@oxagen/cli 3.0.1-155 → 3.0.1-162

package/oxagen.mjs

@@ -5704,7 +5704,7 @@ function bundleSkillChars(skill) {
     chars += file.path.length + file.content.length;
   return chars;
 }
-var protocolTimestampSchema, machineGroupSchema, TACHO_BATCH_SCHEMA, TACHO_BUNDLE_SCHEMA, TACHO_COMMANDS_SCHEMA, TACHO_ENROLLMENT_CLAIMS_SCHEMA, TACHO_ENROLLMENT_CERTIFICATE_DOMAIN, ENROLLMENT_KEY_SET_SCHEMA, ENROLLMENT_KEY_SET_DOMAIN, TACHO_ENFORCEMENT_TIER_ATTR, TACHO_GATEWAY_TIER, TACHO_GATEWAY_SESSION_HEADER, TACHO_GATEWAY_GENESIS_HEADER, REFUSAL_META_KEY, DECISION_META_KEY, MCP_STREAMABLE_HTTP_ACCEPT, TACHO_METERING_ATTR, TACHO_METERING_OBSERVED, TACHO_MODEL_SESSION_HEADER, TACHO_CREDENTIAL_BASIS_ATTR, TACHO_CREDENTIAL_GATEWAY_BROKERED, TACHO_CREDENTIAL_HARNESS_HELD, TACHO_CREDENTIAL_BASES, TACHO_RUN_TOKEN_ATTR, BUNDLE_FEATURE_GATEWAY_TOOLS, BUNDLE_FEATURE_MODEL_PRICES, BUNDLE_FEATURE_MODEL_ALLOWLIST, BUNDLE_FEATURE_INDEPENDENT_MODELS, BUNDLE_FEATURE_HOOK_FAIL_OPEN, BUNDLE_FEATURE_STEERING_MANIFEST, BUNDLE_FEATURE_STEERING_INCOMPLETE, BUNDLE_FEATURE_STEERING_LOAD, BUNDLE_FEATURE_CONTAINMENT, BUNDLE_FEATURE_DAILY_BUDGET, BUNDLE_FEATURE_STEER_NEXT_STEP, BUNDLE_FEATURE_UNBOUND_REPO, BUNDLE_FEATURE_CEDAR, BUNDLE_FEATURE_CEDAR_TOOLBELT, BUNDLE_FEATURE_SKILLS, BUNDLE_FEATURE_CACHE_KEEP_ALIVE, BUNDLE_FEATURE_WORK_ORDERS, BUNDLE_FEATURE_WORK_ORDER_START, BUNDLE_FEATURE_WORK_ORDER_STAFFING, BUNDLE_FEATURE_MCP_RULE_SETS, BUNDLE_FEATURE_GATEWAY_POLICY, BUNDLE_FEATURE_TENANT_SCOPE, BUNDLE_FEATURE_MEMORY_SCOPE, MEMORY_SCOPE_DIGESTS_MAX, GATEWAY_MIN_TIERS, SENSITIVE_SCREEN_ACTIONS, PROMPT_REDACTION_PATTERNS_MAX, PROMPT_REDACTION_PATTERN_MAX, gatewayPolicySchema, TACHO_BUNDLE_FEATURES, bundleFeaturesSchema, hostEnrollmentIdSchema, MEMBER_PUBLIC_ID, tachoPlatformSchema, tachoHarnessSchema, TACHO_HARNESS_LABELS, TACHO_HARNESS_TIERS, WRAPPED_HARNESSES, CONNECTED_HARNESSES, HARNESS_BINARY, MODEL_HARNESS_ROUTES, MODEL_ROUTED_HARNESSES, BROKERABLE_HARNESSES, BROKERABLE_HARNESS_PROVIDER, TACHO_TIER_SUMMARY, CUSTOM_AGENT_NAME_PATTERN, RESERVED_AGENT_NAMES, tachoHostStatusSchema, tachoBundleModeSchema, tachoCommandSchema, tachoCommandStatusSchema, tachoCommandAckStatusSchema, tachoDeliveryModeSchema, denyGenerationSchema, deliveredCommandSchema, deliveredCommandResponseSchema, WORK_ORDER_ID_PATTERN, WORK_ITEM_ID_PATTERN, workOrderCommandPayloadSchema, WORK_ORDER_CLAIM_PATH, CERTIFICATE_RENEW_PATH, ENROLLMENT_KEYS_PATH, WORK_ORDER_REJECT_PATH, workOrderClaimResponseSchema, workOrderRejectResponseSchema, WORK_CRITERION_CLAIM_PATH, WORK_CRITERION_ID_PATTERN, WORK_HEAD_SHA_PATTERN, WORK_CRITERION_CLAIM_TEXT_MAX, workCriterionClaimResponseSchema, WORK_ORDER_ENV, WORK_ORDER_ATTR, steeringForceSchema, steeringItemKindSchema, steeringCutReasonSchema, steeringHeldLoadSchema, steeringManifestItemSchema, STEERING_MANIFEST_SCHEMA, steeringManifestSchema, steeringManifestFrameSchema, workOrderSteeringReasonSchema, workOrderSteeringItemSchema, workOrderSteeringManifestSchema, cedarIdSchema, cedarPrincipalSchema, cedarToolSchema, cedarBundleSchema, BUNDLE_SKILLS_MAX, BUNDLE_SKILL_FILES_MAX, BUNDLE_SKILLS_CHARS_MAX, BASE64_PATTERN, bundleSkillFileSchema, bundleSkillSchema, policyBundleSchema, enrollmentClaimsSchema, ed25519SignatureSchema, enrollmentCertificateSchema, enrollmentKeySchema, enrollmentKeySetSchema, tachoEventWireSchema, TACHO_MAX_HELD_SESSIONS, daemonHealthSchema, TACHO_MAX_BATCH, TACHO_MAX_BODY_BYTES, TACHO_MAX_REQUEST_BYTES, TACHO_REQUEST_ENVELOPE_BYTES, SANDBOX_GIT_RELAY_PATH, SANDBOX_GITLAB_RELAY_PATH, SANDBOX_GIT_RELAY_MAX_REQUEST_BYTES, SANDBOX_GIT_RELAY_TIMEOUT_MS, tachoBodySchema, tachoBatchSchema, agentDaySpendSchema, controlEnvelopeSchema, ingestResponseSchema, bundleResponseSchema, commandAcknowledgementSchema, commandsResponseSchema, certificateRenewalResponseSchema, enrollmentKeysResponseSchema, enrollmentResponseSchema, tokenEnrollmentResponseSchema;
+var protocolTimestampSchema, machineGroupSchema, TACHO_BATCH_SCHEMA, TACHO_BUNDLE_SCHEMA, TACHO_COMMANDS_SCHEMA, TACHO_ENROLLMENT_CLAIMS_SCHEMA, TACHO_ENROLLMENT_CERTIFICATE_DOMAIN, ENROLLMENT_KEY_SET_SCHEMA, ENROLLMENT_KEY_SET_DOMAIN, TACHO_ENFORCEMENT_TIER_ATTR, TACHO_GATEWAY_TIER, TACHO_GATEWAY_SESSION_HEADER, TACHO_GATEWAY_GENESIS_HEADER, REFUSAL_META_KEY, DECISION_META_KEY, MCP_STREAMABLE_HTTP_ACCEPT, TACHO_METERING_ATTR, TACHO_METERING_OBSERVED, TACHO_MODEL_SESSION_HEADER, TACHO_CREDENTIAL_BASIS_ATTR, TACHO_CREDENTIAL_GATEWAY_BROKERED, TACHO_CREDENTIAL_HARNESS_HELD, TACHO_CREDENTIAL_BASES, TACHO_RUN_TOKEN_ATTR, BUNDLE_FEATURE_GATEWAY_TOOLS, BUNDLE_FEATURE_MODEL_PRICES, BUNDLE_FEATURE_MODEL_PRICES_LONG_CONTEXT, BUNDLE_FEATURE_MODEL_ALLOWLIST, BUNDLE_FEATURE_INDEPENDENT_MODELS, BUNDLE_FEATURE_HOOK_FAIL_OPEN, BUNDLE_FEATURE_STEERING_MANIFEST, BUNDLE_FEATURE_STEERING_INCOMPLETE, BUNDLE_FEATURE_STEERING_LOAD, BUNDLE_FEATURE_CONTAINMENT, BUNDLE_FEATURE_DAILY_BUDGET, BUNDLE_FEATURE_STEER_NEXT_STEP, BUNDLE_FEATURE_UNBOUND_REPO, BUNDLE_FEATURE_CEDAR, BUNDLE_FEATURE_CEDAR_TOOLBELT, BUNDLE_FEATURE_SKILLS, BUNDLE_FEATURE_CACHE_KEEP_ALIVE, BUNDLE_FEATURE_WORK_ORDERS, BUNDLE_FEATURE_WORK_ORDER_START, BUNDLE_FEATURE_WORK_ORDER_STAFFING, BUNDLE_FEATURE_MCP_RULE_SETS, BUNDLE_FEATURE_GATEWAY_POLICY, BUNDLE_FEATURE_TENANT_SCOPE, BUNDLE_FEATURE_MEMORY_SCOPE, MEMORY_SCOPE_DIGESTS_MAX, GATEWAY_MIN_TIERS, SENSITIVE_SCREEN_ACTIONS, PROMPT_REDACTION_PATTERNS_MAX, PROMPT_REDACTION_PATTERN_MAX, gatewayPolicySchema, TACHO_BUNDLE_FEATURES, bundleFeaturesSchema, hostEnrollmentIdSchema, MEMBER_PUBLIC_ID, tachoPlatformSchema, tachoHarnessSchema, TACHO_HARNESS_LABELS, TACHO_HARNESS_TIERS, WRAPPED_HARNESSES, CONNECTED_HARNESSES, HARNESS_BINARY, MODEL_HARNESS_ROUTES, MODEL_ROUTED_HARNESSES, BROKERABLE_HARNESSES, BROKERABLE_HARNESS_PROVIDER, TACHO_TIER_SUMMARY, CUSTOM_AGENT_NAME_PATTERN, RESERVED_AGENT_NAMES, tachoHostStatusSchema, tachoBundleModeSchema, tachoCommandSchema, tachoCommandStatusSchema, tachoCommandAckStatusSchema, tachoDeliveryModeSchema, denyGenerationSchema, deliveredCommandSchema, deliveredCommandResponseSchema, WORK_ORDER_ID_PATTERN, WORK_ITEM_ID_PATTERN, workOrderCommandPayloadSchema, WORK_ORDER_CLAIM_PATH, CERTIFICATE_RENEW_PATH, ENROLLMENT_KEYS_PATH, WORK_ORDER_REJECT_PATH, workOrderClaimResponseSchema, workOrderRejectResponseSchema, WORK_CRITERION_CLAIM_PATH, WORK_CRITERION_ID_PATTERN, WORK_HEAD_SHA_PATTERN, WORK_CRITERION_CLAIM_TEXT_MAX, workCriterionClaimResponseSchema, WORK_ORDER_ENV, WORK_ORDER_ATTR, steeringForceSchema, steeringItemKindSchema, steeringCutReasonSchema, steeringHeldLoadSchema, steeringManifestItemSchema, STEERING_MANIFEST_SCHEMA, steeringManifestSchema, steeringManifestFrameSchema, workOrderSteeringReasonSchema, workOrderSteeringItemSchema, workOrderSteeringManifestSchema, cedarIdSchema, cedarPrincipalSchema, cedarToolSchema, cedarBundleSchema, BUNDLE_SKILLS_MAX, BUNDLE_SKILL_FILES_MAX, BUNDLE_SKILLS_CHARS_MAX, BASE64_PATTERN, bundleSkillFileSchema, bundleSkillSchema, policyBundleSchema, enrollmentClaimsSchema, ed25519SignatureSchema, enrollmentCertificateSchema, enrollmentKeySchema, enrollmentKeySetSchema, tachoEventWireSchema, TACHO_MAX_HELD_SESSIONS, daemonHealthSchema, TACHO_MAX_BATCH, TACHO_MAX_BODY_BYTES, TACHO_MAX_REQUEST_BYTES, TACHO_REQUEST_ENVELOPE_BYTES, SANDBOX_GIT_RELAY_PATH, SANDBOX_GITLAB_RELAY_PATH, SANDBOX_GIT_RELAY_MAX_REQUEST_BYTES, SANDBOX_GIT_RELAY_TIMEOUT_MS, tachoBodySchema, tachoBatchSchema, agentDaySpendSchema, controlEnvelopeSchema, ingestResponseSchema, bundleResponseSchema, commandAcknowledgementSchema, commandsResponseSchema, certificateRenewalResponseSchema, enrollmentKeysResponseSchema, enrollmentResponseSchema, tokenEnrollmentResponseSchema;
 var init_wire = __esm({
   "../../packages/tacho/src/wire.ts"() {
     "use strict";
@@ -5746,6 +5746,7 @@ var init_wire = __esm({
     TACHO_RUN_TOKEN_ATTR = "oxagen.run_token_id";
     BUNDLE_FEATURE_GATEWAY_TOOLS = "gateway_tools";
     BUNDLE_FEATURE_MODEL_PRICES = "model_prices";
+    BUNDLE_FEATURE_MODEL_PRICES_LONG_CONTEXT = "model_prices_long_context";
     BUNDLE_FEATURE_MODEL_ALLOWLIST = "models";
     BUNDLE_FEATURE_INDEPENDENT_MODELS = "models_independent";
     BUNDLE_FEATURE_HOOK_FAIL_OPEN = "hook_fail_open";
@@ -5818,7 +5819,8 @@ var init_wire = __esm({
       BUNDLE_FEATURE_MCP_RULE_SETS,
       BUNDLE_FEATURE_GATEWAY_POLICY,
       BUNDLE_FEATURE_TENANT_SCOPE,
-      BUNDLE_FEATURE_MEMORY_SCOPE
+      BUNDLE_FEATURE_MEMORY_SCOPE,
+      BUNDLE_FEATURE_MODEL_PRICES_LONG_CONTEXT
     ];
     bundleFeaturesSchema = external_exports.array(external_exports.string().max(64)).max(32);
     hostEnrollmentIdSchema = external_exports.string().regex(/^tch_[a-z0-9]{22}$/, "a host enrollment public id");
@@ -6319,6 +6321,12 @@ var init_wire = __esm({
        * call it cannot price costs the session budget nothing and is recorded as
        * `observed_unpriced`. Emitted only to a host that advertised
        * `BUNDLE_FEATURE_MODEL_PRICES`.
+       *
+       * `long_context` holds the rates a call pays once its whole prompt
+       * (uncached input, cache reads and cache writes) is more than
+       * `above_prompt_tokens` tokens, in the same unit. A row without it pays
+       * one rate at every size. Emitted only to a host that also advertised
+       * `BUNDLE_FEATURE_MODEL_PRICES_LONG_CONTEXT`.
        */
       model_prices: external_exports.array(
         external_exports.object({
@@ -6328,7 +6336,15 @@ var init_wire = __esm({
           output: external_exports.number().int().nonnegative(),
           cache_read: external_exports.number().int().nonnegative(),
           cache_write: external_exports.number().int().nonnegative(),
-          cache_write_1h: external_exports.number().int().nonnegative().optional()
+          cache_write_1h: external_exports.number().int().nonnegative().optional(),
+          long_context: external_exports.object({
+            above_prompt_tokens: external_exports.number().int().positive(),
+            input: external_exports.number().int().nonnegative(),
+            output: external_exports.number().int().nonnegative(),
+            cache_read: external_exports.number().int().nonnegative(),
+            cache_write: external_exports.number().int().nonnegative(),
+            cache_write_1h: external_exports.number().int().nonnegative().optional()
+          }).strict().optional()
         }).strict()
       ).max(1024).optional(),
       /**
@@ -28443,7 +28459,7 @@ import { readFileSync as readFileSync10 } from "node:fs";
 import { dirname as dirname9, resolve as resolve3 } from "node:path";
 import { fileURLToPath } from "node:url";
 function readVersion() {
-  if (true) return "3.0.1-155";
+  if (true) return "3.0.1-162";
   try {
     const here = dirname9(fileURLToPath(import.meta.url));
     const pkg = JSON.parse(
@@ -55171,6 +55187,10 @@ function resolveModelPriceMatch(prices, provider, model) {
 function resolveModelPrice(prices, provider, model) {
   return resolveModelPriceMatch(prices, provider, model)?.price;
 }
+function ratesForPromptTokens(price, promptTokens) {
+  const long = price.long_context;
+  return long !== void 0 && promptTokens > long.above_prompt_tokens ? long : price;
+}
 function priceObservedUsage(prices, provider, usage2) {
   const price = resolveModelPrice(prices, provider, usage2.model);
   if (price === void 0) return void 0;
@@ -55180,7 +55200,8 @@ function priceObservedUsage(prices, provider, usage2) {
   const cacheWrite = usage2.cacheCreationTokens ?? 0;
   const oneHour = Math.min(cacheWrite, usage2.cacheCreation1hTokens ?? 0);
   const fiveMinute = cacheWrite - oneHour;
-  const micros = (input * price.input + output * price.output + cacheRead * price.cache_read + fiveMinute * price.cache_write + oneHour * (price.cache_write_1h ?? price.cache_write)) / MILLION;
+  const rates = ratesForPromptTokens(price, input + cacheRead + cacheWrite);
+  const micros = (input * rates.input + output * rates.output + cacheRead * rates.cache_read + fiveMinute * rates.cache_write + oneHour * (rates.cache_write_1h ?? rates.cache_write)) / MILLION;
   return Math.round(micros);
 }
 function callCeilingMicros(prices, provider, model, requestBytes, maxOutputTokens) {
@@ -55188,12 +55209,14 @@ function callCeilingMicros(prices, provider, model, requestBytes, maxOutputToken
   if (price === void 0) return 0;
   const input = Math.ceil(requestBytes / CEILING_BYTES_PER_TOKEN);
   const output = maxOutputTokens ?? DEFAULT_OUTPUT_CEILING_TOKENS;
-  return Math.round((input * price.input + output * price.output) / MILLION);
+  const mostPromptTokens = Math.ceil(requestBytes / MIN_BYTES_PER_TOKEN);
+  const rates = ratesForPromptTokens(price, mostPromptTokens);
+  return Math.round((input * rates.input + output * rates.output) / MILLION);
 }
 function usdToMicros(usd2) {
   return Math.round(usd2 * MILLION);
 }
-var MODEL_ID_BOUNDARY, POINT_IN_TIME_STAMP, MILLION, CEILING_BYTES_PER_TOKEN, DEFAULT_OUTPUT_CEILING_TOKENS;
+var MODEL_ID_BOUNDARY, POINT_IN_TIME_STAMP, MILLION, CEILING_BYTES_PER_TOKEN, MIN_BYTES_PER_TOKEN, DEFAULT_OUTPUT_CEILING_TOKENS;
 var init_model_pricing = __esm({
   "../../packages/tacho/src/collector/model-pricing.ts"() {
     "use strict";
@@ -55201,6 +55224,7 @@ var init_model_pricing = __esm({
     POINT_IN_TIME_STAMP = /^(?:latest|\d{4}(?:[-._/]?\d{2}[-._/]?\d{2})?)$/;
     MILLION = 1e6;
     CEILING_BYTES_PER_TOKEN = 3.5;
+    MIN_BYTES_PER_TOKEN = 1;
     DEFAULT_OUTPUT_CEILING_TOKENS = 4096;
   }
 });
@@ -70501,7 +70525,7 @@ var init_package = __esm({
   "package.json"() {
     package_default = {
       name: "@oxagen/cli",
-      version: "3.0.1-155",
+      version: "3.0.1-162",
       description: "Oxagen governance CLI: spend, source tracking, grounding, credentials, and audit (agentic coding lives in the stella CLI)",
       type: "module",
       main: "./dist/index.js",
@@ -71818,6 +71842,20 @@ var init_rate_card = __esm({
           cacheWritePer1M: 18.75
         }
       },
+      // Fable 5 holds the family's price. Its own row keeps it off the family
+      // row, and it sorts after the Fable 5.1 rows, which "claude-fable-5" would
+      // otherwise match first.
+      {
+        family: "claude-fable-5",
+        label: "Claude Fable 5",
+        vendor: "anthropic",
+        rate: {
+          inputPer1M: 10,
+          outputPer1M: 50,
+          cachedInputPer1M: 1,
+          cacheWritePer1M: 12.5
+        }
+      },
       {
         family: "claude-fable",
         label: "Claude Fable",
@@ -71917,6 +71955,30 @@ var init_rate_card = __esm({
           longContext: HAIKU_5_5_LONG_CONTEXT
         }
       },
+      // Haiku 4.5 holds the family's price, in both spellings.
+      {
+        family: "claude-haiku-4-5",
+        label: "Claude Haiku 4.5",
+        vendor: "anthropic",
+        rate: {
+          inputPer1M: 1,
+          outputPer1M: 5,
+          cachedInputPer1M: 0.1,
+          cacheWritePer1M: 1.25
+        }
+      },
+      {
+        family: "claude-haiku-4.5",
+        label: "Claude Haiku 4.5",
+        vendor: "anthropic",
+        alias: true,
+        rate: {
+          inputPer1M: 1,
+          outputPer1M: 5,
+          cachedInputPer1M: 0.1,
+          cacheWritePer1M: 1.25
+        }
+      },
       {
         family: "claude-haiku",
         label: "Claude Haiku",
@@ -72042,7 +72104,7 @@ var init_rate_card = __esm({
         rate: {
           inputPer1M: 1.1,
           outputPer1M: 4.4,
-          cachedInputPer1M: 0.28,
+          cachedInputPer1M: 0.275,
           cacheWritePer1M: 1.1
         }
       },
@@ -72214,6 +72276,33 @@ var init_rate_card = __esm({
           cacheWritePer1M: 2
         }
       },
+      // Gemini 3 Flash and Gemini 3.1 Flash Image at Google's list price
+      // (https://ai.google.dev/gemini-api/docs/pricing, 2026-10-17), with the
+      // image model's cache read from the gateway's model list. Both MUST sort
+      // before the generic "gemini" row. `pricing.ts` cites each price and says
+      // what each row leaves out.
+      {
+        family: "gemini-3-flash",
+        label: "Gemini 3 Flash",
+        vendor: "google",
+        rate: {
+          inputPer1M: 0.5,
+          outputPer1M: 3,
+          cachedInputPer1M: 0.05,
+          cacheWritePer1M: 0.5
+        }
+      },
+      {
+        family: "gemini-3.1-flash-image",
+        label: "Gemini 3.1 Flash Image",
+        vendor: "google",
+        rate: {
+          inputPer1M: 0.5,
+          outputPer1M: 3,
+          cachedInputPer1M: 0.05,
+          cacheWritePer1M: 0.5
+        }
+      },
       {
         family: "gemini",
         label: "Gemini",
@@ -72224,6 +72313,31 @@ var init_rate_card = __esm({
           cachedInputPer1M: 0.3125,
           cacheWritePer1M: 1.25
         }
+      },
+      // Llama 4 Maverick and Mistral Large 3 at the gateway's price
+      // (vercel.com/ai-gateway/models and ai-gateway.vercel.sh/v1/models,
+      // 2026-10-17). `pricing.ts` cites each page.
+      {
+        family: "llama-4-maverick",
+        label: "Llama 4 Maverick",
+        vendor: "meta",
+        rate: {
+          inputPer1M: 0.24,
+          outputPer1M: 0.97,
+          cachedInputPer1M: 0.24,
+          cacheWritePer1M: 0.24
+        }
+      },
+      {
+        family: "mistral-large-3",
+        label: "Mistral Large 3",
+        vendor: "mistral",
+        rate: {
+          inputPer1M: 0.5,
+          outputPer1M: 1.5,
+          cachedInputPer1M: 0.05,
+          cacheWritePer1M: 0.5
+        }
       }
     ];
     FALLBACK_RATE = {
@@ -72692,7 +72806,8 @@ function entryRow(row) {
   return [
     row.provider,
     row.model,
-    row.tokenClass,
+    // A long-context row prices only prompts past a size, so it says which.
+    row.abovePromptTokens === void 0 ? row.tokenClass : `${row.tokenClass} (prompts over ${row.abovePromptTokens.toLocaleString("en-US")} tokens)`,
     row.region ?? "\u2014",
     formatPerMillion(row),
     row.source,
@@ -141735,11 +141850,13 @@ var init_env = __esm({
       STORAGE_FS_ROOT: external_exports.string().min(1).optional(),
       // Vercel AI Gateway — the platform's default AI auth boundary for language
       // models. Embeddings do not use it: they go to Voyage on VOYAGE_API_KEY.
-      // The OXAGEN_LLM_* tiers are white-labeled model handles ("Oxagen
-      // Fast/Balanced/Precise") resolving to concrete model ids in `creator/model`
-      // form. Defaults mirror the registry staticValues so local dev and tests
-      // resolve a tier without extra configuration. ADR-043 removed image and video
-      // generation, so text is the only tier family left.
+      // The OXAGEN_LLM_FAST/BALANCED/PRECISE settings name the models Oxagen's own
+      // calls run on (titles, drafts, summaries) in `creator/model` form. A stella
+      // turn's model, and a workspace's or a person's default tier, come from the
+      // organization's model tiers instead (#7197). Defaults mirror the registry
+      // staticValues so local dev and tests resolve a model without extra
+      // configuration. ADR-043 removed image and video generation, so text is the
+      // only family left.
       AI_GATEWAY_API_KEY: external_exports.string().optional(),
       // Which provider serves LANGUAGE models. The gateway is the default and the
       // metered path; "openrouter" selects a direct OpenAI-compatible provider for

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@oxagen/cli",
-  "version": "3.0.1-155",
+  "version": "3.0.1-162",
   "description": "Oxagen governance CLI: spend, source tracking, grounding, credentials, and audit (agentic coding lives in the stella CLI)",
   "type": "module",
   "bin": {