@iowarp/clio-coder 0.3.8 → 0.4.0

package/dist/{chunk-NMPKI6XL.js → chunk-6DB53AJS.js}

@@ -1,12 +1,12 @@
 import { createRequire as __clioCreateRequire } from "node:module"; const require = __clioCreateRequire(import.meta.url);
 import {
   THINKING_LEVELS
-} from "./chunk-4DGYLA73.js";
+} from "./chunk-GXNLGKAB.js";
 import {
   listCatalogModelsForRuntime,
   mergeCapabilities,
   synthesizeCatalogBackedModel
-} from "./chunk-PT7HYKEM.js";
+} from "./chunk-VO2LKSTM.js";
 import {
   init_esm_shims
 } from "./chunk-3R73A4XB.js";
@@ -414,8 +414,8 @@ function localAnthropicCompat() {
   };
 }
 function synthLocalModel(input) {
-  const { target, wireModelId, kb, defaultCapabilities: defaultCapabilities25, apiFamily, provider } = input;
-  const caps = mergeCapabilities(defaultCapabilities25, kb?.entry.capabilities ?? null, null, target.capabilities ?? null);
+  const { target, wireModelId, kb, defaultCapabilities: defaultCapabilities26, apiFamily, provider } = input;
+  const caps = mergeCapabilities(defaultCapabilities26, kb?.entry.capabilities ?? null, null, target.capabilities ?? null);
   const rawUrl = target.url ?? "";
   const baseUrl = rawUrl.length > 0 ? input.baseUrlForTarget(rawUrl) : "";
   const pricing = target.pricing;
@@ -865,6 +865,9 @@ async function probeLlamaCppModelStatus(base, target, ctx) {
   if (flags.reasoning === true || flags.reasoningBudget !== void 0) caps.reasoning = true;
   if (flags.mmproj) caps.vision = true;
   if (flags.jinja === true) caps.tools = true;
+  if (flags.parallel !== void 0 && Number.isInteger(flags.parallel) && flags.parallel > 0) {
+    caps.parallelSlots = flags.parallel;
+  }
   const enrichment = { modelId: selected.id, serverFlags: flags };
   if (Object.keys(caps).length > 0) enrichment.discoveredCapabilities = caps;
   const notes = statusNotes(selected.id, selected.status);
@@ -885,11 +888,16 @@ async function detectModelMismatch(base, target, ctx) {
   const loaded = ids[0] ?? "(unknown)";
   return `wire model id ${expected} does not match server's loaded model ${loaded}; llama.cpp serves a single fixed model`;
 }
-async function probeLlamaCppProps(base, ctx) {
-  const opts = { url: `${base}/props`, timeoutMs: ctx.httpTimeoutMs };
-  const result = await (ctx.signal ? probeJson({ ...opts, signal: ctx.signal }) : probeJson(opts));
-  if (!result.ok || !result.data) return {};
-  const data = result.data;
+async function probeLlamaCppProps(base, ctx, modelId) {
+  const request = async (url2) => {
+    const opts = { url: url2, timeoutMs: ctx.httpTimeoutMs };
+    const result = await (ctx.signal ? probeJson({ ...opts, signal: ctx.signal }) : probeJson(opts));
+    return result.ok && result.data ? result.data : null;
+  };
+  const router = await request(`${base}/props`);
+  if (router === null) return {};
+  const selected = typeof router.total_slots === "number" || !modelId ? null : await request(`${base}/props?model=${encodeURIComponent(modelId)}`);
+  const data = selected ?? router;
   const enrichment = {};
   const caps = {};
   const nCtx = data.default_generation_settings?.n_ctx;
@@ -898,9 +906,12 @@ async function probeLlamaCppProps(base, ctx) {
   if (typeof nPredict === "number" && nPredict > 0) caps.maxTokens = nPredict;
   const vision = data.modalities?.vision;
   if (typeof vision === "boolean") caps.vision = vision;
+  const totalSlots = data.total_slots;
+  if (typeof totalSlots === "number" && Number.isInteger(totalSlots) && totalSlots > 0) caps.parallelSlots = totalSlots;
   if (Object.keys(caps).length > 0) enrichment.discoveredCapabilities = caps;
-  if (typeof data.build_info === "string" && data.build_info.length > 0) {
-    enrichment.serverVersion = data.build_info;
+  const buildInfo = data.build_info ?? router.build_info;
+  if (typeof buildInfo === "string" && buildInfo.length > 0) {
+    enrichment.serverVersion = buildInfo;
   }
   return enrichment;
 }
@@ -1785,8 +1796,8 @@ var llamacppRuntime = {
     const healthOpts = { url: `${base}/health`, timeoutMs: ctx.httpTimeoutMs };
     const health = await (ctx.signal ? probeHttp({ ...healthOpts, signal: ctx.signal }) : probeHttp(healthOpts));
     if (!health.ok) return health;
-    const props = await probeLlamaCppProps(base, ctx);
     const status = await probeLlamaCppModelStatus(base, target, ctx);
+    const props = await probeLlamaCppProps(base, ctx, status.modelId ?? target.defaultModel);
     const catalog = await probeOpenAIModelCatalog(base, ctx);
     const result = { ok: true };
     if (catalog.models.length > 0) result.models = catalog.models;
@@ -1795,7 +1806,8 @@ var llamacppRuntime = {
     if (typeof health.latencyMs === "number") result.latencyMs = health.latencyMs;
     const discoveredCapabilities = {
       ...props.discoveredCapabilities ?? {},
-      ...status.discoveredCapabilities ?? {}
+      ...status.discoveredCapabilities ?? {},
+      ...props.discoveredCapabilities?.parallelSlots !== void 0 ? { parallelSlots: props.discoveredCapabilities.parallelSlots } : {}
     };
     if (Object.keys(discoveredCapabilities).length > 0) {
       result.discoveredCapabilities = discoveredCapabilities;
@@ -1866,7 +1878,7 @@ var llamacppAnthropicRuntime = {
     };
     const head = await (ctx.signal ? probeHttp({ ...headOpts, signal: ctx.signal }) : probeHttp(headOpts));
     if (!head.ok) return head;
-    const props = await probeLlamaCppProps(base, ctx);
+    const props = await probeLlamaCppProps(base, ctx, target.defaultModel);
     const enriched = { ...head };
     if (props.discoveredCapabilities) enriched.discoveredCapabilities = props.discoveredCapabilities;
     if (props.serverVersion) enriched.serverVersion = props.serverVersion;
@@ -2018,7 +2030,7 @@ var llamacppCompletionRuntime = {
     const healthOpts = { url: `${base}/health`, timeoutMs: ctx.httpTimeoutMs };
     const health = await (ctx.signal ? probeHttp({ ...healthOpts, signal: ctx.signal }) : probeHttp(healthOpts));
     if (!health.ok) return health;
-    const props = await probeLlamaCppProps(base, ctx);
+    const props = await probeLlamaCppProps(base, ctx, target.defaultModel);
     const status = await probeLlamaCppModelStatus(base, target, ctx);
     const enriched = { ...health };
     const discoveredCapabilities = {
@@ -2140,7 +2152,7 @@ var llamacppEmbedRuntime = {
     if (!probeResponse.ok) {
       return { ok: false, error: `/embedding not available: HTTP ${probeResponse.status}` };
     }
-    const props = await probeLlamaCppProps(base, ctx);
+    const props = await probeLlamaCppProps(base, ctx, target.defaultModel);
     const result = { ok: true };
     if (health.latencyMs !== void 0) result.latencyMs = health.latencyMs;
     if (props.discoveredCapabilities) result.discoveredCapabilities = props.discoveredCapabilities;
@@ -2257,7 +2269,7 @@ var llamacppRerankRuntime = {
     if (!(probeResponse.status === 200 || probeResponse.status === 202)) {
       return { ok: false, error: `/reranking not available: HTTP ${probeResponse.status}` };
     }
-    const props = await probeLlamaCppProps(base, ctx);
+    const props = await probeLlamaCppProps(base, ctx, target.defaultModel);
     const result = { ok: true };
     if (health.latencyMs !== void 0) result.latencyMs = health.latencyMs;
     if (props.discoveredCapabilities) result.discoveredCapabilities = props.discoveredCapabilities;
@@ -2470,15 +2482,15 @@ function foldLmStudioModels(models) {
   return [...byKey.values()];
 }
 function rememberHostCatalog(target, models) {
-  const rootUrl = lmStudioRootUrl(target.url ?? "");
-  if (!rootUrl) return;
-  catalogsByHost.set(rootUrl, { targetId: target.id, rootUrl, models: foldLmStudioModels(models) });
+  const rootUrl2 = lmStudioRootUrl(target.url ?? "");
+  if (!rootUrl2) return;
+  catalogsByHost.set(rootUrl2, { targetId: target.id, rootUrl: rootUrl2, models: foldLmStudioModels(models) });
 }
 function peerTargetsFor(target, instanceId) {
-  const rootUrl = lmStudioRootUrl(target.url ?? "");
+  const rootUrl2 = lmStudioRootUrl(target.url ?? "");
   const peers = [];
   for (const catalog of catalogsByHost.values()) {
-    if (catalog.rootUrl === rootUrl) continue;
+    if (catalog.rootUrl === rootUrl2) continue;
     if (catalog.models.some((model) => model.loadedInstances.some((instance) => instance.id === instanceId))) {
       peers.push(catalog.targetId);
     }
@@ -2694,6 +2706,8 @@ function capabilities(model, instance) {
     out.reasoning = model.reasoningOptions.some((option) => option !== "off");
   const contextWindow = loadedContextLength(instance) ?? model.maxContextLength;
   if (contextWindow !== void 0) out.contextWindow = contextWindow;
+  const parallel = instance?.config.parallel;
+  if (typeof parallel === "number" && Number.isInteger(parallel) && parallel > 0) out.parallelSlots = parallel;
   return out;
 }
 function statusFor(model, instance, reasoningLevels) {
@@ -2745,6 +2759,7 @@ function probeFromCatalog(catalog, target) {
   const selected = resolveModel(models, configuredModel(target));
   const result = {
     ok: true,
+    discoveredCapabilities: { parallelSlots: 1 },
     models: ids,
     modelCapabilities,
     modelStates,
@@ -2757,7 +2772,7 @@ function probeFromCatalog(catalog, target) {
   };
   if (catalog.latencyMs !== void 0) result.latencyMs = catalog.latencyMs;
   if (selected) {
-    result.discoveredCapabilities = capabilities(selected.model, selected.instance);
+    result.discoveredCapabilities = { parallelSlots: 1, ...capabilities(selected.model, selected.instance) };
     result.capabilityModelId = configuredModel(target) ?? selected.model.key;
   }
   const loaded = models.flatMap(
@@ -2838,6 +2853,10 @@ var defaultCapabilities22 = {
 function positiveNumber3(value) {
   return typeof value === "number" && Number.isFinite(value) && value > 0 ? value : void 0;
 }
+function ollamaParallelSlots() {
+  const parsed = Number(process.env.OLLAMA_NUM_PARALLEL);
+  return Number.isInteger(parsed) && parsed > 0 ? parsed : 1;
+}
 async function probeResidentModelStates(base, ctx) {
   const opts = { url: `${base}/api/ps`, timeoutMs: ctx.httpTimeoutMs };
   const result = await (ctx.signal ? probeJson({ ...opts, signal: ctx.signal }) : probeJson(opts));
@@ -2874,7 +2893,7 @@ var ollamaNativeRuntime = {
       if (result.latencyMs !== void 0) failed.latencyMs = result.latencyMs;
       return failed;
     }
-    const out = { ok: true };
+    const out = { ok: true, discoveredCapabilities: { parallelSlots: ollamaParallelSlots() } };
     if (result.latencyMs !== void 0) out.latencyMs = result.latencyMs;
     const modelStates = await probeResidentModelStates(base, ctx);
     if (modelStates) out.modelStates = modelStates;
@@ -2951,6 +2970,164 @@ var vllm_default = makeOpenAICompatRuntime({
   healthPath: "/health"
 });
 
+// src/domains/providers/runtimes/protocol/litellm.ts
+init_esm_shims();
+var LITELLM_PROVIDER = "litellm";
+var defaultCapabilities25 = {
+  chat: true,
+  tools: false,
+  reasoning: false,
+  vision: false,
+  audio: false,
+  embeddings: false,
+  rerank: false,
+  fim: false,
+  structuredOutputs: "json-schema",
+  contextWindow: CLIO_MIN_CONTEXT_WINDOW,
+  maxTokens: CLIO_MIN_MAX_OUTPUT_TOKENS
+};
+function rootUrl(target) {
+  if (!target.url) return null;
+  const trimmed = stripTrailingSlash(target.url);
+  return trimmed.endsWith("/v1") ? trimmed.slice(0, -"/v1".length) : trimmed;
+}
+function bearerHeaders(target, ctx) {
+  const headers = { ...target.auth?.headers ?? {} };
+  const envName = target.auth?.apiKeyEnvVar;
+  const fromEnv = envName && ctx.credentialsPresent.has(envName) ? process.env[envName]?.trim() : void 0;
+  const token = ctx.authToken?.trim() || fromEnv;
+  if (token) headers.authorization = `Bearer ${token}`;
+  return headers;
+}
+function positiveInteger(value) {
+  return typeof value === "number" && Number.isFinite(value) && value > 0 ? Math.floor(value) : void 0;
+}
+function boolean(value) {
+  return typeof value === "boolean" ? value : void 0;
+}
+function capabilitiesFromLiteLLMModelInfo(info) {
+  const caps = {};
+  const contextWindow = positiveInteger(info.max_input_tokens) ?? positiveInteger(info.max_tokens);
+  if (contextWindow !== void 0) caps.contextWindow = contextWindow;
+  const maxTokens = positiveInteger(info.max_output_tokens);
+  if (maxTokens !== void 0) caps.maxTokens = maxTokens;
+  const tools = boolean(info.supports_function_calling);
+  if (tools !== void 0) caps.tools = tools;
+  const vision = boolean(info.supports_vision);
+  if (vision !== void 0) caps.vision = vision;
+  const reasoning = boolean(info.supports_reasoning);
+  if (reasoning !== void 0) caps.reasoning = reasoning;
+  const audio = boolean(info.supports_audio_input);
+  if (audio !== void 0) caps.audio = audio;
+  if (info.mode === "embedding") {
+    caps.chat = false;
+    caps.embeddings = true;
+  }
+  if (info.mode === "rerank") {
+    caps.chat = false;
+    caps.rerank = true;
+  }
+  if (info.mode === "chat") caps.chat = true;
+  if (boolean(info.supports_response_schema) === true) caps.structuredOutputs = "json-schema";
+  return caps;
+}
+var EMPTY_CATALOG = { models: [], modelCapabilities: {}, modelStates: {}, deployments: {} };
+async function fetchCatalog(base, ctx, headers) {
+  const detailOpts = { url: `${base}/v1/model/info`, timeoutMs: ctx.httpTimeoutMs, headers };
+  const detail = await (ctx.signal ? probeJson({ ...detailOpts, signal: ctx.signal }) : probeJson(detailOpts));
+  if (detail.ok && Array.isArray(detail.data?.data)) {
+    const catalog = { models: [], modelCapabilities: {}, modelStates: {}, deployments: {} };
+    for (const row of detail.data.data) {
+      if (typeof row?.model_name !== "string" || row.model_name.length === 0) continue;
+      const alias = row.model_name;
+      if (catalog.deployments[alias] !== void 0) continue;
+      catalog.models.push(alias);
+      const caps = row.model_info ? capabilitiesFromLiteLLMModelInfo(row.model_info) : {};
+      if (Object.keys(caps).length > 0) catalog.modelCapabilities[alias] = caps;
+      const upstream = row.litellm_params?.model;
+      const apiBase = row.litellm_params?.api_base;
+      catalog.deployments[alias] = {
+        model: typeof upstream === "string" ? upstream : alias,
+        ...typeof apiBase === "string" ? { apiBase } : {}
+      };
+    }
+    if (catalog.models.length > 0) return catalog;
+  }
+  const listOpts = { url: `${base}/v1/models`, timeoutMs: ctx.httpTimeoutMs, headers };
+  const list = await (ctx.signal ? probeJson({ ...listOpts, signal: ctx.signal }) : probeJson(listOpts));
+  if (!list.ok || !Array.isArray(list.data?.data)) return EMPTY_CATALOG;
+  const models = list.data.data.map((row) => typeof row?.id === "string" ? row.id : null).filter((id) => id !== null && id.length > 0);
+  return { models, modelCapabilities: {}, modelStates: {}, deployments: {} };
+}
+var litellmRuntime = {
+  id: "litellm",
+  displayName: "LiteLLM gateway",
+  kind: "http",
+  tier: "protocol",
+  apiFamily: "openai-completions",
+  auth: "api-key",
+  defaultCapabilities: defaultCapabilities25,
+  async probe(target, ctx) {
+    const base = rootUrl(target);
+    if (!base) return { ok: false, error: "target has no url" };
+    const headers = bearerHeaders(target, ctx);
+    const liveOpts = { url: `${base}/health/liveliness`, timeoutMs: ctx.httpTimeoutMs };
+    const live = await (ctx.signal ? probeJson({ ...liveOpts, signal: ctx.signal }) : probeJson(liveOpts));
+    if (!live.ok) return { ok: false, error: live.error ?? "gateway is not reachable" };
+    const catalog = await fetchCatalog(base, ctx, headers);
+    if (catalog.models.length === 0) {
+      const result2 = {
+        ok: false,
+        error: "gateway is live but served no model catalog; check the target's API key"
+      };
+      if (live.latencyMs !== void 0) result2.latencyMs = live.latencyMs;
+      return result2;
+    }
+    const result = { ok: true, models: catalog.models };
+    if (live.latencyMs !== void 0) result.latencyMs = live.latencyMs;
+    if (Object.keys(catalog.modelCapabilities).length > 0) {
+      result.modelCapabilities = catalog.modelCapabilities;
+      const selected = target.defaultModel?.trim();
+      const selectedCaps = selected ? catalog.modelCapabilities[selected] : void 0;
+      if (selected && selectedCaps) {
+        result.discoveredCapabilities = selectedCaps;
+        result.capabilityModelId = selected;
+      }
+    }
+    const notes = [];
+    const configured = target.defaultModel?.trim();
+    if (configured && !catalog.models.includes(configured)) {
+      notes.push(`configured model '${configured}' is not in the gateway catalog`);
+    }
+    const routed = Object.entries(catalog.deployments).map(([alias, deployment]) => `${alias}=${deployment.model}`).sort();
+    if (routed.length > 0) notes.push(`routes: ${routed.join(", ")}`);
+    if (notes.length > 0) result.notes = notes;
+    return result;
+  },
+  async probeModels(target, ctx) {
+    const base = rootUrl(target);
+    if (!base) return [];
+    return (await fetchCatalog(base, ctx, bearerHeaders(target, ctx))).models;
+  },
+  synthesizeModel(target, wireModelId, kb) {
+    return synthLocalModel({
+      // A LiteLLM target is a gateway whether or not settings said so. The
+      // residency layer reads this to stay observe-only: the models behind an
+      // alias are the proxy's to load and evict, and a Clio stream that tried
+      // to manage them would be issuing load commands to a host that does not
+      // take them.
+      target: { ...target, gateway: true },
+      wireModelId,
+      kb,
+      defaultCapabilities: defaultCapabilities25,
+      apiFamily: "openai-completions",
+      provider: LITELLM_PROVIDER,
+      baseUrlForTarget: withV1
+    });
+  }
+};
+var litellm_default = litellmRuntime;
+
 // src/domains/providers/runtimes/builtins.ts
 var BUILTIN_RUNTIMES = [
   alcf_default,
@@ -2974,6 +3151,7 @@ var BUILTIN_RUNTIMES = [
   lmstudio_default,
   ollama_native_default,
   anthropic_compat_default,
+  litellm_default,
   openai_compat_default,
   sglang_default,
   vllm_default,
@@ -3003,4 +3181,4 @@ export {
   BUILTIN_RUNTIMES,
   registerBuiltinRuntimes
 };
-//# sourceMappingURL=chunk-NMPKI6XL.js.map
+//# sourceMappingURL=chunk-6DB53AJS.js.map

package/dist/{chunk-3BINW3FP.js → chunk-76ONBSIA.js}

@@ -9,7 +9,7 @@ import {
 import {
   clioStateDir,
   stateRootRemoved
-} from "./chunk-BNAZZHFG.js";
+} from "./chunk-BYP5D4HI.js";
 import {
   fsyncDirectory,
   safeResourceWrite
@@ -567,4 +567,4 @@ export {
   openSession,
   resumeSession
 };
-//# sourceMappingURL=chunk-3BINW3FP.js.map
+//# sourceMappingURL=chunk-76ONBSIA.js.map

package/dist/{chunk-26LEYJZH.js → chunk-7MCTRUCE.js}

@@ -5,7 +5,7 @@ import {
   settingsPath,
   updateSavedSettingsDocument,
   validateSettings
-} from "./chunk-IJ7RPIYJ.js";
+} from "./chunk-VFA6GDY5.js";
 import {
   require_dist
 } from "./chunk-APJ265NV.js";
@@ -226,4 +226,4 @@ export {
   updateLayeredSettings,
   settingsSourceFor
 };
-//# sourceMappingURL=chunk-26LEYJZH.js.map
+//# sourceMappingURL=chunk-7MCTRUCE.js.map

package/dist/{chunk-TYPGUK6W.js → chunk-AB44T6BB.js}

@@ -1,12 +1,14 @@
 import { createRequire as __clioCreateRequire } from "node:module"; const require = __clioCreateRequire(import.meta.url);
 import {
   inspectEvidence
-} from "./chunk-TLQJPP24.js";
+} from "./chunk-BUMFYQFY.js";
 import {
   TASK_MEMORY_CONTENT_MAX_CHARS,
   TASK_MEMORY_DEFAULT_KNOWLEDGE_CAP,
-  TASK_MEMORY_DEFAULT_PROCEDURAL_CAP
-} from "./chunk-ZVJ5BLO2.js";
+  TASK_MEMORY_DEFAULT_PROCEDURAL_CAP,
+  TASK_MEMORY_TELEMETRY_FILE,
+  parseTaskMemoryTelemetryRecord
+} from "./chunk-BKFJHQCA.js";
 import {
   createRedactionTally,
   redactSecretsText
@@ -16,7 +18,10 @@ import {
 } from "./chunk-ECH6PKUQ.js";
 import {
   ceilChars
-} from "./chunk-RWSI4YD7.js";
+} from "./chunk-UM7N4G5A.js";
+import {
+  clioStateDir
+} from "./chunk-BYP5D4HI.js";
 import {
   safeResourceWrite
 } from "./chunk-WEPFGWHJ.js";
@@ -1116,6 +1121,30 @@ function compareStrings3(a, b) {
   return a.localeCompare(b);
 }
 
+// src/domains/memory/task-bank-promotion.ts
+init_esm_shims();
+async function proposeInjectedTaskMemory(dataDir, input) {
+  const result = { records: [], errors: [] };
+  const sessionId = input.sessionId;
+  const repository = canonicalMemoryRepositoryIdentity(input.cwd);
+  if (sessionId === null || sessionId.length === 0 || repository === null) return result;
+  for (const entry of input.entries) {
+    if (entry.kind === "status") continue;
+    try {
+      const proposal = await proposeMemoryPromotion(
+        dataDir,
+        { kind: "task-bank-entry", sessionId, evidenceRefs: [`session:${sessionId}`], entry },
+        { scope: "repo", repository },
+        input.now ?? /* @__PURE__ */ new Date()
+      );
+      result.records.push(proposal.record);
+    } catch (error) {
+      result.errors.push(`${entry.id}: ${error instanceof Error ? error.message : String(error)}`);
+    }
+  }
+  return result;
+}
+
 // src/domains/memory/task-memory-handoff.ts
 init_esm_shims();
 import { existsSync, readdirSync, readFileSync as readFileSync2, statSync } from "node:fs";
@@ -1367,6 +1396,80 @@ function compareStrings4(left, right) {
   return left < right ? -1 : left > right ? 1 : 0;
 }
 
+// src/domains/memory/task-memory-spend.ts
+init_esm_shims();
+import { readFileSync as readFileSync3, statSync as statSync2 } from "node:fs";
+import { join as join3 } from "node:path";
+function taskMemoryStepsPath(stateDir = clioStateDir()) {
+  return join3(stateDir, "memory", TASK_MEMORY_TELEMETRY_FILE);
+}
+function emptyTaskMemorySpendSummary() {
+  return {
+    llmSteps: 0,
+    injections: 0,
+    hitRate: 0,
+    inputTokens: 0,
+    outputTokens: 0,
+    totalTokens: 0,
+    modelMs: 0,
+    slowestStepMs: 0,
+    timeouts: 0,
+    endpointBusySkips: 0,
+    firstAt: null,
+    lastAt: null
+  };
+}
+function foldTaskMemorySpend(lines) {
+  const summary = emptyTaskMemorySpendSummary();
+  for (const line of lines) {
+    if (line.trim().length === 0) continue;
+    let parsed;
+    try {
+      parsed = JSON.parse(line);
+    } catch {
+      continue;
+    }
+    const record = parseTaskMemoryTelemetryRecord(parsed);
+    if (record === null) continue;
+    if (summary.firstAt === null || record.at < summary.firstAt) summary.firstAt = record.at;
+    if (summary.lastAt === null || record.at > summary.lastAt) summary.lastAt = record.at;
+    if (record.reason === "endpoint_busy") summary.endpointBusySkips += 1;
+    if (record.tier !== "llm") continue;
+    summary.llmSteps += 1;
+    if (record.decision === "injected") summary.injections += 1;
+    if (record.decision === "timeout") summary.timeouts += 1;
+    summary.inputTokens += record.tokenCost.input;
+    summary.outputTokens += record.tokenCost.output;
+    summary.totalTokens += record.tokenCost.total;
+    summary.modelMs += record.latencyMs;
+    summary.slowestStepMs = Math.max(summary.slowestStepMs, record.latencyMs);
+  }
+  summary.hitRate = summary.llmSteps === 0 ? 0 : summary.injections / summary.llmSteps;
+  return summary;
+}
+var spendCache = /* @__PURE__ */ new Map();
+function readTaskMemorySpendSummary(stateDir = clioStateDir()) {
+  const path = taskMemoryStepsPath(stateDir);
+  try {
+    const info = statSync2(path);
+    const cached = spendCache.get(path);
+    if (cached !== void 0 && cached.mtimeMs === info.mtimeMs && cached.size === info.size) return cached.summary;
+    const summary = foldTaskMemorySpend(readFileSync3(path, "utf8").split("\n"));
+    spendCache.set(path, { mtimeMs: info.mtimeMs, size: info.size, summary });
+    return summary;
+  } catch {
+    return emptyTaskMemorySpendSummary();
+  }
+}
+function formatTaskMemorySpend(summary) {
+  if (summary.llmSteps === 0) return "";
+  const tokens = summary.totalTokens >= 1e3 ? `${(summary.totalTokens / 1e3).toFixed(1)}k tok` : `${summary.totalTokens} tok`;
+  const seconds = summary.modelMs / 1e3;
+  const time = seconds >= 60 ? `${(seconds / 60).toFixed(1)}m` : `${seconds.toFixed(1)}s`;
+  const rate = `${Math.round(summary.hitRate * 100)}%`;
+  return `spend ${summary.llmSteps} steps \xB7 ${tokens} \xB7 ${time} \xB7 hit ${summary.injections}/${summary.llmSteps} (${rate})`;
+}
+
 // src/domains/memory/task-memory-status.ts
 init_esm_shims();
 function describeTaskMemoryActivity(event) {
@@ -1441,13 +1544,16 @@ export {
   proposeMemoryPromotion,
   buildMemoryPromptSection,
   proposeMemoryFromEvidence,
+  proposeInjectedTaskMemory,
   renderTaskMemoryHandoffSource,
   parseTaskMemoryHandoffSnapshot,
   taskMemoryHandoffSeedOffer,
   seedTaskMemoryFromNewestHandoff,
+  readTaskMemorySpendSummary,
+  formatTaskMemorySpend,
   describeTaskMemoryActivity,
   taskMemoryBankSize,
   taskMemoryTracePath,
   createTaskMemoryTrace
 };
-//# sourceMappingURL=chunk-TYPGUK6W.js.map
+//# sourceMappingURL=chunk-AB44T6BB.js.map