2014 wall-time cap reached" : ""}.` : `Flock failed \u2014 no worker produced a usable result.`;
+  const header = status === "completed" ? `Subagents finished: ${plan.summary}.` : status === "partial" ? `Subagents finished with partial results (${plan.summary}).` : status === "cancelled" ? `Subagents stopped early${wallTimeHit ? " \u2014 wall-time cap reached" : ""}.` : `Subagents failed \u2014 no worker produced a usable result.`;
   const lines = outcomes.map((outcome) => {
     const mark = outcome.status === "done" ? "done" : outcome.status === "failed" ? "failed" : outcome.status;
-    const detail = outcome.detail ? ` \u2014 ${clampLine(outcome.detail, 160)}` : "";
+    const detailText = friendlyOutcomeDetail(outcome.detail);
+    const detail = detailText ? ` \u2014 ${clampLine(detailText, 160)}` : "";
     return `- ${outcome.worker.displayName} (${outcome.worker.nickname}): ${mark}${detail}`;
   });
   const footer = `Used ${toolCallsUsed}/${plan.caps.maxTotalToolCalls} tool calls across ${outcomes.length} workers.`;
   return [header, ...lines, footer].join("\n");
 }
+function friendlyOutcomeDetail(detail) {
+  const clean = clampLine(detail, 220);
+  if (!clean) return "";
+  if (/Reached maximum \d+ tool-call iterations/i.test(clean)) {
+    return "finished from gathered evidence";
+  }
+  if (/source-read budget|read-only source|source-analysis/i.test(clean)) {
+    return "used gathered evidence";
+  }
+  if (/tool-call budget exhausted|Tool budget reached/i.test(clean)) {
+    return "tool limit reached; finished from evidence";
+  }
+  return clean;
+}
 function clampLine(text, max2) {
   const clean = (text ?? "").replace(/\s+/g, " ").trim();
   return clean.length > max2 ? `${clean.slice(0, max2 - 1)}\u2026` : clean;
@@ -221510,13 +221595,282 @@ var init_runFlockTurn = __esm({
   }
 });
 
+// features/perchTerminal/runtime/usage/usageCommand.ts
+function parseUsageCommand(raw) {
+  const trimmed = raw.trim();
+  if (!/^\/usage(?:\s|$)/i.test(trimmed)) return null;
+  const task = trimmed.slice("/usage".length).trim();
+  return task ? { kind: "usage", task } : { kind: "usage" };
+}
+var init_usageCommand = __esm({
+  "features/perchTerminal/runtime/usage/usageCommand.ts"() {
+    "use strict";
+  }
+});
+
+// lib/perch-ai/access.ts
+function hasPrivilegedPerchAiRole(memberships) {
+  return memberships.some(
+    (membership) => membership.status === "active" && (membership.role === "admin" || membership.role === "internal")
+  );
+}
+function canShowPerchDevTools(membershipRole) {
+  if (!membershipRole) return false;
+  return hasPrivilegedPerchAiRole([
+    {
+      role: membershipRole,
+      status: "active"
+    }
+  ]);
+}
+var init_access = __esm({
+  "lib/perch-ai/access.ts"() {
+  }
+});
+
+// features/perchTerminal/runtime/usage/usageSummary.ts
+function collectUsageRunIds(input) {
+  const runIds = /* @__PURE__ */ new Set();
+  for (const runId of input.usageRunIds ?? []) {
+    const cleaned = runId.trim();
+    if (cleaned) runIds.add(cleaned);
+  }
+  for (const message of input.recentMessages ?? []) {
+    if (message.kind !== "text" || message.role !== "assistant") continue;
+    const stateRunId = message.operatorState?.runId?.trim();
+    if (stateRunId) runIds.add(stateRunId);
+    for (const event of message.operatorState?.events ?? []) {
+      if (event.type === "turn_started" && event.runId.trim()) runIds.add(event.runId.trim());
+    }
+    for (const event of message.transcriptEvents ?? []) {
+      if (event.type === "turn_started" && event.runId.trim()) runIds.add(event.runId.trim());
+    }
+  }
+  return Array.from(runIds).slice(-250);
+}
+function summarizeInferenceUsageRows(rows, options = {}) {
+  const promptTokens = sumRows(rows, "prompt_tokens");
+  const completionTokens = sumRows(rows, "completion_tokens");
+  const cachedTokens = sumRows(rows, "cached_tokens");
+  const totalTokens = sumRows(rows, "total_tokens");
+  const estimatedCostUsd = rows.reduce((sum, row) => sum + readNumber(row.estimated_cost_usd), 0);
+  const created = rows.map((row) => row.created_at).filter((value) => typeof value === "string" && value.length > 0).sort();
+  const summary = {
+    ok: true,
+    calls: rows.length,
+    promptTokens,
+    completionTokens,
+    cachedTokens,
+    totalTokens,
+    estimatedCostUsd,
+    runCount: new Set(options.runIds?.length ? options.runIds : rows.map((row) => row.run_id).filter(Boolean)).size,
+    since: created[0] ?? null,
+    through: created[created.length - 1] ?? null,
+    showModelDetails: Boolean(options.showModelDetails)
+  };
+  if (summary.showModelDetails) {
+    summary.modelBreakdown = buildModelBreakdown(rows);
+  }
+  return summary;
+}
+function formatUsageSummaryText(summary) {
+  const lines = [
+    "Usage this session",
+    `- Calls: ${formatInteger(summary.calls)}`,
+    `- Tokens: ${formatInteger(summary.totalTokens)} total (${formatInteger(summary.promptTokens)} input \xB7 ${formatInteger(summary.completionTokens)} output${summary.cachedTokens ? ` \xB7 ${formatInteger(summary.cachedTokens)} cached` : ""})`,
+    `- Estimated cost: ${formatUsd(summary.estimatedCostUsd)}`
+  ];
+  if (summary.calls === 0) {
+    lines.push("- Nothing has been recorded for this session yet.");
+  }
+  if (summary.showModelDetails && summary.modelBreakdown?.length) {
+    lines.push("", "Internal detail:");
+    for (const row of summary.modelBreakdown.slice(0, 8)) {
+      lines.push(
+        `- ${row.label}: ${formatInteger(row.totalTokens)} tokens \xB7 ${formatUsd(row.estimatedCostUsd)} \xB7 ${formatInteger(row.calls)} calls`
+      );
+    }
+  }
+  return lines.join("\n");
+}
+function buildModelBreakdown(rows) {
+  const byModel = /* @__PURE__ */ new Map();
+  for (const row of rows) {
+    const label = [row.model_option_id, row.provider, row.model].filter((value) => typeof value === "string" && value.trim()).join(" \xB7 ") || "unknown model";
+    const current = byModel.get(label) ?? {
+      label,
+      calls: 0,
+      promptTokens: 0,
+      completionTokens: 0,
+      cachedTokens: 0,
+      totalTokens: 0,
+      estimatedCostUsd: 0
+    };
+    current.calls += 1;
+    current.promptTokens += readNumber(row.prompt_tokens);
+    current.completionTokens += readNumber(row.completion_tokens);
+    current.cachedTokens += readNumber(row.cached_tokens);
+    current.totalTokens += readNumber(row.total_tokens);
+    current.estimatedCostUsd += readNumber(row.estimated_cost_usd);
+    byModel.set(label, current);
+  }
+  return Array.from(byModel.values()).sort((a, b2) => b2.estimatedCostUsd - a.estimatedCostUsd);
+}
+function sumRows(rows, key) {
+  return rows.reduce((sum, row) => sum + readNumber(row[key]), 0);
+}
+function readNumber(value) {
+  if (typeof value === "number" && Number.isFinite(value)) return value;
+  if (typeof value === "string") {
+    const parsed = Number(value);
+    return Number.isFinite(parsed) ? parsed : 0;
+  }
+  return 0;
+}
+function formatInteger(value) {
+  return Math.max(0, Math.round(value)).toLocaleString("en-US");
+}
+function formatUsd(value) {
+  if (!Number.isFinite(value) || value <= 0) return "$0.00";
+  return value < 0.01 ? `$${value.toFixed(4)}` : `$${value.toFixed(2)}`;
+}
+var init_usageSummary = __esm({
+  "features/perchTerminal/runtime/usage/usageSummary.ts"() {
+    "use strict";
+  }
+});
+
+// features/perchTerminal/runtime/usage/runUsageTurn.ts
+async function runUsageTurn(input, deps) {
+  const startedAt = Date.now();
+  const runId = input.clientRunId ?? `usage-${Date.now()}-${Math.random().toString(36).slice(2, 8)}`;
+  const now14 = () => (/* @__PURE__ */ new Date()).toISOString();
+  const events = [];
+  const emit = (event) => {
+    events.push(event);
+    deps.onEvent?.(event);
+  };
+  emit({ type: "turn_started", runId, ts: now14(), isSystemNotification: false });
+  let userMessageId = null;
+  if (input.workspaceId && input.skipUserMessagePersistence !== true) {
+    userMessageId = await deps.persistUserMessage({
+      workspaceId: input.workspaceId,
+      threadId: input.threadId,
+      role: "user",
+      mode: input.chatMode,
+      content: input.trimmedInput
+    });
+    if (userMessageId) {
+      emit({ type: "user_message_persisted", messageId: userMessageId, ts: now14() });
+    }
+  }
+  const runIds = collectUsageRunIds({
+    usageRunIds: input.usageRunIds,
+    recentMessages: input.recentMessages
+  });
+  const showModelDetails = canShowPerchDevTools(input.sessionContext.membershipRole);
+  const summary = await resolveUsageSummary(input, runIds, showModelDetails);
+  const assistantText = formatUsageSummaryText(summary);
+  let assistantMessageId = null;
+  if (input.workspaceId) {
+    assistantMessageId = await deps.persistAssistantMessage({
+      workspaceId: input.workspaceId,
+      threadId: input.threadId,
+      role: "assistant",
+      mode: input.chatMode,
+      content: assistantText,
+      contentJson: {
+        slashCommand: "usage",
+        turnSummary: assistantText,
+        usageSummary: summary,
+        transcriptEvents: events
+      }
+    });
+  }
+  emit({
+    type: "assistant_response",
+    text: assistantText,
+    messageId: assistantMessageId ?? runId,
+    ts: now14()
+  });
+  const durationMs = Date.now() - startedAt;
+  emit({ type: "turn_completed", runId, durationMs, status: "completed", ts: now14() });
+  return {
+    status: "completed",
+    assistantText,
+    messageId: assistantMessageId ?? runId,
+    runId,
+    durationMs
+  };
+}
+async function resolveUsageSummary(input, runIds, showModelDetails) {
+  if (runIds.length === 0) {
+    return summarizeInferenceUsageRows([], { showModelDetails, runIds });
+  }
+  if (input.supabase && input.userId && input.workspaceId) {
+    const rows = await fetchUsageRows(input.supabase, {
+      userId: input.userId,
+      workspaceId: input.workspaceId,
+      runIds
+    });
+    return summarizeInferenceUsageRows(rows, { showModelDetails, runIds });
+  }
+  const viaServer = await fetchUsageSummaryFromServer(input, runIds);
+  if (viaServer) return viaServer;
+  return summarizeInferenceUsageRows([], { showModelDetails, runIds });
+}
+async function fetchUsageRows(supabase, input) {
+  const { data, error } = await supabase.from("perch_ai_inference_usage").select("*").eq("user_id", input.userId).eq("workspace_id", input.workspaceId).in("run_id", input.runIds).order("created_at", { ascending: true }).limit(1e3);
+  if (error) throw error;
+  return data ?? [];
+}
+async function fetchUsageSummaryFromServer(input, runIds) {
+  const appUrl = input.cliServerAppUrl?.trim();
+  const token = input.cliServerAccessToken?.trim();
+  if (!appUrl || !token || typeof fetch !== "function") return null;
+  const controller = new AbortController();
+  const timeout = setTimeout(() => controller.abort(), 5e3);
+  try {
+    const response = await fetch(`${appUrl.replace(/\/+$/, "")}/api/perch-terminal/usage`, {
+      method: "POST",
+      headers: {
+        Accept: "application/json",
+        "Content-Type": "application/json",
+        Authorization: `Bearer ${token}`
+      },
+      body: JSON.stringify({ runIds }),
+      signal: controller.signal
+    });
+    if (!response.ok) return null;
+    const payload = await response.json();
+    return payload.ok ? payload.summary : null;
+  } catch {
+    return null;
+  } finally {
+    clearTimeout(timeout);
+  }
+}
+var init_runUsageTurn = __esm({
+  "features/perchTerminal/runtime/usage/runUsageTurn.ts"() {
+    "use strict";
+    init_access();
+    init_usageSummary();
+  }
+});
+
 // features/perchTerminal/runtime/commands/sharedSlashCommands.ts
 function parseSharedSlashCommand(raw) {
-  return parseFlockCommand(raw);
+  return parseUsageCommand(raw) ?? parseFlockCommand(raw);
+}
+function sharedSlashCommandRunsAsTurn(command) {
+  if (!command) return false;
+  if (command.kind === "usage") return true;
+  return Boolean(command.task);
 }
 function resolveTurnRunnerForPrompt(raw) {
   const parsed = parseSharedSlashCommand(raw);
   if (!parsed) return null;
+  if (parsed.kind === "usage") return runUsageTurn;
   return runFlockTurn;
 }
 var init_sharedSlashCommands = __esm({
@@ -221524,6 +221878,8 @@ var init_sharedSlashCommands = __esm({
     "use strict";
     init_flockCommand();
     init_runFlockTurn();
+    init_usageCommand();
+    init_runUsageTurn();
   }
 });
 
@@ -230123,6 +230479,7 @@ function buildCliTurnInput(input, resolved) {
     userId,
     selectedSourceId: input.selectedSourceId ?? null,
     recentMessages: input.recentMessages ?? [],
+    usageRunIds: input.usageRunIds ?? [],
     sessionContext: emptyPerchTerminalSessionContext({
       userId,
       workspaceId
@@ -230793,6 +231150,10 @@ var init_cliStandaloneOAuth = __esm({
 });
 
 // features/perchTerminal/runtime/contextMeterDisplay.ts
+function positiveInt3(value) {
+  if (typeof value !== "number" || !Number.isFinite(value) || value <= 0) return null;
+  return Math.floor(value);
+}
 function formatTokenK(n) {
   if (n < 1e3) return `${Math.max(0, Math.round(n))}`;
   return `${(n / 1e3).toFixed(n < 1e4 ? 1 : 0)}k`;
@@ -230811,12 +231172,17 @@ function resolvePrimaryContextTokens(snapshot) {
   if (rowTokens > 0) return rowTokens;
   return 0;
 }
-function resolveContextMeterDisplay(snapshot) {
-  const limitTokens = snapshot?.rawContextWindowTokens ?? snapshot?.effectiveLimitTokens ?? snapshot?.contextLimitTokens ?? getEffectiveContextWindow();
+function resolveContextMeterDisplay(snapshot, activeModel) {
+  const activeRawLimitTokens = positiveInt3(activeModel?.contextWindow) ?? positiveInt3(activeModel?.contextWindowTokens);
+  const activeEffectiveLimitTokens = activeRawLimitTokens ? getEffectiveContextWindow({
+    contextWindow: activeRawLimitTokens,
+    maxOutputTokens: activeModel?.maxOutputTokens ?? null
+  }) : null;
+  const limitTokens = activeRawLimitTokens ?? snapshot?.rawContextWindowTokens ?? activeEffectiveLimitTokens ?? snapshot?.effectiveLimitTokens ?? snapshot?.contextLimitTokens ?? getEffectiveContextWindow();
   const threadTokens = snapshot?.threadContextTokens ?? snapshot?.totalContextTokens ?? snapshot?.latestSendTokens ?? 0;
   const primaryTokens = resolvePrimaryContextTokens(snapshot);
   const rawLimitTokens = Math.max(1, limitTokens);
-  const effectiveLimitTokens = snapshot?.effectiveLimitTokens ?? snapshot?.contextLimitTokens ?? rawLimitTokens;
+  const effectiveLimitTokens = activeEffectiveLimitTokens ?? snapshot?.effectiveLimitTokens ?? snapshot?.contextLimitTokens ?? rawLimitTokens;
   const reservedOutputTokens = Math.max(0, rawLimitTokens - effectiveLimitTokens);
   const committedTokens = Math.min(rawLimitTokens, primaryTokens + reservedOutputTokens);
   const workerUsageTokens = snapshot?.workerUsage?.reduce((sum, worker) => sum + Math.max(0, worker.totalTokens), 0) ?? 0;
@@ -230836,6 +231202,8 @@ function resolveContextMeterDisplay(snapshot) {
     workerTokens,
     jobTotalTokens: job,
     limitTokens,
+    effectiveLimitTokens,
+    reservedOutputTokens,
     composerLabel,
     hasWorkers
   };
@@ -285305,7 +285673,8 @@ ${HELP_TEXT}`);
         trimRecentMessages(nextRecentMessages);
         writeCliTurnResult(result3, parsed.json, writer, {
           personaId: parsed.personaId,
-          color: shouldUseCliColor()
+          color: shouldUseCliColor(),
+          activeContextModel: resolveCliActiveContextModel(connection)
         });
         await admitCliLearningMemory(connection, {
           prompt: parsed.prompt,
@@ -285610,7 +285979,7 @@ async function runReadlineInteractivePerchCli(writer, deps, options) {
       if (!prompt) continue;
       if (prompt === "/exit" || prompt === "exit" || prompt === "quit") break;
       const sharedCommand = parseSharedSlashCommand(prompt);
-      const runsAsSharedTurn = Boolean(sharedCommand?.task);
+      const runsAsSharedTurn = sharedSlashCommandRunsAsTurn(sharedCommand);
       if (prompt.startsWith("/") && !runsAsSharedTurn) {
         try {
           const commandResult = await runInteractiveSlashCommand({
@@ -285648,6 +286017,7 @@ async function runReadlineInteractivePerchCli(writer, deps, options) {
         permissionMode: state.permissionMode,
         threadId: state.threadId,
         recentMessages: state.recentMessages,
+        usageRunIds: state.usageRunIds,
         userId: hostedContext.userId,
         workspaceId: hostedContext.workspaceId,
         permanentMemories: hostedContext.permanentMemories,
@@ -285658,13 +286028,15 @@ async function runReadlineInteractivePerchCli(writer, deps, options) {
       });
       writeCliTurnResult(result2, false, writer, {
         personaId: state.personaId,
-        color: shouldUseCliColor()
+        color: shouldUseCliColor(),
+        activeContextModel: resolveCliActiveContextModel(connection)
       });
       appendRecentMessage(state.recentMessages, "user", prompt);
       if (result2.assistantText.trim()) {
         appendRecentMessage(state.recentMessages, "assistant", result2.assistantText.trim());
       }
       trimRecentMessages(state.recentMessages);
+      recordCliUsageRunId(state, result2.runId);
       state.contextSnapshot = result2.contextSnapshot ?? state.contextSnapshot;
       const admitted = await admitCliLearningMemory(connection, {
         prompt,
@@ -285803,7 +286175,7 @@ async function runInkInteractivePerchCli(writer, deps, options) {
             return;
           }
           const sharedTurnCommand = parseSharedSlashCommand(prompt);
-          if (prompt.startsWith("/") && !sharedTurnCommand?.task) {
+          if (prompt.startsWith("/") && !sharedSlashCommandRunsAsTurn(sharedTurnCommand)) {
             const commandWriter = {
               stdout: (text) => {
                 const clean = text.trim();
@@ -285844,6 +286216,8 @@ async function runInkInteractivePerchCli(writer, deps, options) {
           const toolNamesById = /* @__PURE__ */ new Map();
           const flockWorkerNames = /* @__PURE__ */ new Map();
           const flockLimitMeta = /* @__PURE__ */ new Map();
+          const aggregateToolIds = /* @__PURE__ */ new Map();
+          const aggregateToolMeta = /* @__PURE__ */ new Map();
           const clientRunId = createCliRunId();
           const externalController = new AbortController();
           const runtimeRun = registerRuntimeRun({
@@ -285875,6 +286249,7 @@ async function runInkInteractivePerchCli(writer, deps, options) {
               permissionMode: state.permissionMode,
               threadId: state.threadId,
               recentMessages: state.recentMessages,
+              usageRunIds: state.usageRunIds,
               userId: hostedContext.userId,
               workspaceId: hostedContext.workspaceId,
               permanentMemories: hostedContext.permanentMemories,
@@ -285907,11 +286282,28 @@ async function runInkInteractivePerchCli(writer, deps, options) {
                   case "tool_call_started": {
                     const toolId = event.toolCallId ?? `${event.toolName}-${Date.now()}`;
                     const name = humanizeCliToolName(event.toolName);
+                    const aggregateKind = cliToolAggregateKind(event.toolName);
                     toolInputsById.current.set(toolId, {
                       toolName: event.toolName,
                       input: event.input ?? {}
                     });
                     toolNamesById.set(toolId, name);
+                    if (aggregateKind) {
+                      aggregateToolIds.set(toolId, aggregateKind);
+                      const meta2 = getCliToolAggregateMeta(aggregateToolMeta, aggregateKind);
+                      meta2.started += 1;
+                      meta2.pending += 1;
+                      appendCliToolAggregateDetail(meta2, event.toolName, event.input ?? {}, "started");
+                      updateToolItem(cliToolAggregateItemId(aggregateKind), {
+                        label: "tool",
+                        text: renderCliToolAggregateText(aggregateKind, meta2),
+                        tone: "muted",
+                        detailLines: meta2.detailLines,
+                        expanded: false
+                      });
+                      setWorkingText(cliToolAggregateWorkingText(aggregateKind));
+                      break;
+                    }
                     const fileStart = buildFileToolDisplay(event.toolName, event.input ?? {}, "running");
                     if (fileStart) richToolIds.current.add(toolId);
                     updateToolItem(toolId, {
@@ -285927,6 +286319,22 @@ async function runInkInteractivePerchCli(writer, deps, options) {
                   case "tool_call_completed": {
                     const toolId = event.toolCallId ?? `${event.toolName}-${Date.now()}`;
                     const stored = toolInputsById.current.get(toolId);
+                    const aggregateKind = aggregateToolIds.get(toolId);
+                    if (aggregateKind) {
+                      const meta2 = getCliToolAggregateMeta(aggregateToolMeta, aggregateKind);
+                      meta2.pending = Math.max(0, meta2.pending - 1);
+                      meta2.done += 1;
+                      appendCliToolAggregateDetail(meta2, event.toolName, stored?.input ?? {}, "done");
+                      updateToolItem(cliToolAggregateItemId(aggregateKind), {
+                        label: "tool",
+                        text: renderCliToolAggregateText(aggregateKind, meta2),
+                        tone: meta2.failed > 0 ? "danger" : "success",
+                        detailLines: meta2.detailLines,
+                        expanded: false
+                      });
+                      setWorkingText("thinking");
+                      break;
+                    }
                     const fileDone = buildFileToolDisplay(event.toolName, stored?.input ?? {}, "done");
                     if (fileDone) {
                       richToolIds.current.add(toolId);
@@ -285956,6 +286364,22 @@ async function runInkInteractivePerchCli(writer, deps, options) {
                   case "tool_call_failed": {
                     const toolId = event.toolCallId ?? `${event.toolName}-${Date.now()}`;
                     const name = toolNamesById.get(toolId) ?? humanizeCliToolName(event.toolName);
+                    const stored = toolInputsById.current.get(toolId);
+                    const aggregateKind = aggregateToolIds.get(toolId);
+                    if (aggregateKind) {
+                      const meta2 = getCliToolAggregateMeta(aggregateToolMeta, aggregateKind);
+                      meta2.pending = Math.max(0, meta2.pending - 1);
+                      meta2.failed += 1;
+                      appendCliToolAggregateDetail(meta2, event.toolName, stored?.input ?? {}, "failed", event.error);
+                      updateToolItem(cliToolAggregateItemId(aggregateKind), {
+                        label: "tool",
+                        text: renderCliToolAggregateText(aggregateKind, meta2),
+                        tone: "danger",
+                        detailLines: meta2.detailLines,
+                        expanded: false
+                      });
+                      break;
+                    }
                     updateToolItem(toolId, {
                       label: "need",
                       text: `${name} \xB7 ${event.error ?? "needs attention"}`,
@@ -286202,6 +286626,7 @@ async function runInkInteractivePerchCli(writer, deps, options) {
             appendRecentMessage(state.recentMessages, "user", prompt);
             if (assistantText) appendRecentMessage(state.recentMessages, "assistant", assistantText);
             trimRecentMessages(state.recentMessages);
+            recordCliUsageRunId(state, result2.runId);
             state.contextSnapshot = result2.contextSnapshot ?? state.contextSnapshot;
             const admitted = await admitCliLearningMemory(connection, {
               prompt,
@@ -286515,7 +286940,10 @@ async function runInteractiveSlashCommand(input) {
 `);
       return "continue";
     case "context":
-      input.writer.stdout(renderCliContextDetails(input.state.contextSnapshot));
+      input.writer.stdout(renderCliContextDetails(
+        input.state.contextSnapshot,
+        resolveCliActiveContextModel(input.getConnection?.() ?? null)
+      ));
       return "continue";
     case "memoryAudit": {
       const audit = explainLearningAdmission({
@@ -286590,6 +287018,7 @@ async function runInteractiveSlashCommand(input) {
       }
       input.state.threadId = parsed.value === "new" ? `cli-${Date.now()}` : parsed.value;
       input.state.recentMessages = [];
+      input.state.usageRunIds = [];
       input.state.contextSnapshot = null;
       input.state.persistedThreadUpdatedAt = null;
       await hydrateInteractiveCliState(input.state);
@@ -286597,6 +287026,7 @@ async function runInteractiveSlashCommand(input) {
       return "continue";
     case "clear":
       input.state.recentMessages = [];
+      input.state.usageRunIds = [];
       input.state.contextSnapshot = null;
       input.state.persistedThreadUpdatedAt = null;
       clearThreadSession(input.state.threadId);
@@ -286685,6 +287115,7 @@ function renderInteractiveStatus(state, connection, session, workspaceId) {
   const signedIn = session === void 0 ? isCliModelConnectionReady(connection) : isStoredCliAuthSessionUsable(session);
   const connectionStatus = isCliModelConnectionReady(connection) ? "connected" : "locked \xB7 run /login";
   const color = shouldUseCliColor();
+  const activeContextModel = resolveCliActiveContextModel(connection);
   const lines = [
     ["version", CLI_PACKAGE_VERSION],
     ["cwd", state.cwd],
@@ -286697,7 +287128,7 @@ function renderInteractiveStatus(state, connection, session, workspaceId) {
     ["mode", state.chatMode],
     ["persona", state.personaId],
     ["thread", state.threadId],
-    ["context", renderCliContextMeter(state.contextSnapshot)],
+    ["context", renderCliContextMeter(state.contextSnapshot, activeContextModel)],
     ["saved", state.persistedThreadUpdatedAt ? new Date(state.persistedThreadUpdatedAt).toLocaleString() : "(new thread)"],
     ["local-tools", state.cliLocalTools ? "on" : "off"]
   ];
@@ -286846,6 +287277,7 @@ function createInteractiveCliState(options) {
     cliLocalTools: options.cliLocalTools ?? true,
     appUrl: resolveCliAppUrl(options.appUrl ?? null, null),
     recentMessages: [],
+    usageRunIds: [],
     contextSnapshot: null,
     persistedThreadUpdatedAt: null
   };
@@ -286862,6 +287294,7 @@ async function syncInteractiveCliThreadScope(state, connection, options = {}) {
   if (!options.force && state.threadScopeKey === nextScopeKey) return;
   state.threadScopeKey = nextScopeKey;
   state.recentMessages = [];
+  state.usageRunIds = [];
   state.contextSnapshot = null;
   state.persistedThreadUpdatedAt = null;
   clearThreadSession(state.threadId);
@@ -287045,21 +287478,21 @@ function isPermanentMemoryLike2(value) {
   const memory = value;
   return typeof memory.id === "string" && typeof memory.title === "string" && typeof memory.body === "string";
 }
-function renderCliContextMeter(snapshot) {
+function renderCliContextMeter(snapshot, activeModel) {
   if (!snapshot) return "Context";
-  const meter = resolveContextMeterDisplay(snapshot);
+  const meter = resolveContextMeterDisplay(snapshot, activeModel);
   const percent2 = snapshot.contextPercentage ?? Math.round(meter.committedTokens / Math.max(1, meter.limitTokens) * 100);
   const parts = [`${meter.composerLabel}`, `${Math.max(0, Math.min(100, percent2))}%`];
   const compacted = snapshot.compactedRowCount ?? 0;
   if (snapshot.compacted || compacted > 0) parts.push(`compacted ${compacted}`);
   return parts.join(" \xB7 ");
 }
-function renderCliContextDetails(snapshot) {
+function renderCliContextDetails(snapshot, activeModel) {
   if (!snapshot) return "context     no meter yet\n";
-  const meter = resolveContextMeterDisplay(snapshot);
+  const meter = resolveContextMeterDisplay(snapshot, activeModel);
   const percent2 = snapshot.contextPercentage ?? Math.round(meter.committedTokens / Math.max(1, meter.limitTokens) * 100);
   const lines = [
-    ["context", renderCliContextMeter(snapshot)],
+    ["context", renderCliContextMeter(snapshot, activeModel)],
     ["thread", `${formatCliTokenCount(meter.threadTokens)} tokens`],
     ["limit", `${formatCliTokenCount(meter.limitTokens)} tokens`],
     ["fill", `${Math.max(0, Math.min(100, percent2))}%`],
@@ -287079,6 +287512,10 @@ function humanizeCliToolName(name) {
 function isCliModelConnectionReady(connection) {
   return Boolean(connection?.authenticated && connection.founderModelSelection);
 }
+function resolveCliActiveContextModel(connection) {
+  if (!connection?.founderModelSelection) return null;
+  return resolveModelForLane(connection.founderModelSelection, "chat");
+}
 function renderCliAuthSummary(connection) {
   if (isCliModelConnectionReady(connection)) {
     return `signed in${connection.email ? ` as ${connection.email}` : ""}`;
@@ -287415,6 +287852,67 @@ function buildFileToolDisplay(toolName, input, phase, summary) {
     ]
   };
 }
+function cliToolAggregateKind(toolName) {
+  const normalized = toolName.toLowerCase();
+  if (normalized === "readlocalfile" || normalized === "readlocalsourcefile") return "read";
+  if (normalized === "glob" || normalized === "grep" || normalized === "listlocalsources" || normalized === "statpath" || normalized === "validateworkspaceroot") return "search";
+  return null;
+}
+function getCliToolAggregateMeta(store, kind) {
+  const existing = store.get(kind);
+  if (existing) return existing;
+  const created = {
+    started: 0,
+    done: 0,
+    failed: 0,
+    pending: 0,
+    detailLines: []
+  };
+  store.set(kind, created);
+  return created;
+}
+function cliToolAggregateItemId(kind) {
+  return `tool-aggregate-${kind}`;
+}
+function cliToolAggregateWorkingText(kind) {
+  return kind === "read" ? "reading files" : "searching";
+}
+function renderCliToolAggregateText(kind, meta) {
+  const completed = meta.done + meta.failed;
+  const count = Math.max(completed, meta.started);
+  const noun = kind === "read" ? `file${count === 1 ? "" : "s"}` : `path${count === 1 ? "" : "s"}`;
+  const verb = kind === "read" ? "read" : "searched";
+  const status = meta.pending > 0 ? `${meta.pending} running` : "done";
+  const failed = meta.failed > 0 ? ` \xB7 ${meta.failed} failed` : "";
+  return `${verb} ${count} ${noun} \xB7 ${status}${failed} \xB7 ctrl-e`;
+}
+function appendCliToolAggregateDetail(meta, toolName, input, status, error) {
+  const label = humanizeCliToolName(toolName);
+  const target = describeCliToolAggregateInput(toolName, input);
+  const tail = error ? ` \xB7 ${error}` : "";
+  meta.detailLines.push({
+    tone: "meta",
+    text: `${label}${target ? ` \xB7 ${target}` : ""} \xB7 ${status}${tail}`
+  });
+  if (meta.detailLines.length > INK_DETAIL_LINE_LIMIT) {
+    meta.detailLines = meta.detailLines.slice(-INK_DETAIL_LINE_LIMIT);
+  }
+}
+function describeCliToolAggregateInput(toolName, input) {
+  const normalized = toolName.toLowerCase();
+  if (normalized === "glob") {
+    const pattern = stringValue8(input.pattern) ?? stringValue8(input.glob) ?? "*";
+    const base = stringValue8(input.path) ?? stringValue8(input.cwd) ?? ".";
+    return `${pattern} in ${shortFilePath(base)}`;
+  }
+  if (normalized === "grep") {
+    const pattern = stringValue8(input.pattern) ?? stringValue8(input.query) ?? "";
+    const base = stringValue8(input.path) ?? stringValue8(input.cwd) ?? ".";
+    return `${pattern || "pattern"} in ${shortFilePath(base)}`;
+  }
+  const filePath = stringValue8(input.path) ?? stringValue8(input.filePath) ?? stringValue8(input.localSourceId) ?? stringValue8(input.cwd);
+  return filePath ? shortFilePath(filePath) : null;
+}
 function describeChangeSummary(summary) {
   const kind = summary.changeKind ?? "changed";
   const added = summary.linesAdded ?? 0;
@@ -287535,6 +288033,11 @@ function trimRecentMessages(messages) {
     messages.splice(0, messages.length - 30);
   }
 }
+function recordCliUsageRunId(state, runId) {
+  const cleaned = runId?.trim();
+  if (!cleaned) return;
+  state.usageRunIds = [.../* @__PURE__ */ new Set([...state.usageRunIds, cleaned])].slice(-250);
+}
 function createCliRunId() {
   return `cli-turn-${Date.now()}-${Math.random().toString(36).slice(2, 8)}`;
 }
@@ -287657,7 +288160,7 @@ function writeCliTurnResult(result2, json, writer, options = {}) {
     color: options.color,
     streamLabels: true
   });
-  const contextLine = `${paint("context".padEnd(12), "muted", options.color === true)} ${renderCliContextMeter(result2.contextSnapshot)}
+  const contextLine = `${paint("context".padEnd(12), "muted", options.color === true)} ${renderCliContextMeter(result2.contextSnapshot, options.activeContextModel)}
 `;
   writer.stdout(`${rendered}${rendered.endsWith("\n") ? "" : "\n"}${contextLine}`);
 }
@@ -287813,6 +288316,7 @@ var init_perch_cli = __esm({
     init_cliAuthSession();
     init_cliStandaloneOAuth();
     init_contextMeterDisplay();
+    init_modelRegistry();
     init_threadSession();
     init_runRegistry();
     init_learningMemory();
@@ -287875,6 +288379,7 @@ Commands:
   /help                         Show this list.
   /status                       Show cwd, auth, mode, persona, permission, and thread.
   /skills [name]                 List Perch core skills, or show one skill body.
+  /usage                        Show tokens and estimated cost for this session.
   /cwd [dir]                    Show or change the working directory.
   /permission [mode]            Show or set default, auto_review, take_the_wheel, or plan.
   /permissions [mode]           Alias for /permission.
@@ -287934,6 +288439,7 @@ Commands:
     ];
     PERCH_SPLASH_COMMANDS = [
       ["/status", "show auth, route, tools, and thread"],
+      ["/usage", "show session tokens and estimated cost"],
       ["/skills", "show reusable operator skills"],
       ["/persona", "swap saffron or quill"],
       ["/permission", "change autonomy for the next turns"],

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "perchai-cli",
-  "version": "2.4.26",
+  "version": "2.4.29",
   "description": "Perch AI command-line interface",
   "bin": {
     "perch": "bin/perch"