@tryarcanist/cli 0.1.335 → 0.1.337

package/dist/index.js

@@ -7370,7 +7370,7 @@ import { Command } from "commander";
 // package.json
 var package_default = {
   name: "@tryarcanist/cli",
-  version: "0.1.335",
+  version: "0.1.337",
   description: "CLI for Arcanist - create and manage coding agent sessions",
   type: "module",
   bin: {
@@ -23355,14 +23355,6 @@ var OpenAIModel = {
   GPT52ChatLatest: "gpt-5.2-chat-latest",
   GPT52Codex: "gpt-5.2-codex"
 };
-var DeepSeekModel = {
-  V4Flash: "deepseek-v4-flash",
-  V4Pro: "deepseek-v4-pro"
-};
-var GlmModel = {
-  Flash53: "glm-5.3-flash",
-  G53: "glm-5.3"
-};
 var MODEL_PROVIDERS_SET = /* @__PURE__ */ new Set(["openai", "anthropic"]);
 var BACKEND_DESKTOP_IMAGE_FEEDBACK_CONFIGS = {
   [CODEX_AGENT_RUNTIME_BACKEND]: {
@@ -23622,110 +23614,6 @@ var MODEL_REGISTRY = [
     contextWindow: 4e5,
     reasoning: { efforts: ["low", "medium", "high", "xhigh"], default: "high" },
     pricing: { inputPerMillion: 1.75, outputPerMillion: 14, cacheReadPerMillion: 0.175 }
-  },
-  {
-    id: DeepSeekModel.V4Flash,
-    name: "DeepSeek V4 Flash",
-    // `provider` names the WIRE PROTOCOL the gateway speaks (OpenAI Responses),
-    // not the vendor. DeepSeek V4 is served by Baseten: the gateway maps these
-    // registry ids to Baseten wire ids and swaps the upstream host, while every
-    // request/response stays OpenAI-shaped. Do NOT "fix" this to a new
-    // "baseten" provider — `ModelProvider` is what review-model validation, the
-    // credential gate (`PROVIDER_ENV_VAR` / spawn credential resolution), and
-    // the sandbox bridge's model guard all key off, so a new provider value
-    // silently fails those closed instead of routing anywhere.
-    provider: "openai",
-    backends: [CODEX_AGENT_RUNTIME_BACKEND],
-    codexProviderRoute: "baseten",
-    // Required for codex session-start eligibility: the codex backend is the
-    // sole consumer of this flag and drops any model without it from the
-    // session-start set (see modelSupportsRequiredSessionStartCapabilities).
-    capabilities: { codexToolSearch: true },
-    contextWindow: 1048576,
-    // Only the effort our live Baseten probes actually proved. Registry efforts
-    // are treated as supported capabilities and are selectable by other paths
-    // (isValidReasoningEffort, review config, CLI overrides), so widen this list
-    // only after probing each added value end-to-end against Baseten.
-    reasoning: { efforts: ["high"], default: "high" },
-    // Verified against https://www.baseten.co/pricing on 2026-08-11 (Model APIs
-    // token pricing, row `deepseek-ai/DeepSeek-V4-Flash-0731`): $0.13/M input,
-    // $0.028/M cached input, $0.26/M output. Baseten publishes a cached-input
-    // rate for this model, so cacheReadPerMillion carries it rather than
-    // mirroring the input rate. No longContext and no flex: both are OpenAI-only
-    // pricing axes and Baseten prices a single flat tier.
-    pricing: { inputPerMillion: 0.13, outputPerMillion: 0.26, cacheReadPerMillion: 0.028 },
-    sessionStart: { eligible: true },
-    visibility: "internal_probe"
-    // Deliberately no overloadFallback. A Baseten 429 must not silently re-run
-    // an A/B arm on a different model; with no fallback the loop retries the
-    // same model, which is what an A/B comparison requires.
-  },
-  {
-    id: DeepSeekModel.V4Pro,
-    name: "DeepSeek V4 Pro",
-    // See the DeepSeek V4 Flash entry above: "openai" is the wire protocol
-    // (OpenAI Responses), not the vendor. Changing it to a "baseten" provider
-    // breaks review-model validation, the credential gate, and the sandbox
-    // bridge's model guard.
-    provider: "openai",
-    backends: [CODEX_AGENT_RUNTIME_BACKEND],
-    codexProviderRoute: "baseten",
-    // Required for codex session-start eligibility (sole consumer of the flag).
-    capabilities: { codexToolSearch: true },
-    contextWindow: 262144,
-    // Only the probed effort; widen only after probing each value against
-    // Baseten, since registry efforts are selectable capabilities.
-    reasoning: { efforts: ["high"], default: "high" },
-    // Verified against https://www.baseten.co/pricing on 2026-08-11 (Model APIs
-    // token pricing, row `DeepSeek V4 Pro`): $1.74/M input, $0.145/M cached
-    // input, $3.48/M output. Cached-input rate is published, so it is carried
-    // verbatim. No longContext and no flex (OpenAI-only axes).
-    pricing: { inputPerMillion: 1.74, outputPerMillion: 3.48, cacheReadPerMillion: 0.145 },
-    sessionStart: { eligible: true },
-    visibility: "internal_probe"
-    // Deliberately no overloadFallback — see DeepSeek V4 Flash: an A/B arm must
-    // retry the same model on a Baseten 429, never silently swap models.
-  },
-  {
-    id: GlmModel.Flash53,
-    name: "GLM-5.3 Flash",
-    // Same contract as the DeepSeek V4 entries: "openai" is the wire protocol
-    // (OpenAI Responses), not the vendor. GLM-5.3-Flash is served by Baseten.
-    provider: "openai",
-    backends: [CODEX_AGENT_RUNTIME_BACKEND],
-    codexProviderRoute: "baseten",
-    capabilities: { codexToolSearch: true },
-    contextWindow: 1e6,
-    // Only the effort the live streamed Baseten probe proved on 2026-08-27.
-    reasoning: { efforts: ["medium"], default: "medium" },
-    // Verified against https://www.baseten.co/pricing on 2026-08-27 (Model APIs
-    // token pricing, row `zai-org/GLM-5.3-Flash`): $0.15/M input, $0.03/M cached
-    // input, $0.50/M output. Cached-input rate is published, so it is carried
-    // verbatim. No longContext and no flex (OpenAI-only axes).
-    pricing: { inputPerMillion: 0.15, outputPerMillion: 0.5, cacheReadPerMillion: 0.03 },
-    sessionStart: { eligible: true },
-    visibility: "internal_probe"
-  },
-  {
-    id: GlmModel.G53,
-    name: "GLM-5.3",
-    // Same contract as the DeepSeek V4 entries: "openai" is the wire protocol
-    // (OpenAI Responses), not the vendor. GLM-5.3 is served by Baseten.
-    provider: "openai",
-    backends: [CODEX_AGENT_RUNTIME_BACKEND],
-    codexProviderRoute: "baseten",
-    capabilities: { codexToolSearch: true },
-    contextWindow: 1e6,
-    // Official GLM-5.3 efforts are low / high / max. Do not list medium.
-    // Widen only after probing each value against Baseten.
-    reasoning: { efforts: ["high"], default: "high" },
-    // Verified against https://www.baseten.co/pricing on 2026-08-28 (Model APIs
-    // token pricing, row `zai-org/GLM-5.3`): $1.40/M input, $0.14/M cached
-    // input, $4.40/M output. Cached-input rate is published, so it is carried
-    // verbatim. No longContext and no flex (OpenAI-only axes).
-    pricing: { inputPerMillion: 1.4, outputPerMillion: 4.4, cacheReadPerMillion: 0.14 },
-    sessionStart: { eligible: true },
-    visibility: "internal_probe"
   }
 ];
 var MODEL_PROVIDER_NAMES = {
@@ -26229,6 +26117,7 @@ var ERROR_CODES = [
   "policy_block",
   "handled_automatically",
   "review_publish_blocked",
+  "review_finalization_failed",
   "output_length",
   "context_overflow",
   "aborted",
@@ -26278,6 +26167,7 @@ var ERROR_CODE_LABELS = {
   policy_block: "Policy block",
   handled_automatically: "Handled automatically",
   review_publish_blocked: "Review publication blocked",
+  review_finalization_failed: "Review finalization failed",
   output_length: "Output was too long",
   context_overflow: "Context window exceeded",
   aborted: "Stopped by user",

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@tryarcanist/cli",
-  "version": "0.1.335",
+  "version": "0.1.337",
   "description": "CLI for Arcanist - create and manage coding agent sessions",
   "type": "module",
   "bin": {