@mars-sea/dsh-commandcode-provider 0.11.14 → 0.11.16

package/lib/index.d.ts

@@ -7,7 +7,7 @@ import { Context } from "@deepseek-ai/cordis";
 import { AttachmentStore } from "@deepseek-ai/dsh-attachment";
 import { CommandDefinition } from "@deepseek-ai/dsh-commands";
 //#region src/adapter.d.ts
-declare const COMMAND_CODE_CLI_VERSION = "1.65.0";
+declare const COMMAND_CODE_CLI_VERSION = "1.65.2";
 declare const DEFAULT_API_BASE = "https://api.commandcode.ai";
 declare const DEFAULT_GENERATE_MAX_TOKENS = 64000;
 declare const DEFAULT_MAX_OUTPUT_TOKENS = 65536;
@@ -789,8 +789,9 @@ declare const KNOWN_EFFORTS: Readonly<Record<string, readonly string[]>>;
 declare const KNOWN_IMAGE_MODELS: ReadonlySet<string>;
 /**
  * Models WITHOUT a zero-data-retention upstream, per the official CLI's own
- * registry (`command-code@1.65.0` `dist/cli.mjs`: `modelSupportsZdr(id)` is
- * exactly `!nonZdrSet.has(canonicalize(id))`, and `knownModelSupportsZdr`
+ * registry (`command-code@1.65.2` `dist/cli.mjs`, unchanged from 1.65.0:
+ * `modelSupportsZdr(id)` is exactly `!nonZdrSet.has(canonicalize(id))`, and
+ * `knownModelSupportsZdr`
  * carries the same membership in the sibling route table — the union is this
  * set). The official docs (commandcode.ai/docs/resources/zdr) put it in prose
  * — "99% of our models have ZDR-capable upstreams … only a small handful of

package/lib/index.js

@@ -395,8 +395,9 @@ const KNOWN_IMAGE_MODELS = /* @__PURE__ */ new Set([
 ]);
 /**
 * Models WITHOUT a zero-data-retention upstream, per the official CLI's own
-* registry (`command-code@1.65.0` `dist/cli.mjs`: `modelSupportsZdr(id)` is
-* exactly `!nonZdrSet.has(canonicalize(id))`, and `knownModelSupportsZdr`
+* registry (`command-code@1.65.2` `dist/cli.mjs`, unchanged from 1.65.0:
+* `modelSupportsZdr(id)` is exactly `!nonZdrSet.has(canonicalize(id))`, and
+* `knownModelSupportsZdr`
 * carries the same membership in the sibling route table — the union is this
 * set). The official docs (commandcode.ai/docs/resources/zdr) put it in prose
 * — "99% of our models have ZDR-capable upstreams … only a small handful of
@@ -3004,20 +3005,19 @@ function openStreamTrace(scope, options = {}) {
 * and API key or subscription, and Command Code's terms apply.
 *
 * Wire protocol (reverse-engineered by the pi plugin, command-code@1.28.4;
-* re-verified against command-code@1.62.0 — endpoints, request shape, and
-* stream events unchanged):
+* re-verified through command-code@1.65.2):
 *   POST {apiBase}/alpha/generate
 *   body: { config, memory, taste, skills, params: { model, messages, tools,
 *          system, max_tokens, temperature, stream, reasoning_effort? }, threadId }
 *   SSE-ish JSONL events: text-delta | reasoning-start/delta/end | tool-call
-*                         | tool-result | finish | error
+*                         | tool-result | cache-write-tokens | finish | error
 *   Model catalog: GET {apiBase}/provider/v1/models -> { object: 'list', data: [...] }
 *
 * The adapter is deliberately free of cordis/schemastery: it receives a
 * per-request options thunk and an API-key resolver from the plugin entry
 * (src/index.ts), so a settings change reaches the very next request.
 */
-const COMMAND_CODE_CLI_VERSION = "1.65.0";
+const COMMAND_CODE_CLI_VERSION = "1.65.2";
 const DEFAULT_API_BASE = "https://api.commandcode.ai";
 const DEFAULT_GENERATE_MAX_TOKENS = 64e3;
 const DEFAULT_MAX_OUTPUT_TOKENS = 65536;
@@ -4102,6 +4102,7 @@ function createBlockAssembler() {
 		reasoningContent: "",
 		sawContent: false,
 		finishReason: void 0,
+		cliCacheWriteTokens: void 0,
 		openAiToolCalls: []
 	};
 }
@@ -4304,6 +4305,7 @@ var CommandCodeAdapter = class extends LlmAdapter {
 			name: entry ? `${entry.name} (CC)` : model,
 			description: capabilityDescription(model, entry?.contextWindow),
 			inputModalities: vision ? ["text", "image"] : ["text"],
+			toolUpdate: "addition-only",
 			...entry ? {
 				context: { contextWindow: entry.contextWindow },
 				defaultMaxTokens: Math.min(entry.maxTokens, DEFAULT_GENERATE_MAX_TOKENS)
@@ -5029,22 +5031,29 @@ function handleCliEvent(asm, event) {
 			});
 			break;
 		}
+		case "cache-write-tokens": {
+			const cacheWrite = numberValue(event.cacheWriteTokens);
+			if (cacheWrite !== void 0 && cacheWrite >= 0) asm.cliCacheWriteTokens = cacheWrite;
+			break;
+		}
 		case "finish": {
 			asm.finishReason = stringValue(event.finishReason);
 			chunks.push(...closeText(asm), ...closeReasoning(asm));
 			const usage = isRecord(event.totalUsage) ? event.totalUsage : void 0;
-			if (usage) {
-				const details = isRecord(usage.inputTokenDetails) ? usage.inputTokenDetails : void 0;
-				const totalInput = numberValue(usage.inputTokens) ?? 0;
+			const capturedCacheWrite = asm.cliCacheWriteTokens;
+			if (usage || capturedCacheWrite !== void 0) {
+				const details = isRecord(usage?.inputTokenDetails) ? usage.inputTokenDetails : void 0;
+				const totalInput = numberValue(usage?.inputTokens) ?? 0;
 				const cacheRead = numberValue(details?.cacheReadTokens) ?? 0;
-				const cacheWrite = numberValue(details?.cacheWriteTokens) ?? 0;
+				const reportedCacheWrite = numberValue(details?.cacheWriteTokens) ?? 0;
+				const cacheWrite = reportedCacheWrite === 0 && capturedCacheWrite !== void 0 ? capturedCacheWrite : reportedCacheWrite;
 				const tokenUsage = {
 					inputTokens: numberValue(details?.noCacheTokens) ?? Math.max(0, totalInput - cacheRead - cacheWrite),
-					outputTokens: numberValue(usage.outputTokens) ?? 0,
+					outputTokens: numberValue(usage?.outputTokens) ?? 0,
 					cacheReadTokens: cacheRead,
 					cacheWriteTokens: cacheWrite
 				};
-				const reasoningTokens = numberValue((isRecord(usage.outputTokenDetails) ? usage.outputTokenDetails : void 0)?.reasoningTokens) ?? numberValue(usage.reasoningTokens);
+				const reasoningTokens = numberValue((isRecord(usage?.outputTokenDetails) ? usage.outputTokenDetails : void 0)?.reasoningTokens) ?? numberValue(usage?.reasoningTokens);
 				if (reasoningTokens !== void 0) tokenUsage.reasoningTokens = reasoningTokens;
 				chunks.push({
 					type: "usage",
@@ -7277,7 +7286,7 @@ function transportBudgetMessage(failureMessage, maxRetries) {
 * as an internal advisor instead: the command guard in `./command-guard.ts`
 * asks it whether a shell command is safe to auto-approve.
 *
-* The three question types are the whole vocabulary (command-code@1.65.0
+* The three question types are the whole vocabulary (command-code@1.65.2
 * `systemOneRequestSchema`, unchanged since 1.64.0):
 *
 * - `noul` — "is this true?", optional `criteria: { true, false }` describing
@@ -8123,7 +8132,7 @@ function applyCommandGuard(ctx, deps) {
 * Mark one schema field volatile.
 *
 * `.volatile()` is unconditional: the mark is a schemastery 3.18.3 feature and
-* the only engine this bundle supports (`dsh 0.1.7-rc.1`) pins `~3.18.4`, so
+* the only engine this bundle supports (`dsh 0.1.7-rc.2`) pins `~3.18.4`, so
 * the method is always present. It is reached through a structural member
 * rather than schemastery's own typing so the function can keep the caller's
 * declared field type, which is what lets `Config` stay the plain shape every