@iowarp/clio-coder 0.3.9 → 0.4.0

package/dist/{chunk-MIX5N5AC.js → chunk-DQITNCXG.js}

@@ -5,24 +5,24 @@ import {
 } from "./chunk-CFGTUFWB.js";
 import {
   credentialsPresent
-} from "./chunk-LU7P4LHA.js";
+} from "./chunk-P3FOHJT4.js";
 import {
   ceilChars,
   estimateAgentMessageTokens,
   toolSchemaChars
-} from "./chunk-AEYBF3TB.js";
-import {
-  readSettings
-} from "./chunk-PNY46YEY.js";
-import {
-  sleep
-} from "./chunk-FQ4SKYE4.js";
+} from "./chunk-UM7N4G5A.js";
 import {
   getSharedBus
 } from "./chunk-ZGVHUX3M.js";
 import {
   BusChannels
-} from "./chunk-RKRLDWD3.js";
+} from "./chunk-HLW2MRKE.js";
+import {
+  readSettings
+} from "./chunk-VFA6GDY5.js";
+import {
+  sleep
+} from "./chunk-FQ4SKYE4.js";
 import {
   CLIO_CONTEXT_WINDOW_WARN_BELOW,
   CLIO_MIN_CONTEXT_WINDOW,
@@ -34,17 +34,17 @@ import {
   registerBuiltinRuntimes,
   requestLmStudioJson,
   resolveLmStudioInstance
-} from "./chunk-JEQQR47K.js";
+} from "./chunk-6DB53AJS.js";
 import {
   listKnownModelsForRuntime
-} from "./chunk-774ILSRL.js";
+} from "./chunk-HI63TFOG.js";
 import {
   authNotRequiredStatus,
   openAuthStorage,
   registerClioOAuthProviders,
   resolveAuthTarget,
   targetRequiresAuth
-} from "./chunk-EQ63NRB7.js";
+} from "./chunk-BBVYXMFO.js";
 import {
   withStateFileLock
 } from "./chunk-IKCO5N3L.js";
@@ -52,10 +52,16 @@ import {
   require_dist
 } from "./chunk-APJ265NV.js";
 import {
+  clioCacheDir,
   clioConfigDir,
+  clioStateDir,
   clioStatePath,
-  resolveClioDirs
-} from "./chunk-BNAZZHFG.js";
+  resolveClioDirs,
+  stateRootRemoved
+} from "./chunk-BYP5D4HI.js";
+import {
+  safeResourceWrite
+} from "./chunk-WEPFGWHJ.js";
 import {
   getRuntimeRegistry
 } from "./chunk-NUGM5KR6.js";
@@ -102,6 +108,88 @@ function canonicalEndpointUrl(raw) {
   }
 }
 
+// src/domains/providers/endpoint-slots-store.ts
+init_esm_shims();
+import { existsSync, readFileSync } from "node:fs";
+import { join } from "node:path";
+var DEFAULT_ENDPOINT_SLOTS_TTL_MS = 24 * 60 * 60 * 1e3;
+var ENDPOINT_SLOTS_TTL_ENV_VAR = "CLIO_CODER_ENDPOINT_SLOTS_TTL_MS";
+var MAX_ENDPOINT_SLOT_RECORDS = 64;
+function endpointSlotsPath() {
+  return join(clioStateDir(), "endpoint-slots.json");
+}
+function endpointSlotsTtlMs(env = process.env) {
+  const raw = env[ENDPOINT_SLOTS_TTL_ENV_VAR]?.trim();
+  if (raw !== void 0 && /^[1-9]\d*$/.test(raw)) {
+    const parsed = Number(raw);
+    if (Number.isSafeInteger(parsed)) return parsed;
+  }
+  return DEFAULT_ENDPOINT_SLOTS_TTL_MS;
+}
+function validRecord(key, value) {
+  if (typeof value !== "object" || value === null || Array.isArray(value)) return false;
+  const record = value;
+  return record.endpointKey === key && typeof record.runtimeId === "string" && record.runtimeId.length > 0 && typeof record.slots === "number" && Number.isInteger(record.slots) && record.slots > 0 && typeof record.observedAt === "string" && Number.isFinite(Date.parse(record.observedAt));
+}
+function readRecords() {
+  const path = endpointSlotsPath();
+  if (!existsSync(path)) return [];
+  try {
+    const parsed = JSON.parse(readFileSync(path, "utf8"));
+    if (parsed?.version !== 1 || typeof parsed.endpoints !== "object" || parsed.endpoints === null) return [];
+    return Object.entries(parsed.endpoints).filter(([key, value]) => validRecord(key, value)).map(([, value]) => ({ ...value }));
+  } catch {
+    return [];
+  }
+}
+function fresh(record, nowMs, ttlMs) {
+  const observedMs = Date.parse(record.observedAt);
+  return observedMs <= nowMs && nowMs - observedMs < ttlMs;
+}
+function readDiscoveredEndpointSlots(options) {
+  const nowMs = options?.nowMs ?? Date.now();
+  const ttlMs = options?.ttlMs ?? endpointSlotsTtlMs();
+  const out = {};
+  for (const record of readRecords()) {
+    if (!fresh(record, nowMs, ttlMs)) continue;
+    out[record.endpointKey] = record;
+  }
+  return Object.freeze(out);
+}
+function writeRecords(records) {
+  const kept = [...records].sort((a, b) => Date.parse(b.observedAt) - Date.parse(a.observedAt)).slice(0, MAX_ENDPOINT_SLOT_RECORDS);
+  const file = {
+    version: 1,
+    endpoints: Object.fromEntries(kept.map((record) => [record.endpointKey, { ...record }]))
+  };
+  safeResourceWrite(endpointSlotsPath(), `${JSON.stringify(file, null, 2)}
+`, { encoding: "utf8" });
+}
+async function recordDiscoveredEndpointSlots(entry, options) {
+  if (!Number.isInteger(entry.slots) || entry.slots <= 0) return;
+  if (entry.endpointKey.length === 0 || entry.runtimeId.length === 0) return;
+  if (stateRootRemoved()) return;
+  const nowMs = options?.nowMs ?? Date.now();
+  const ttlMs = options?.ttlMs ?? endpointSlotsTtlMs();
+  try {
+    await withStateFileLock(endpointSlotsPath(), () => {
+      if (stateRootRemoved()) return;
+      const kept = readRecords().filter(
+        (record) => record.endpointKey !== entry.endpointKey && fresh(record, nowMs, ttlMs)
+      );
+      kept.push({
+        endpointKey: entry.endpointKey,
+        runtimeId: entry.runtimeId,
+        slots: entry.slots,
+        observedAt: new Date(nowMs).toISOString()
+      });
+      writeRecords(kept);
+    });
+  } catch (error) {
+    options?.onError?.(error);
+  }
+}
+
 // src/domains/providers/endpoint-capacity.ts
 var UNBOUNDED_LOCAL_NATIVE_RUNTIMES = /* @__PURE__ */ new Set(["vllm", "sglang"]);
 function canonicalEndpointKey(value) {
@@ -118,31 +206,80 @@ function endpointLabel(endpointKey) {
 function positiveInteger(value) {
   return typeof value === "number" && Number.isInteger(value) && value > 0 ? value : void 0;
 }
-function endpointCapacityForStatus(status) {
-  const key = canonicalEndpointKey(status.target);
+function endpointCapacityFor(input, priors = readDiscoveredEndpointSlots()) {
+  const key = canonicalEndpointKey(input.target);
   if (key === null) return null;
-  const override = positiveInteger(status.target.maxConcurrentRequests);
-  if (override !== void 0) return { key, label: endpointLabel(key), limit: override, source: "override" };
-  const discovered = positiveInteger(status.probeCapabilities?.parallelSlots);
-  if (discovered !== void 0) return { key, label: endpointLabel(key), limit: discovered, source: "discovered" };
-  if (status.runtime?.tier !== "local-native" || UNBOUNDED_LOCAL_NATIVE_RUNTIMES.has(status.runtime.id)) return null;
-  return { key, label: endpointLabel(key), limit: 1, source: "local-native-default" };
+  const label = endpointLabel(key);
+  const override = positiveInteger(input.target.maxConcurrentRequests);
+  if (override !== void 0) return { key, label, limit: override, source: "override" };
+  const discovered = positiveInteger(input.discoveredSlots);
+  if (discovered !== void 0) return { key, label, limit: discovered, source: "discovered" };
+  const prior = priors[key];
+  if (prior !== void 0 && prior.runtimeId === (input.runtime?.id ?? input.target.runtime)) {
+    return { key, label, limit: prior.slots, source: "discovered-prior" };
+  }
+  if (input.runtime?.tier !== "local-native" || UNBOUNDED_LOCAL_NATIVE_RUNTIMES.has(input.runtime.id)) return null;
+  return { key, label, limit: 1, source: "local-native-default" };
+}
+function endpointCapacityForStatus(status, priors = readDiscoveredEndpointSlots()) {
+  return endpointCapacityFor(
+    {
+      target: status.target,
+      runtime: status.runtime ?? null,
+      discoveredSlots: status.probeCapabilities?.parallelSlots
+    },
+    priors
+  );
 }
 function sourceRank(source) {
-  return source === "override" ? 2 : source === "discovered" ? 1 : 0;
+  switch (source) {
+    case "override":
+      return 3;
+    case "discovered":
+      return 2;
+    case "discovered-prior":
+      return 1;
+    default:
+      return 0;
+  }
+}
+function foldCapacity(capacities, candidate) {
+  if (candidate === null) return;
+  const current = capacities[candidate.key];
+  if (current === void 0 || sourceRank(candidate.source) > sourceRank(current.source) || sourceRank(candidate.source) === sourceRank(current.source) && candidate.limit < current.limit) {
+    capacities[candidate.key] = candidate;
+  }
 }
 function endpointCapacitiesForStatuses(statuses) {
+  const priors = readDiscoveredEndpointSlots();
   const capacities = {};
-  for (const status of statuses) {
-    const candidate = endpointCapacityForStatus(status);
-    if (candidate === null) continue;
-    const current = capacities[candidate.key];
-    if (current === void 0 || sourceRank(candidate.source) > sourceRank(current.source) || sourceRank(candidate.source) === sourceRank(current.source) && candidate.limit < current.limit) {
-      capacities[candidate.key] = candidate;
-    }
+  for (const status of statuses) foldCapacity(capacities, endpointCapacityForStatus(status, priors));
+  return capacities;
+}
+function resolveEndpointCapacities(input) {
+  const priors = readDiscoveredEndpointSlots();
+  const capacities = {};
+  const covered = /* @__PURE__ */ new Set();
+  for (const status of input.statuses) {
+    covered.add(status.target.id);
+    foldCapacity(capacities, endpointCapacityForStatus(status, priors));
+  }
+  for (const target of input.targets ?? []) {
+    if (covered.has(target.id)) continue;
+    foldCapacity(
+      capacities,
+      endpointCapacityFor({ target, runtime: input.runtimeFor?.(target.runtime) ?? null }, priors)
+    );
   }
   return capacities;
 }
+function recordEndpointSlotsFromStatus(status, options) {
+  const key = canonicalEndpointKey(status.target);
+  const slots = positiveInteger(status.probeCapabilities?.parallelSlots);
+  const runtimeId = status.runtime?.id ?? status.target.runtime;
+  if (key === null || slots === void 0 || runtimeId === void 0) return Promise.resolve();
+  return recordDiscoveredEndpointSlots({ endpointKey: key, runtimeId, slots }, options);
+}
 var foregroundStreams = /* @__PURE__ */ new Map();
 function registerForegroundStream(endpointKey) {
   foregroundStreams.set(endpointKey, (foregroundStreams.get(endpointKey) ?? 0) + 1);
@@ -164,11 +301,11 @@ init_esm_shims();
 
 // src/engine/apis/residency-lock.ts
 init_esm_shims();
-import { join } from "node:path";
+import { join as join2 } from "node:path";
 var LOCK_WAIT_MS = 12e4;
 function lockTargetFor(targetKey) {
   const key = canonicalEndpointKey(targetKey) ?? targetKey;
-  return join(clioStatePath(), "residency-locks", key.replace(/[^a-zA-Z0-9._-]+/g, "_"));
+  return join2(clioStatePath(), "residency-locks", key.replace(/[^a-zA-Z0-9._-]+/g, "_"));
 }
 async function withResidencyLock(targetKey, fn) {
   return withStateFileLock(lockTargetFor(targetKey), fn, {
@@ -905,6 +1042,13 @@ function harmonyReasoningEffort(level) {
   if (level === "medium") return "medium";
   return "low";
 }
+var MEMORY_ALWAYS_ON_MODEL_MIN_OUTPUT_TOKENS = 4e3;
+function memoryInterventionModelMaxTokens(input) {
+  const configured = Number.isSafeInteger(input.configuredMaxTokens) && input.configuredMaxTokens > 0 ? input.configuredMaxTokens : 1;
+  const requested = input.thinkingMechanism === "always-on" ? Math.max(configured * 2, MEMORY_ALWAYS_ON_MODEL_MIN_OUTPUT_TOKENS) : configured;
+  const modelCap = input.modelMaxTokens;
+  return typeof modelCap === "number" && Number.isSafeInteger(modelCap) && modelCap > 0 ? Math.min(requested, modelCap) : requested;
+}
 function reasoningClassForMechanism(mechanism) {
   if (mechanism === "none") return "never";
   if (mechanism === "always-on") return "always";
@@ -1295,6 +1439,91 @@ function resolveTargetRuntimeCapabilities(target, runtime, wireModelId, capabili
   });
 }
 
+// src/domains/providers/target-model-cache.ts
+init_esm_shims();
+import { createHash } from "node:crypto";
+import { existsSync as existsSync2, readFileSync as readFileSync2 } from "node:fs";
+import { join as join3 } from "node:path";
+var TARGET_MODEL_CACHE_TTL_MS = 24 * 60 * 60 * 1e3;
+var TARGET_MODEL_CACHE_VERSION = 1;
+var TARGET_MODEL_CACHE_MAX_MODELS = 4096;
+var TARGET_MODEL_CACHE_MAX_MODEL_ID_CHARS = 1024;
+function targetCacheKey(targetId) {
+  return createHash("sha256").update(targetId).digest("hex").slice(0, 24);
+}
+function targetModelSnapshotPath(targetId, cacheDir = clioCacheDir()) {
+  return join3(cacheDir, "provider-target-models", `${targetCacheKey(targetId)}.json`);
+}
+function normalizedModels(models) {
+  const seen = /* @__PURE__ */ new Set();
+  const out = [];
+  for (const raw of models) {
+    const model = raw.trim();
+    if (model.length === 0 || model.length > TARGET_MODEL_CACHE_MAX_MODEL_ID_CHARS || seen.has(model)) continue;
+    seen.add(model);
+    out.push(model);
+    if (out.length >= TARGET_MODEL_CACHE_MAX_MODELS) break;
+  }
+  return out;
+}
+function parseSnapshot(value) {
+  if (typeof value !== "object" || value === null || Array.isArray(value)) return null;
+  const row = value;
+  if (row.version !== TARGET_MODEL_CACHE_VERSION) return null;
+  if (typeof row.targetId !== "string" || row.targetId.length === 0) return null;
+  if (typeof row.runtimeId !== "string" || row.runtimeId.length === 0) return null;
+  if (row.url !== null && typeof row.url !== "string") return null;
+  if (!Array.isArray(row.models) || row.models.some((model) => typeof model !== "string")) return null;
+  if (typeof row.observedAt !== "string" || !Number.isFinite(Date.parse(row.observedAt))) return null;
+  const models = normalizedModels(row.models);
+  if (models.length !== row.models.length) return null;
+  return {
+    version: TARGET_MODEL_CACHE_VERSION,
+    targetId: row.targetId,
+    runtimeId: row.runtimeId,
+    url: row.url,
+    models,
+    observedAt: row.observedAt
+  };
+}
+function readTargetModelSnapshot(target, options = {}) {
+  const path = targetModelSnapshotPath(target.id, options.cacheDir);
+  if (!existsSync2(path)) return null;
+  try {
+    const snapshot = parseSnapshot(JSON.parse(readFileSync2(path, "utf8")));
+    if (snapshot === null) return null;
+    if (snapshot.targetId !== target.id || snapshot.runtimeId !== target.runtime) return null;
+    if (snapshot.url !== (target.url ?? null)) return null;
+    const observedMs = Date.parse(snapshot.observedAt);
+    const nowMs = options.nowMs ?? Date.now();
+    const ttlMs = options.ttlMs ?? TARGET_MODEL_CACHE_TTL_MS;
+    if (observedMs > nowMs || nowMs - observedMs >= ttlMs) return null;
+    return snapshot;
+  } catch {
+    return null;
+  }
+}
+function recordTargetModelSnapshot(target, models, options = {}) {
+  if (target.id.length === 0 || target.runtime.length === 0) return false;
+  const snapshot = {
+    version: TARGET_MODEL_CACHE_VERSION,
+    targetId: target.id,
+    runtimeId: target.runtime,
+    url: target.url ?? null,
+    models: normalizedModels(models),
+    observedAt: new Date(options.nowMs ?? Date.now()).toISOString()
+  };
+  try {
+    safeResourceWrite(targetModelSnapshotPath(target.id, options.cacheDir), `${JSON.stringify(snapshot, null, 2)}
+`, {
+      encoding: "utf8"
+    });
+    return true;
+  } catch {
+    return false;
+  }
+}
+
 // src/domains/providers/eligibility.ts
 init_esm_shims();
 function isTargetEligibleRuntime(runtime) {
@@ -1400,6 +1629,105 @@ function canonicalizeWireModelId(status, requested) {
   return matches.length === 1 ? matches[0] ?? requested : requested;
 }
 
+// src/domains/providers/resolver.ts
+init_esm_shims();
+var GLOB_CHARS = /[*?[]/;
+function isThinkingLevel(token) {
+  return VALID_THINKING_LEVELS.includes(token);
+}
+function splitThinkingSuffix(pattern) {
+  const idx = pattern.lastIndexOf(":");
+  if (idx === -1) return { base: pattern };
+  const suffix = pattern.slice(idx + 1).trim();
+  if (!isThinkingLevel(suffix)) return { base: pattern };
+  const base = pattern.slice(0, idx).trim();
+  return base ? { base, thinkingLevel: suffix } : { base: pattern };
+}
+function globToRegExp(glob) {
+  let body = "^";
+  for (let i = 0; i < glob.length; i++) {
+    const ch = glob[i] ?? "";
+    if (ch === "*") {
+      body += ".*";
+    } else if (ch === "?") {
+      body += ".";
+    } else if (ch === "[") {
+      const close = glob.indexOf("]", i + 1);
+      if (close === -1) {
+        body += "\\[";
+        continue;
+      }
+      body += `[${glob.slice(i + 1, close)}]`;
+      i = close;
+    } else if (/[.+^${}()|\\]/.test(ch)) {
+      body += `\\${ch}`;
+    } else {
+      body += ch;
+    }
+  }
+  body += "$";
+  return new RegExp(body, "i");
+}
+function collectCandidates(providers) {
+  const out = [];
+  const seen = /* @__PURE__ */ new Set();
+  for (const status of providers.list()) {
+    for (const model of modelIdsForStatus(status)) {
+      const key = `${status.target.id}/${model}`;
+      if (seen.has(key)) continue;
+      seen.add(key);
+      out.push({ target: status.target.id, model, full: key });
+    }
+  }
+  return out;
+}
+function pickFirst(candidates) {
+  if (candidates.length === 0) return null;
+  const head = candidates[0];
+  if (!head) return null;
+  return { ref: head, ambiguous: candidates.length > 1 };
+}
+function buildResult(pattern, pick, thinkingLevel) {
+  if (!pick) {
+    return { ref: null, error: `no models match pattern "${pattern}"` };
+  }
+  const ref = { target: pick.ref.target, model: pick.ref.model };
+  if (thinkingLevel) ref.thinkingLevel = thinkingLevel;
+  const result = { ref };
+  if (pick.ambiguous) {
+    result.warning = `pattern "${pattern}" matched multiple models; using ${pick.ref.full}`;
+  }
+  return result;
+}
+function resolveModelReference(rawPattern, providers) {
+  const trimmed = rawPattern.trim();
+  if (!trimmed) return { ref: null, error: "empty pattern" };
+  const { base, thinkingLevel } = splitThinkingSuffix(trimmed);
+  const candidates = collectCandidates(providers);
+  if (candidates.length === 0) {
+    return { ref: null, error: "no targets configured; add one with `clio-coder configure` or `clio-coder targets add`" };
+  }
+  const lower = base.toLowerCase();
+  const exactFull = candidates.filter((c) => c.full.toLowerCase() === lower);
+  if (exactFull.length > 0) return buildResult(base, pickFirst(exactFull), thinkingLevel);
+  if (GLOB_CHARS.test(base)) {
+    const re = globToRegExp(base);
+    const globMatches = candidates.filter((c) => re.test(c.full) || re.test(c.model));
+    return buildResult(base, pickFirst(globMatches), thinkingLevel);
+  }
+  const slashIdx = base.indexOf("/");
+  if (slashIdx !== -1) {
+    const ep = base.slice(0, slashIdx).toLowerCase();
+    const mdl = base.slice(slashIdx + 1).toLowerCase();
+    const epModel = candidates.filter((c) => c.target.toLowerCase() === ep && c.model.toLowerCase() === mdl);
+    if (epModel.length > 0) return buildResult(base, pickFirst(epModel), thinkingLevel);
+  }
+  const exactModel = candidates.filter((c) => c.model.toLowerCase() === lower);
+  if (exactModel.length > 0) return buildResult(base, pickFirst(exactModel), thinkingLevel);
+  const partial = candidates.filter((c) => c.model.toLowerCase().includes(lower));
+  return buildResult(base, pickFirst(partial), thinkingLevel);
+}
+
 // src/domains/providers/runtime-resolution.ts
 init_esm_shims();
 function diagnostic(severity, code, message) {
@@ -3754,8 +4082,8 @@ function registerClioApiProviders() {
 
 // src/domains/providers/knowledge-base-path.ts
 init_esm_shims();
-import { existsSync, statSync } from "node:fs";
-import { delimiter, dirname, join as join2 } from "node:path";
+import { existsSync as existsSync3, statSync } from "node:fs";
+import { delimiter, dirname, join as join4 } from "node:path";
 import { fileURLToPath } from "node:url";
 var MODEL_CATALOG_OVERLAY_DIR = "model-catalog.d";
 var MODEL_CATALOG_DIRS_ENV = "CLIO_CODER_MODEL_CATALOG_DIRS";
@@ -3769,19 +4097,19 @@ function isDirectory(path) {
 function resolveProvidersModelsDir(importMetaUrl) {
   const start = dirname(fileURLToPath(importMetaUrl));
   const directCandidates = [
-    join2(start, "models"),
-    join2(start, "..", "domains", "providers", "models"),
-    join2(start, "..", "providers-models")
+    join4(start, "models"),
+    join4(start, "..", "domains", "providers", "models"),
+    join4(start, "..", "providers-models")
   ];
   for (const candidate of directCandidates) {
     if (isDirectory(candidate)) return candidate;
   }
   let cursor = start;
   for (let i = 0; i < 8; i++) {
-    const packageJson = join2(cursor, "package.json");
-    const sourceModels = join2(cursor, "src", "domains", "providers", "models");
-    const distModels = join2(cursor, "dist", "providers-models");
-    if (existsSync(packageJson)) {
+    const packageJson = join4(cursor, "package.json");
+    const sourceModels = join4(cursor, "src", "domains", "providers", "models");
+    const distModels = join4(cursor, "dist", "providers-models");
+    if (existsSync3(packageJson)) {
       if (isDirectory(sourceModels)) return sourceModels;
       if (isDirectory(distModels)) return distModels;
     }
@@ -3811,8 +4139,8 @@ function resolveProviderModelCatalogDirs(importMetaUrl, options = {}) {
   const bundled = resolveProvidersModelsDir(importMetaUrl);
   const cwd = options.cwd ?? process.cwd();
   const overlays = uniqueExistingDirs([
-    join2(resolveClioDirs().config, MODEL_CATALOG_OVERLAY_DIR),
-    join2(cwd, ".clio-coder", MODEL_CATALOG_OVERLAY_DIR),
+    join4(resolveClioDirs().config, MODEL_CATALOG_OVERLAY_DIR),
+    join4(cwd, ".clio-coder", MODEL_CATALOG_OVERLAY_DIR),
     ...envOverlayDirs()
   ]);
   return {
@@ -3831,7 +4159,7 @@ function resolveProviderKnowledgeBaseRoots(importMetaUrl, options = {}) {
 
 // src/domains/providers/plugins.ts
 init_esm_shims();
-import { join as join3 } from "node:path";
+import { join as join5 } from "node:path";
 function extractPluginPackages(settings) {
   if (!settings || typeof settings !== "object") return [];
   const raw = settings.runtimePlugins;
@@ -3840,7 +4168,7 @@ function extractPluginPackages(settings) {
 }
 async function loadPluginRuntimes(registry, settings) {
   const loaded = [];
-  const pluginDir = join3(clioConfigDir(), "runtimes");
+  const pluginDir = join5(clioConfigDir(), "runtimes");
   const packages = extractPluginPackages(settings);
   try {
     const ids = await registry.loadFromDir(pluginDir, activateExternalPluginApiBridge);
@@ -3868,8 +4196,8 @@ async function loadPluginRuntimes(registry, settings) {
 // src/domains/providers/types/knowledge-base.ts
 init_esm_shims();
 var import_yaml = __toESM(require_dist(), 1);
-import { readdirSync, readFileSync, statSync as statSync2 } from "node:fs";
-import { join as join4 } from "node:path";
+import { readdirSync, readFileSync as readFileSync3, statSync as statSync2 } from "node:fs";
+import { join as join6 } from "node:path";
 var FileKnowledgeBase = class {
   roots;
   loaded = [];
@@ -3883,7 +4211,7 @@ var FileKnowledgeBase = class {
       const files = collectYamlFiles(root.dir);
       for (const file of files) {
         const name = root.label ? `${root.label}:${file.name}` : file.name;
-        const raw = readFileSync(file.path, "utf8");
+        const raw = readFileSync3(file.path, "utf8");
         const parsed = (0, import_yaml.parse)(raw);
         if (!Array.isArray(parsed)) {
           throw new Error(`knowledge base file ${name} must be a YAML list of KnowledgeBaseEntry`);
@@ -3941,7 +4269,7 @@ function collectYamlFiles(dir, prefix = "") {
   const out = [];
   const entries = readdirSync(dir, { withFileTypes: true }).sort((a, b) => a.name.localeCompare(b.name));
   for (const entry of entries) {
-    const path = join4(dir, entry.name);
+    const path = join6(dir, entry.name);
     const name = prefix ? `${prefix}/${entry.name}` : entry.name;
     if (entry.isDirectory()) {
       out.push(...collectYamlFiles(path, name));
@@ -4047,16 +4375,17 @@ function uniqueModels2(ids) {
   }
   return out;
 }
-function discoveredModelsSource(probe, preservePreviousProbe, previous, desc) {
+function discoveredModelsSource(probe, preservePreviousProbe, previous, cachedModels, desc) {
   if (probe?.models !== void 0) return "probe";
   if (preservePreviousProbe && previous?.discoveredModels && previous.discoveredModels.length > 0) return "cache";
+  if (cachedModels.length > 0) return "cache";
   if (desc.knownModels && desc.knownModels.length > 0) return "runtime";
   return "none";
 }
 function sameProbeIdentity(previous, next) {
   return previous.id === next.id && previous.runtime === next.runtime && (previous.url ?? "") === (next.url ?? "") && (previous.defaultModel ?? "") === (next.defaultModel ?? "");
 }
-function mergeProbeResult(desc, target, probe, previous) {
+function mergeProbeResult(desc, target, probe, previous, cachedModels = []) {
   const probeSucceeded = probe?.ok ?? false;
   const preservePrevious = !probeSucceeded && previous !== void 0 && sameProbeIdentity(previous.target, target);
   const probeCapabilities = probe?.discoveredCapabilities ?? (preservePrevious ? previous.probeCapabilities : null) ?? null;
@@ -4065,12 +4394,12 @@ function mergeProbeResult(desc, target, probe, previous) {
   const probeNotes = probe?.notes && probe.notes.length > 0 ? probe.notes : preservePrevious ? previous.probeNotes : void 0;
   const probeSurfaces = probe?.surfaces ?? (preservePrevious ? previous.probeSurfaces : void 0);
   const discoveredModels = uniqueModels2(
-    probe?.models ?? (preservePrevious ? previous.discoveredModels : void 0) ?? desc.knownModels ?? []
+    probe?.models ?? (preservePrevious ? previous.discoveredModels : void 0) ?? (cachedModels.length > 0 ? cachedModels : void 0) ?? desc.knownModels ?? []
   );
   const discoveredModelStates = probe?.modelStates ?? (preservePrevious ? previous.discoveredModelStates : null) ?? null;
   const merge = {
     discoveredModels,
-    discoveredModelsSource: discoveredModelsSource(probe, preservePrevious, previous, desc),
+    discoveredModelsSource: discoveredModelsSource(probe, preservePrevious, previous, cachedModels, desc),
     discoveredModelStates,
     probeCapabilities,
     probeModelCapabilities,
@@ -4163,7 +4492,8 @@ function createProvidersBundle(context) {
       return out2;
     }
     const availability = availabilityFor(desc, target, authStatusFor);
-    const merge = mergeProbeResult(desc, target, probe, previous);
+    const cachedModels = readTargetModelSnapshot(target)?.models ?? [];
+    const merge = mergeProbeResult(desc, target, probe, previous, cachedModels);
     const { capabilities, contextWindowProvenance } = capabilitiesFor(desc, target, merge, kb);
     const healthy = probe !== null ? probe.ok : null;
     const unservedDefault = probe?.ok ? unservedDefaultModelReason(desc, target, merge) : null;
@@ -4251,6 +4581,10 @@ function createProvidersBundle(context) {
     }
     const status = buildStatus(target, desc, probeResult, previous);
     statuses.set(target.id, status);
+    if (probeResult.ok && probeResult.models !== void 0) {
+      recordTargetModelSnapshot(target, probeResult.models);
+    }
+    void recordEndpointSlotsFromStatus(status);
     context.bus.emit(BusChannels.ProviderHealth, { id: target.id, status });
     return status;
   }
@@ -4413,105 +4747,6 @@ var ProvidersManifest = {
   dependsOn: ["config"]
 };
 
-// src/domains/providers/resolver.ts
-init_esm_shims();
-var GLOB_CHARS = /[*?[]/;
-function isThinkingLevel(token) {
-  return VALID_THINKING_LEVELS.includes(token);
-}
-function splitThinkingSuffix(pattern) {
-  const idx = pattern.lastIndexOf(":");
-  if (idx === -1) return { base: pattern };
-  const suffix = pattern.slice(idx + 1).trim();
-  if (!isThinkingLevel(suffix)) return { base: pattern };
-  const base = pattern.slice(0, idx).trim();
-  return base ? { base, thinkingLevel: suffix } : { base: pattern };
-}
-function globToRegExp(glob) {
-  let body = "^";
-  for (let i = 0; i < glob.length; i++) {
-    const ch = glob[i] ?? "";
-    if (ch === "*") {
-      body += ".*";
-    } else if (ch === "?") {
-      body += ".";
-    } else if (ch === "[") {
-      const close = glob.indexOf("]", i + 1);
-      if (close === -1) {
-        body += "\\[";
-        continue;
-      }
-      body += `[${glob.slice(i + 1, close)}]`;
-      i = close;
-    } else if (/[.+^${}()|\\]/.test(ch)) {
-      body += `\\${ch}`;
-    } else {
-      body += ch;
-    }
-  }
-  body += "$";
-  return new RegExp(body, "i");
-}
-function collectCandidates(providers) {
-  const out = [];
-  const seen = /* @__PURE__ */ new Set();
-  for (const status of providers.list()) {
-    for (const model of modelIdsForStatus(status)) {
-      const key = `${status.target.id}/${model}`;
-      if (seen.has(key)) continue;
-      seen.add(key);
-      out.push({ target: status.target.id, model, full: key });
-    }
-  }
-  return out;
-}
-function pickFirst(candidates) {
-  if (candidates.length === 0) return null;
-  const head = candidates[0];
-  if (!head) return null;
-  return { ref: head, ambiguous: candidates.length > 1 };
-}
-function buildResult(pattern, pick, thinkingLevel) {
-  if (!pick) {
-    return { ref: null, error: `no models match pattern "${pattern}"` };
-  }
-  const ref = { target: pick.ref.target, model: pick.ref.model };
-  if (thinkingLevel) ref.thinkingLevel = thinkingLevel;
-  const result = { ref };
-  if (pick.ambiguous) {
-    result.warning = `pattern "${pattern}" matched multiple models; using ${pick.ref.full}`;
-  }
-  return result;
-}
-function resolveModelReference(rawPattern, providers) {
-  const trimmed = rawPattern.trim();
-  if (!trimmed) return { ref: null, error: "empty pattern" };
-  const { base, thinkingLevel } = splitThinkingSuffix(trimmed);
-  const candidates = collectCandidates(providers);
-  if (candidates.length === 0) {
-    return { ref: null, error: "no targets configured; add one with `clio-coder configure` or `clio-coder targets add`" };
-  }
-  const lower = base.toLowerCase();
-  const exactFull = candidates.filter((c) => c.full.toLowerCase() === lower);
-  if (exactFull.length > 0) return buildResult(base, pickFirst(exactFull), thinkingLevel);
-  if (GLOB_CHARS.test(base)) {
-    const re = globToRegExp(base);
-    const globMatches = candidates.filter((c) => re.test(c.full) || re.test(c.model));
-    return buildResult(base, pickFirst(globMatches), thinkingLevel);
-  }
-  const slashIdx = base.indexOf("/");
-  if (slashIdx !== -1) {
-    const ep = base.slice(0, slashIdx).toLowerCase();
-    const mdl = base.slice(slashIdx + 1).toLowerCase();
-    const epModel = candidates.filter((c) => c.target.toLowerCase() === ep && c.model.toLowerCase() === mdl);
-    if (epModel.length > 0) return buildResult(base, pickFirst(epModel), thinkingLevel);
-  }
-  const exactModel = candidates.filter((c) => c.model.toLowerCase() === lower);
-  if (exactModel.length > 0) return buildResult(base, pickFirst(exactModel), thinkingLevel);
-  const partial = candidates.filter((c) => c.model.toLowerCase().includes(lower));
-  return buildResult(base, pickFirst(partial), thinkingLevel);
-}
-
 // src/domains/providers/index.ts
 var ProvidersDomainModule = {
   manifest: ProvidersManifest,
@@ -4524,8 +4759,10 @@ export {
   resolveReservedOutputTokens,
   canonicalEndpointKey,
   endpointLabel,
+  endpointCapacityFor,
   endpointCapacityForStatus,
   endpointCapacitiesForStatuses,
+  resolveEndpointCapacities,
   registerForegroundStream,
   foregroundStreamUsage,
   setResidencyNoticeSink,
@@ -4536,6 +4773,7 @@ export {
   applyModelCapabilityPatch,
   probeCapabilitiesForModel,
   resolveModelCapabilities,
+  memoryInterventionModelMaxTokens,
   thinkingLevelChoiceLabel,
   thinkingLevelFromChoiceLabel,
   resolveModelRuntimeCapabilitiesForProviders,
@@ -4543,6 +4781,8 @@ export {
   registerClioApiProviders,
   resolveProviderKnowledgeBaseRoots,
   loadPluginRuntimes,
+  readTargetModelSnapshot,
+  recordTargetModelSnapshot,
   FileKnowledgeBase,
   isOrchestratorEligibleRuntime,
   isDispatchEligibleRuntime,
@@ -4560,4 +4800,4 @@ export {
   normalizeCostProvenance,
   ProvidersDomainModule
 };
-//# sourceMappingURL=chunk-MIX5N5AC.js.map
+//# sourceMappingURL=chunk-DQITNCXG.js.map