@skein-code/cli 0.3.21 → 0.3.22

package/README.md

@@ -57,8 +57,10 @@ inspectable index for code retrieval and no retrieval service dependency.
   artifact for bounded readback.
 - **Measurable token economy:** every model request records a privacy-safe
   receipt for stable, dynamic, history, retrieval, tool-result, and tool-schema
-  estimates, alongside actual provider usage when available. Receipts retain
-  only counts and runtime decisions, never prompt or source content.
+  estimates, alongside actual provider usage when available. OpenAI, Anthropic,
+  and Gemini cache/reasoning counters are normalized into session totals and
+  structured events when the provider reports them. Receipts retain only counts
+  and runtime decisions, never prompt or source content.
 - **Reversible work:** Skein snapshots affected files before mutation without
   touching your Git history.
 - **Resumable by default:** conversations, tasks, usage, and changed files live
@@ -94,7 +96,7 @@ To build, verify, and install a local package artifact from this checkout:
 
 ```bash
 npm run verify:package -- --output-dir artifacts/package
-npm install -g ./artifacts/package/skein-code-cli-0.3.21.tgz
+npm install -g ./artifacts/package/skein-code-cli-0.3.22.tgz
 ```
 
 To install the published package from npm:

package/dist/cli.js

@@ -220,7 +220,7 @@ function isSqliteBusy(error) {
 // package.json
 var package_default = {
   name: "@skein-code/cli",
-  version: "0.3.21",
+  version: "0.3.22",
   description: "A context-first, model-agnostic coding agent with an auditable terminal workspace.",
   type: "module",
   license: "MIT",
@@ -236,6 +236,9 @@ var package_default = {
   },
   skein: {
     releaseNotes: [
+      "OpenAI, Anthropic, and Gemini usage now normalizes cached input, cache-write input, and reasoning tokens when reported",
+      "Token Ledger, session totals, and structured usage events preserve provider cache and reasoning counts without prompt content",
+      "Streaming, non-streaming, explicit-zero, and legacy-session compatibility paths are covered without claiming cache activation",
       "Failed configured verification output yields bounded current-run diagnostic ranking without persisting paths",
       "Diagnostic hints cannot create zero-relevance hits and are cleared on success or the next agent run",
       "Python absolute and relative module imports now contribute deterministic offline graph adjacency",
@@ -5349,10 +5352,7 @@ var AnthropicProvider = class {
         name: block.name ?? "unknown",
         arguments: safeJsonArguments(block.input)
       })),
-      usage: {
-        ...data.usage?.input_tokens !== void 0 ? { inputTokens: data.usage.input_tokens } : {},
-        ...data.usage?.output_tokens !== void 0 ? { outputTokens: data.usage.output_tokens } : {}
-      },
+      usage: normalizeAnthropicUsage(data.usage),
       ...data.stop_reason ? { stopReason: data.stop_reason } : {}
     };
   }
@@ -5393,6 +5393,8 @@ var AnthropicProvider = class {
     let content = "";
     let inputTokens;
     let outputTokens;
+    let cachedInputTokens;
+    let cacheWriteInputTokens;
     let stopReason;
     const calls = /* @__PURE__ */ new Map();
     for await (const event of parseServerSentEvents(response)) {
@@ -5402,6 +5404,13 @@ var AnthropicProvider = class {
       }
       if (chunk.usage?.input_tokens !== void 0) inputTokens = chunk.usage.input_tokens;
       if (chunk.usage?.output_tokens !== void 0) outputTokens = chunk.usage.output_tokens;
+      const usage = chunk.usage ?? chunk.message?.usage;
+      if (usage?.cache_read_input_tokens !== void 0) {
+        cachedInputTokens = usage.cache_read_input_tokens;
+      }
+      if (usage?.cache_creation_input_tokens !== void 0) {
+        cacheWriteInputTokens = usage.cache_creation_input_tokens;
+      }
       if (chunk.delta?.stop_reason) stopReason = chunk.delta.stop_reason;
       const index = chunk.index ?? 0;
       if (chunk.type === "content_block_start" && chunk.content_block?.type === "text") {
@@ -5440,7 +5449,9 @@ var AnthropicProvider = class {
         })),
         usage: {
           ...inputTokens !== void 0 ? { inputTokens } : {},
-          ...outputTokens !== void 0 ? { outputTokens } : {}
+          ...outputTokens !== void 0 ? { outputTokens } : {},
+          ...cachedInputTokens !== void 0 ? { cachedInputTokens } : {},
+          ...cacheWriteInputTokens !== void 0 ? { cacheWriteInputTokens } : {}
         },
         ...stopReason ? { stopReason } : {}
       }
@@ -5456,13 +5467,18 @@ function normalizeAnthropicResponse(data) {
       name: block.name ?? "unknown",
       arguments: safeJsonArguments(block.input)
     })),
-    usage: {
-      ...data.usage?.input_tokens !== void 0 ? { inputTokens: data.usage.input_tokens } : {},
-      ...data.usage?.output_tokens !== void 0 ? { outputTokens: data.usage.output_tokens } : {}
-    },
+    usage: normalizeAnthropicUsage(data.usage),
     ...data.stop_reason ? { stopReason: data.stop_reason } : {}
   };
 }
+function normalizeAnthropicUsage(usage) {
+  return {
+    ...usage?.input_tokens !== void 0 ? { inputTokens: usage.input_tokens } : {},
+    ...usage?.output_tokens !== void 0 ? { outputTokens: usage.output_tokens } : {},
+    ...usage?.cache_read_input_tokens !== void 0 ? { cachedInputTokens: usage.cache_read_input_tokens } : {},
+    ...usage?.cache_creation_input_tokens !== void 0 ? { cacheWriteInputTokens: usage.cache_creation_input_tokens } : {}
+  };
+}
 function toAnthropicMessage(message2) {
   if (message2.role === "tool") {
     return {
@@ -5536,10 +5552,7 @@ var GeminiProvider = class {
         name: part.functionCall?.name ?? "unknown",
         arguments: safeJsonArguments(part.functionCall?.args)
       })),
-      usage: {
-        ...data.usageMetadata?.promptTokenCount !== void 0 ? { inputTokens: data.usageMetadata.promptTokenCount } : {},
-        ...data.usageMetadata?.candidatesTokenCount !== void 0 ? { outputTokens: data.usageMetadata.candidatesTokenCount } : {}
-      },
+      usage: normalizeGeminiUsage(data.usageMetadata),
       ...candidate?.finishReason ? { stopReason: candidate.finishReason } : {}
     };
   }
@@ -5579,6 +5592,8 @@ var GeminiProvider = class {
     let content = "";
     let inputTokens;
     let outputTokens;
+    let cachedInputTokens;
+    let reasoningTokens;
     let stopReason;
     const toolCalls = [];
     for await (const event of parseServerSentEvents(response)) {
@@ -5590,6 +5605,12 @@ var GeminiProvider = class {
       if (chunk.usageMetadata?.candidatesTokenCount !== void 0) {
         outputTokens = chunk.usageMetadata.candidatesTokenCount;
       }
+      if (chunk.usageMetadata?.cachedContentTokenCount !== void 0) {
+        cachedInputTokens = chunk.usageMetadata.cachedContentTokenCount;
+      }
+      if (chunk.usageMetadata?.thoughtsTokenCount !== void 0) {
+        reasoningTokens = chunk.usageMetadata.thoughtsTokenCount;
+      }
       const candidate = chunk.candidates?.[0];
       if (candidate?.finishReason) stopReason = candidate.finishReason;
       for (const part of candidate?.content?.parts ?? []) {
@@ -5613,7 +5634,9 @@ var GeminiProvider = class {
         toolCalls,
         usage: {
           ...inputTokens !== void 0 ? { inputTokens } : {},
-          ...outputTokens !== void 0 ? { outputTokens } : {}
+          ...outputTokens !== void 0 ? { outputTokens } : {},
+          ...cachedInputTokens !== void 0 ? { cachedInputTokens } : {},
+          ...reasoningTokens !== void 0 ? { reasoningTokens } : {}
         },
         ...stopReason ? { stopReason } : {}
       }
@@ -5630,13 +5653,18 @@ function normalizeGeminiResponse(data) {
       name: part.functionCall?.name ?? "unknown",
       arguments: safeJsonArguments(part.functionCall?.args)
     })),
-    usage: {
-      ...data.usageMetadata?.promptTokenCount !== void 0 ? { inputTokens: data.usageMetadata.promptTokenCount } : {},
-      ...data.usageMetadata?.candidatesTokenCount !== void 0 ? { outputTokens: data.usageMetadata.candidatesTokenCount } : {}
-    },
+    usage: normalizeGeminiUsage(data.usageMetadata),
     ...candidate?.finishReason ? { stopReason: candidate.finishReason } : {}
   };
 }
+function normalizeGeminiUsage(usage) {
+  return {
+    ...usage?.promptTokenCount !== void 0 ? { inputTokens: usage.promptTokenCount } : {},
+    ...usage?.candidatesTokenCount !== void 0 ? { outputTokens: usage.candidatesTokenCount } : {},
+    ...usage?.cachedContentTokenCount !== void 0 ? { cachedInputTokens: usage.cachedContentTokenCount } : {},
+    ...usage?.thoughtsTokenCount !== void 0 ? { reasoningTokens: usage.thoughtsTokenCount } : {}
+  };
+}
 function toGeminiMessage(message2) {
   if (message2.role === "tool") {
     return {
@@ -5764,6 +5792,8 @@ var OpenAIProvider = class {
     let content = "";
     let inputTokens;
     let outputTokens;
+    let cachedInputTokens;
+    let reasoningTokens;
     let stopReason;
     const calls = /* @__PURE__ */ new Map();
     for await (const event of parseServerSentEvents(response)) {
@@ -5771,6 +5801,12 @@ var OpenAIProvider = class {
       const chunk = JSON.parse(event.data);
       if (chunk.usage?.prompt_tokens !== void 0) inputTokens = chunk.usage.prompt_tokens;
       if (chunk.usage?.completion_tokens !== void 0) outputTokens = chunk.usage.completion_tokens;
+      if (chunk.usage?.prompt_tokens_details?.cached_tokens !== void 0) {
+        cachedInputTokens = chunk.usage.prompt_tokens_details.cached_tokens;
+      }
+      if (chunk.usage?.completion_tokens_details?.reasoning_tokens !== void 0) {
+        reasoningTokens = chunk.usage.completion_tokens_details.reasoning_tokens;
+      }
       const choice = chunk.choices?.[0];
       if (choice?.finish_reason) stopReason = choice.finish_reason;
       const delta = choice?.delta;
@@ -5798,7 +5834,9 @@ var OpenAIProvider = class {
         })),
         usage: {
           ...inputTokens !== void 0 ? { inputTokens } : {},
-          ...outputTokens !== void 0 ? { outputTokens } : {}
+          ...outputTokens !== void 0 ? { outputTokens } : {},
+          ...cachedInputTokens !== void 0 ? { cachedInputTokens } : {},
+          ...reasoningTokens !== void 0 ? { reasoningTokens } : {}
         },
         ...stopReason ? { stopReason } : {}
       }
@@ -5816,13 +5854,18 @@ function normalizeOpenAIResponse(data) {
       name: call.function?.name ?? "unknown",
       arguments: safeJsonArguments(call.function?.arguments)
     })),
-    usage: {
-      ...data.usage?.prompt_tokens !== void 0 ? { inputTokens: data.usage.prompt_tokens } : {},
-      ...data.usage?.completion_tokens !== void 0 ? { outputTokens: data.usage.completion_tokens } : {}
-    },
+    usage: normalizeOpenAIUsage(data.usage),
     ...choice?.finish_reason ? { stopReason: choice.finish_reason } : {}
   };
 }
+function normalizeOpenAIUsage(usage) {
+  return {
+    ...usage?.prompt_tokens !== void 0 ? { inputTokens: usage.prompt_tokens } : {},
+    ...usage?.completion_tokens !== void 0 ? { outputTokens: usage.completion_tokens } : {},
+    ...usage?.prompt_tokens_details?.cached_tokens !== void 0 ? { cachedInputTokens: usage.prompt_tokens_details.cached_tokens } : {},
+    ...usage?.completion_tokens_details?.reasoning_tokens !== void 0 ? { reasoningTokens: usage.completion_tokens_details.reasoning_tokens } : {}
+  };
+}
 function toOpenAIMessage(message2) {
   if (message2.role === "tool") {
     return {
@@ -6399,6 +6442,9 @@ var sessionSchema = z5.object({
     outputSource: z5.enum(["actual", "estimated", "mixed", "unknown"]).optional(),
     actualInputTokens: z5.number().nonnegative().optional(),
     actualOutputTokens: z5.number().nonnegative().optional(),
+    actualCachedInputTokens: z5.number().nonnegative().optional(),
+    actualCacheWriteInputTokens: z5.number().nonnegative().optional(),
+    actualReasoningTokens: z5.number().nonnegative().optional(),
     estimatedInputTokens: z5.number().nonnegative().optional(),
     estimatedOutputTokens: z5.number().nonnegative().optional()
   }).strict(),
@@ -6419,7 +6465,10 @@ var sessionSchema = z5.object({
     }).strict(),
     actual: z5.object({
       inputTokens: z5.number().nonnegative().optional(),
-      outputTokens: z5.number().nonnegative().optional()
+      outputTokens: z5.number().nonnegative().optional(),
+      cachedInputTokens: z5.number().nonnegative().optional(),
+      cacheWriteInputTokens: z5.number().nonnegative().optional(),
+      reasoningTokens: z5.number().nonnegative().optional()
     }).strict(),
     inputSource: z5.enum(["actual", "estimated"]),
     outputSource: z5.enum(["actual", "estimated"]),
@@ -10830,6 +10879,9 @@ var AgentRunner = class {
         const { inputTokens, outputTokens } = turnUsage;
         const actualInputTokens = validTokenCount(response.usage?.inputTokens);
         const actualOutputTokens = validTokenCount(response.usage?.outputTokens);
+        const actualCachedInputTokens = validTokenCount(response.usage?.cachedInputTokens);
+        const actualCacheWriteInputTokens = validTokenCount(response.usage?.cacheWriteInputTokens);
+        const actualReasoningTokens = validTokenCount(response.usage?.reasoningTokens);
         const receipt = recordTokenLedger(this.session, {
           requestId: userMessage.id,
           turn,
@@ -10840,7 +10892,10 @@ var AgentRunner = class {
           },
           actual: {
             ...actualInputTokens === void 0 ? {} : { inputTokens: actualInputTokens },
-            ...actualOutputTokens === void 0 ? {} : { outputTokens: actualOutputTokens }
+            ...actualOutputTokens === void 0 ? {} : { outputTokens: actualOutputTokens },
+            ...actualCachedInputTokens === void 0 ? {} : { cachedInputTokens: actualCachedInputTokens },
+            ...actualCacheWriteInputTokens === void 0 ? {} : { cacheWriteInputTokens: actualCacheWriteInputTokens },
+            ...actualReasoningTokens === void 0 ? {} : { reasoningTokens: actualReasoningTokens }
           },
           inputSource: actualInputTokens === void 0 ? "estimated" : "actual",
           outputSource: actualOutputTokens === void 0 ? "estimated" : "actual",
@@ -10859,7 +10914,10 @@ var AgentRunner = class {
           outputSource: this.session.usage.outputSource ?? "unknown",
           actual: {
             inputTokens: this.session.usage.actualInputTokens ?? 0,
-            outputTokens: this.session.usage.actualOutputTokens ?? 0
+            outputTokens: this.session.usage.actualOutputTokens ?? 0,
+            ...this.session.usage.actualCachedInputTokens === void 0 ? {} : { cachedInputTokens: this.session.usage.actualCachedInputTokens },
+            ...this.session.usage.actualCacheWriteInputTokens === void 0 ? {} : { cacheWriteInputTokens: this.session.usage.actualCacheWriteInputTokens },
+            ...this.session.usage.actualReasoningTokens === void 0 ? {} : { reasoningTokens: this.session.usage.actualReasoningTokens }
           },
           estimated: {
             inputTokens: this.session.usage.estimatedInputTokens ?? 0,
@@ -11560,6 +11618,18 @@ function recordTokenUsage(session, providerUsage, estimatedInputTokens, estimate
   } else {
     session.usage.estimatedOutputTokens = (session.usage.estimatedOutputTokens ?? 0) + outputTokens;
   }
+  const cachedInputActual = validTokenCount(providerUsage?.cachedInputTokens);
+  const cacheWriteInputActual = validTokenCount(providerUsage?.cacheWriteInputTokens);
+  const reasoningActual = validTokenCount(providerUsage?.reasoningTokens);
+  if (cachedInputActual !== void 0) {
+    session.usage.actualCachedInputTokens = (session.usage.actualCachedInputTokens ?? 0) + cachedInputActual;
+  }
+  if (cacheWriteInputActual !== void 0) {
+    session.usage.actualCacheWriteInputTokens = (session.usage.actualCacheWriteInputTokens ?? 0) + cacheWriteInputActual;
+  }
+  if (reasoningActual !== void 0) {
+    session.usage.actualReasoningTokens = (session.usage.actualReasoningTokens ?? 0) + reasoningActual;
+  }
   session.usage.inputSource = mergeMeasurementSource(
     priorInputSource,
     inputActual === void 0 ? "estimated" : "actual",