docs(llm): fix stale references in protocols/shared.ts and gemini.ts

- subtractTokens JSDoc said the raw payload lives on Usage.native, but that field was renamed to providerMetadata earlier in this PR. - totalTokens JSDoc still described the abandoned "additive" first-pass contract where inputTokens/outputTokens were non-cached / visible only. We landed on inclusive totals; the fallback already covers cache and reasoning. - Removed a duplicate inline comment in Gemini's mapUsage — the function-level comment already explains the visible/reasoning sum and the undefined-when-incomplete rule.
test(llm): re-record golden scenarios against live providers
2026-08-10 03:19:59 -04:00 · 2026-05-10 22:08:12 -04:00 · 2026-05-10 22:03:02 -04:00 · 2026-05-10 21:42:09 -04:00 · 2026-05-10 20:39:22 -04:00 · 2026-05-10 19:29:41 -04:00
46 changed files with 2198 additions and 882 deletions
@@ -53,7 +53,7 @@ export function UsageSection() {
  }

  const calculateTotalOutputTokens = (u: Awaited<ReturnType<typeof getUsageInfo>>[0]) => {
-    return u.outputTokens
+    return u.outputTokens + (u.reasoningTokens ?? 0)
  }

  const goPrev = async () => {
@@ -889,6 +889,10 @@ export async function handler(

    const inputCost = modelCost.input * inputTokens * 100
    const outputCost = modelCost.output * outputTokens * 100
+    const reasoningCost = (() => {
+      if (!reasoningTokens) return undefined
+      return modelCost.output * reasoningTokens * 100
+    })()
    const cacheReadCost = (() => {
      if (!cacheReadTokens) return undefined
      if (!modelCost.cacheRead) return undefined
@@ -905,11 +909,17 @@ export async function handler(
      return modelCost.cacheWrite1h * cacheWrite1hTokens * 100
    })()
    const totalCostInCent =
-      inputCost + outputCost + (cacheReadCost ?? 0) + (cacheWrite5mCost ?? 0) + (cacheWrite1hCost ?? 0)
+      inputCost +
+      outputCost +
+      (reasoningCost ?? 0) +
+      (cacheReadCost ?? 0) +
+      (cacheWrite5mCost ?? 0) +
+      (cacheWrite1hCost ?? 0)
    return {
      totalCostInCent,
      inputCost,
      outputCost,
+      reasoningCost,
      cacheReadCost,
      cacheWrite5mCost,
      cacheWrite1hCost,
@@ -931,7 +941,8 @@ export async function handler(
  ) {
    const { inputTokens, outputTokens, reasoningTokens, cacheReadTokens, cacheWrite5mTokens, cacheWrite1hTokens } =
      usageInfo
-    const { totalCostInCent, inputCost, outputCost, cacheReadCost, cacheWrite5mCost, cacheWrite1hCost } = costInfo
+    const { totalCostInCent, inputCost, outputCost, reasoningCost, cacheReadCost, cacheWrite5mCost, cacheWrite1hCost } =
+      costInfo

    logger.metric({
      "tokens.input": inputTokens,
@@ -942,12 +953,14 @@ export async function handler(
      "tokens.cache_write_1h": cacheWrite1hTokens,
      "cost.input.microcents": centsToMicroCents(inputCost),
      "cost.output.microcents": centsToMicroCents(outputCost),
+      "cost.reasoning.microcents": reasoningCost ? centsToMicroCents(reasoningCost) : undefined,
      "cost.cache_read.microcents": cacheReadCost ? centsToMicroCents(cacheReadCost) : undefined,
      "cost.cache_write.microcents": cacheWrite5mCost ? centsToMicroCents(cacheWrite5mCost) : undefined,
      "cost.total.microcents": centsToMicroCents(totalCostInCent),
      // deprecated - remove after May 20, 2026
      "cost.input": Math.round(inputCost),
      "cost.output": Math.round(outputCost),
+      "cost.reasoning": reasoningCost ? Math.round(reasoningCost) : undefined,
      "cost.cache_read": cacheReadCost ? Math.round(cacheReadCost) : undefined,
      "cost.cache_write_5m": cacheWrite5mCost ? Math.round(cacheWrite5mCost) : undefined,
      "cost.cache_write_1h": cacheWrite1hCost ? Math.round(cacheWrite1hCost) : undefined,
@@ -50,7 +50,7 @@ export const openaiHelper: ProviderHelper = ({ workspaceID }) => ({
    const cacheReadTokens = usage.input_tokens_details?.cached_tokens ?? undefined
    return {
      inputTokens: inputTokens - (cacheReadTokens ?? 0),
-      outputTokens,
+      outputTokens: outputTokens - (reasoningTokens ?? 0),
      reasoningTokens,
      cacheReadTokens,
      cacheWrite5mTokens: undefined,
@@ -364,34 +364,56 @@ const mapFinishReason = (reason: string | null | undefined): FinishReason => {
  return "unknown"
 }

+// Anthropic reports the non-overlapping breakdown natively — its
+// `input_tokens` is the *non-cached* count per the Messages API docs, with
+// cache reads and writes as separate fields. We sum them to derive the
+// inclusive `inputTokens` the rest of the contract expects. Extended
+// thinking tokens are *not* broken out by Anthropic — they're billed as
+// part of `output_tokens`, so `reasoningTokens` stays `undefined` and
+// `outputTokens` carries the combined total.
 const mapUsage = (usage: AnthropicUsage | undefined): Usage | undefined => {
  if (!usage) return undefined
+  const nonCached = usage.input_tokens
+  const cacheRead = usage.cache_read_input_tokens ?? undefined
+  const cacheWrite = usage.cache_creation_input_tokens ?? undefined
+  const inputTokens = ProviderShared.sumTokens(nonCached, cacheRead, cacheWrite)
  return new Usage({
-    inputTokens: usage.input_tokens,
+    inputTokens,
    outputTokens: usage.output_tokens,
-    cacheReadInputTokens: usage.cache_read_input_tokens ?? undefined,
-    cacheWriteInputTokens: usage.cache_creation_input_tokens ?? undefined,
-    totalTokens: ProviderShared.totalTokens(usage.input_tokens, usage.output_tokens, undefined),
-    native: usage,
+    nonCachedInputTokens: nonCached,
+    cacheReadInputTokens: cacheRead,
+    cacheWriteInputTokens: cacheWrite,
+    totalTokens: ProviderShared.totalTokens(inputTokens, usage.output_tokens, undefined),
+    providerMetadata: { anthropic: usage },
  })
 }

 // Anthropic emits usage on `message_start` and again on `message_delta` — the
 // final delta carries the authoritative totals. Right-biased merge: each
-// field prefers `right` when defined, falls back to `left`. `totalTokens` is
-// recomputed from the merged input/output to stay consistent.
+// field prefers `right` when defined, falls back to `left`. `inputTokens` is
+// recomputed from the merged breakdown so the inclusive total stays
+// consistent with `nonCached + cacheRead + cacheWrite`.
 const mergeUsage = (left: Usage | undefined, right: Usage | undefined) => {
  if (!left) return right
  if (!right) return left
-  const inputTokens = right.inputTokens ?? left.inputTokens
+  const nonCachedInputTokens = right.nonCachedInputTokens ?? left.nonCachedInputTokens
+  const cacheReadInputTokens = right.cacheReadInputTokens ?? left.cacheReadInputTokens
+  const cacheWriteInputTokens = right.cacheWriteInputTokens ?? left.cacheWriteInputTokens
+  const inputTokens = ProviderShared.sumTokens(nonCachedInputTokens, cacheReadInputTokens, cacheWriteInputTokens)
  const outputTokens = right.outputTokens ?? left.outputTokens
  return new Usage({
    inputTokens,
    outputTokens,
-    cacheReadInputTokens: right.cacheReadInputTokens ?? left.cacheReadInputTokens,
-    cacheWriteInputTokens: right.cacheWriteInputTokens ?? left.cacheWriteInputTokens,
+    nonCachedInputTokens,
+    cacheReadInputTokens,
+    cacheWriteInputTokens,
    totalTokens: ProviderShared.totalTokens(inputTokens, outputTokens, undefined),
-    native: { ...left.native, ...right.native },
+    providerMetadata: {
+      anthropic: {
+        ...(left.providerMetadata?.["anthropic"] ?? {}),
+        ...(right.providerMetadata?.["anthropic"] ?? {}),
+      },
+    },
  })
 }

@@ -363,15 +363,22 @@ const mapFinishReason = (reason: string): FinishReason => {
  return "unknown"
 }

+// AWS Bedrock Converse reports `inputTokens` (inclusive total) with
+// `cacheReadInputTokens` and `cacheWriteInputTokens` as subsets. Pass
+// the total through and derive the non-cached breakdown. Bedrock does
+// not break reasoning out of `outputTokens` for any current model.
 const mapUsage = (usage: BedrockUsageSchema | undefined): Usage | undefined => {
  if (!usage) return undefined
+  const cacheTotal = (usage.cacheReadInputTokens ?? 0) + (usage.cacheWriteInputTokens ?? 0)
+  const nonCached = ProviderShared.subtractTokens(usage.inputTokens, cacheTotal)
  return new Usage({
    inputTokens: usage.inputTokens,
    outputTokens: usage.outputTokens,
-    totalTokens: ProviderShared.totalTokens(usage.inputTokens, usage.outputTokens, usage.totalTokens),
+    nonCachedInputTokens: nonCached,
    cacheReadInputTokens: usage.cacheReadInputTokens,
    cacheWriteInputTokens: usage.cacheWriteInputTokens,
-    native: usage,
+    totalTokens: ProviderShared.totalTokens(usage.inputTokens, usage.outputTokens, usage.totalTokens),
+    providerMetadata: { bedrock: usage },
  })
 }

@@ -281,15 +281,28 @@ const fromRequest = Effect.fn("Gemini.fromRequest")(function* (request: LLMReque
 // =============================================================================
 // Stream Parsing
 // =============================================================================
+// Gemini reports `promptTokenCount` (inclusive total) with a
+// `cachedContentTokenCount` subset. `candidatesTokenCount` is *exclusive*
+// of `thoughtsTokenCount` — visible-only, not a total — so we sum the two
+// to produce the inclusive `outputTokens` the rest of the contract expects.
+// Output is left undefined when the visible component is missing, so we
+// don't fabricate an inclusive number from a partial breakdown.
 const mapUsage = (usage: GeminiUsage | undefined) => {
  if (!usage) return undefined
+  const cached = usage.cachedContentTokenCount
+  const nonCached = ProviderShared.subtractTokens(usage.promptTokenCount, cached)
+  const outputTokens =
+    usage.candidatesTokenCount !== undefined
+      ? usage.candidatesTokenCount + (usage.thoughtsTokenCount ?? 0)
+      : undefined
  return new Usage({
    inputTokens: usage.promptTokenCount,
-    outputTokens: usage.candidatesTokenCount,
+    outputTokens,
+    nonCachedInputTokens: nonCached,
+    cacheReadInputTokens: cached,
    reasoningTokens: usage.thoughtsTokenCount,
-    cacheReadInputTokens: usage.cachedContentTokenCount,
-    totalTokens: ProviderShared.totalTokens(usage.promptTokenCount, usage.candidatesTokenCount, usage.totalTokenCount),
-    native: usage,
+    totalTokens: ProviderShared.totalTokens(usage.promptTokenCount, outputTokens, usage.totalTokenCount),
+    providerMetadata: { google: usage },
  })
 }

@@ -290,15 +290,24 @@ const mapFinishReason = (reason: string | null | undefined): FinishReason => {
  return "unknown"
 }

+// OpenAI Chat reports `prompt_tokens` (inclusive total) with a
+// `cached_tokens` subset, and `completion_tokens` (inclusive total) with
+// a `reasoning_tokens` subset. We pass the inclusive totals through and
+// derive the non-cached breakdown so the `LLM.Usage` contract is
+// satisfied on both sides.
 const mapUsage = (usage: OpenAIChatEvent["usage"]): Usage | undefined => {
  if (!usage) return undefined
+  const cached = usage.prompt_tokens_details?.cached_tokens
+  const reasoning = usage.completion_tokens_details?.reasoning_tokens
+  const nonCached = ProviderShared.subtractTokens(usage.prompt_tokens, cached)
  return new Usage({
    inputTokens: usage.prompt_tokens,
    outputTokens: usage.completion_tokens,
-    reasoningTokens: usage.completion_tokens_details?.reasoning_tokens,
-    cacheReadInputTokens: usage.prompt_tokens_details?.cached_tokens,
+    nonCachedInputTokens: nonCached,
+    cacheReadInputTokens: cached,
+    reasoningTokens: reasoning,
    totalTokens: ProviderShared.totalTokens(usage.prompt_tokens, usage.completion_tokens, usage.total_tokens),
-    native: usage,
+    providerMetadata: { openai: usage },
  })
 }

@@ -276,15 +276,23 @@ const fromRequest = Effect.fn("OpenAIResponses.fromRequest")(function* (request:
 // =============================================================================
 // Stream Parsing
 // =============================================================================
+// OpenAI Responses reports `input_tokens` (inclusive total) with a
+// `cached_tokens` subset, and `output_tokens` (inclusive total) with a
+// `reasoning_tokens` subset. Pass the totals through and derive the
+// non-cached breakdown.
 const mapUsage = (usage: OpenAIResponsesUsage | null | undefined) => {
  if (!usage) return undefined
+  const cached = usage.input_tokens_details?.cached_tokens
+  const reasoning = usage.output_tokens_details?.reasoning_tokens
+  const nonCached = ProviderShared.subtractTokens(usage.input_tokens, cached)
  return new Usage({
    inputTokens: usage.input_tokens,
    outputTokens: usage.output_tokens,
-    reasoningTokens: usage.output_tokens_details?.reasoning_tokens,
-    cacheReadInputTokens: usage.input_tokens_details?.cached_tokens,
+    nonCachedInputTokens: nonCached,
+    cacheReadInputTokens: cached,
+    reasoningTokens: reasoning,
    totalTokens: ProviderShared.totalTokens(usage.input_tokens, usage.output_tokens, usage.total_tokens),
-    native: usage,
+    providerMetadata: { openai: usage },
  })
 }

@@ -42,6 +42,11 @@ export interface ToolAccumulator {
 * supplied total; otherwise falls back to `inputTokens + outputTokens` only
 * when at least one is defined. Returns `undefined` when neither input nor
 * output is known so routes don't publish a misleading `0`.
+ *
+ * Under the `LLM.Usage` contract, `inputTokens` and `outputTokens` are
+ * inclusive totals, so the computed fallback already covers cache reads /
+ * writes and reasoning — used mainly for Anthropic-style providers that
+ * don't surface a top-level total.
 */
 export const totalTokens = (
  inputTokens: number | undefined,
@@ -53,6 +58,39 @@ export const totalTokens = (
  return (inputTokens ?? 0) + (outputTokens ?? 0)
 }

+/**
+ * Subtract `subtrahend` from `total`, clamping to zero if the provider
+ * reports a non-sensical breakdown (e.g. `cached_tokens > prompt_tokens`).
+ * Used by protocol mappers when deriving a non-overlapping breakdown field
+ * from a provider's inclusive total — `nonCachedInputTokens` from
+ * `inputTokens - cacheReadInputTokens - cacheWriteInputTokens`.
+ *
+ * If `total` is `undefined`, returns `undefined` (we don't fabricate
+ * counts). If `subtrahend` is `undefined`, returns `total` unchanged. The
+ * provider-native breakdown stays available on `Usage.providerMetadata`
+ * for debugging.
+ */
+export const subtractTokens = (
+  total: number | undefined,
+  subtrahend: number | undefined,
+): number | undefined => {
+  if (total === undefined) return undefined
+  if (subtrahend === undefined) return total
+  return Math.max(0, total - subtrahend)
+}
+
+/**
+ * Sum a list of optional token counts, returning `undefined` only when
+ * every value is `undefined` (so we don't fabricate a `0`). Used by
+ * protocol mappers to derive the inclusive `inputTokens` total from a
+ * provider that natively reports a non-overlapping breakdown
+ * (e.g. Anthropic, whose `input_tokens` is already non-cached only).
+ */
+export const sumTokens = (...values: ReadonlyArray<number | undefined>): number | undefined => {
+  if (values.every((value) => value === undefined)) return undefined
+  return values.reduce<number>((acc, value) => acc + (value ?? 0), 0)
+}
+
 export const eventError = (route: string, message: string, raw?: string) =>
  new LLMError({
    module: "ProviderShared",
@@ -3,15 +3,70 @@ import { ContentBlockID, FinishReason, ProtocolID, ProviderMetadata, ResponseID,
 import { ModelRef } from "./options"
 import { ToolResultValue } from "./messages"

+/**
+ * Token usage reported by an LLM provider.
+ *
+ * **Inclusive totals** (match AI SDK / OpenAI / LangChain convention — a
+ * reader from any of those ecosystems sees the number they expect):
+ *
+ * - `inputTokens` — total prompt tokens, *including* cached reads/writes.
+ * - `outputTokens` — total output tokens, *including* reasoning.
+ * - `totalTokens` — provider-supplied total, or `inputTokens + outputTokens`.
+ *
+ * **Non-overlapping breakdown** (every field is independently meaningful;
+ * consumers never have to subtract):
+ *
+ * - `nonCachedInputTokens` — the "fresh" portion of the prompt.
+ * - `cacheReadInputTokens` — input tokens served from cache.
+ * - `cacheWriteInputTokens` — input tokens written to cache.
+ * - `reasoningTokens` — subset of `outputTokens` spent on hidden reasoning.
+ *
+ * **Invariant**: `nonCachedInputTokens + cacheReadInputTokens +
+ * cacheWriteInputTokens = inputTokens`, and `reasoningTokens ≤ outputTokens`.
+ * Each protocol mapper computes whichever side it doesn't get natively,
+ * with `Math.max(0, …)` clamping for defense against provider bugs. Because
+ * every breakdown field is stored independently, downstream consumers can
+ * read whatever they need (cost-by-category, context-pressure, AI-SDK-style
+ * inclusive total) without ever subtracting — eliminating the underflow
+ * class of bug where a clamped difference would silently store the wrong
+ * value.
+ *
+ * **Semantics by provider**:
+ *
+ * - OpenAI Chat / Responses / Gemini / Bedrock: provider reports inclusive
+ *   `inputTokens` and an inclusive `outputTokens`; mapper subtracts to
+ *   derive the breakdown.
+ * - Anthropic: provider reports the breakdown natively (`input_tokens` is
+ *   non-cached only); mapper sums to derive the inclusive `inputTokens`.
+ *   Anthropic does *not* break extended-thinking out of `output_tokens`, so
+ *   `reasoningTokens` is `undefined` and `outputTokens` carries the
+ *   combined total — a documented limitation of the Anthropic API.
+ *
+ * `providerMetadata` always carries the provider's raw usage payload —
+ * keyed by provider name (`{ openai: ... }`, `{ anthropic: ... }`, etc.)
+ * — for fields we don't normalize and for billing-level audit trails.
+ * Matches the same escape-hatch field on `LLMEvent`.
+ */
 export class Usage extends Schema.Class<Usage>("LLM.Usage")({
  inputTokens: Schema.optional(Schema.Number),
  outputTokens: Schema.optional(Schema.Number),
-  reasoningTokens: Schema.optional(Schema.Number),
+  nonCachedInputTokens: Schema.optional(Schema.Number),
  cacheReadInputTokens: Schema.optional(Schema.Number),
  cacheWriteInputTokens: Schema.optional(Schema.Number),
+  reasoningTokens: Schema.optional(Schema.Number),
  totalTokens: Schema.optional(Schema.Number),
-  native: Schema.optional(Schema.Record(Schema.String, Schema.Unknown)),
-}) {}
+  providerMetadata: Schema.optional(ProviderMetadata),
+}) {
+  /**
+   * Visible output tokens — `outputTokens` minus `reasoningTokens`, clamped
+   * to zero. The one place subtraction happens in this contract; the clamp
+   * means a provider reporting `reasoningTokens > outputTokens` produces a
+   * harmless zero rather than a negative that crashes downstream schemas.
+   */
+  get visibleOutputTokens() {
+    return Math.max(0, (this.outputTokens ?? 0) - (this.reasoningTokens ?? 0))
+  }
+}

 export const RequestStart = Schema.Struct({
  type: Schema.tag("request-start"),
@@ -0,0 +1,38 @@
+{
+  "version": 1,
+  "metadata": {
+    "name": "anthropic-messages/anthropic-haiku-4-5-text",
+    "recordedAt": "2026-05-11T02:02:03.804Z",
+    "provider": "anthropic",
+    "route": "anthropic-messages",
+    "transport": "http",
+    "model": "claude-haiku-4-5-20251001",
+    "tags": [
+      "prefix:anthropic-messages",
+      "provider:anthropic",
+      "text",
+      "golden"
+    ]
+  },
+  "interactions": [
+    {
+      "transport": "http",
+      "request": {
+        "method": "POST",
+        "url": "https://api.anthropic.com/v1/messages",
+        "headers": {
+          "anthropic-version": "2023-06-01",
+          "content-type": "application/json"
+        },
+        "body": "{\"model\":\"claude-haiku-4-5-20251001\",\"system\":[{\"type\":\"text\",\"text\":\"You are concise.\"}],\"messages\":[{\"role\":\"user\",\"content\":[{\"type\":\"text\",\"text\":\"Reply exactly with: Hello!\"}]}],\"stream\":true,\"max_tokens\":40,\"temperature\":0}"
+      },
+      "response": {
+        "status": 200,
+        "headers": {
+          "content-type": "text/event-stream; charset=utf-8"
+        },
+        "body": "event: message_start\ndata: {\"type\":\"message_start\",\"message\":{\"model\":\"claude-haiku-4-5-20251001\",\"id\":\"msg_01SvRWwb75gDuhBpVMHjnFaf\",\"type\":\"message\",\"role\":\"assistant\",\"content\":[],\"stop_reason\":null,\"stop_sequence\":null,\"stop_details\":null,\"usage\":{\"input_tokens\":18,\"cache_creation_input_tokens\":0,\"cache_read_input_tokens\":0,\"cache_creation\":{\"ephemeral_5m_input_tokens\":0,\"ephemeral_1h_input_tokens\":0},\"output_tokens\":2,\"service_tier\":\"standard\",\"inference_geo\":\"not_available\"}}   }\n\nevent: content_block_start\ndata: {\"type\":\"content_block_start\",\"index\":0,\"content_block\":{\"type\":\"text\",\"text\":\"\"}            }\n\nevent: ping\ndata: {\"type\": \"ping\"}\n\nevent: content_block_delta\ndata: {\"type\":\"content_block_delta\",\"index\":0,\"delta\":{\"type\":\"text_delta\",\"text\":\"Hello!\"}  }\n\nevent: content_block_stop\ndata: {\"type\":\"content_block_stop\",\"index\":0          }\n\nevent: message_delta\ndata: {\"type\":\"message_delta\",\"delta\":{\"stop_reason\":\"end_turn\",\"stop_sequence\":null,\"stop_details\":null},\"usage\":{\"input_tokens\":18,\"cache_creation_input_tokens\":0,\"cache_read_input_tokens\":0,\"output_tokens\":5}  }\n\nevent: message_stop\ndata: {\"type\":\"message_stop\"          }\n\n"
+      }
+    }
+  ]
+}
@@ -0,0 +1,39 @@
+{
+  "version": 1,
+  "metadata": {
+    "name": "anthropic-messages/anthropic-haiku-4-5-tool-call",
+    "recordedAt": "2026-05-11T02:02:04.363Z",
+    "provider": "anthropic",
+    "route": "anthropic-messages",
+    "transport": "http",
+    "model": "claude-haiku-4-5-20251001",
+    "tags": [
+      "prefix:anthropic-messages",
+      "provider:anthropic",
+      "tool",
+      "tool-call",
+      "golden"
+    ]
+  },
+  "interactions": [
+    {
+      "transport": "http",
+      "request": {
+        "method": "POST",
+        "url": "https://api.anthropic.com/v1/messages",
+        "headers": {
+          "anthropic-version": "2023-06-01",
+          "content-type": "application/json"
+        },
+        "body": "{\"model\":\"claude-haiku-4-5-20251001\",\"system\":[{\"type\":\"text\",\"text\":\"Call tools exactly as requested.\"}],\"messages\":[{\"role\":\"user\",\"content\":[{\"type\":\"text\",\"text\":\"Call get_weather with city exactly Paris.\"}]}],\"tools\":[{\"name\":\"get_weather\",\"description\":\"Get current weather for a city.\",\"input_schema\":{\"type\":\"object\",\"properties\":{\"city\":{\"type\":\"string\"}},\"required\":[\"city\"],\"additionalProperties\":false}}],\"tool_choice\":{\"type\":\"tool\",\"name\":\"get_weather\"},\"stream\":true,\"max_tokens\":80,\"temperature\":0}"
+      },
+      "response": {
+        "status": 200,
+        "headers": {
+          "content-type": "text/event-stream; charset=utf-8"
+        },
+        "body": "event: message_start\ndata: {\"type\":\"message_start\",\"message\":{\"model\":\"claude-haiku-4-5-20251001\",\"id\":\"msg_01Lu38yDM3WD8QBQTcg3dBaF\",\"type\":\"message\",\"role\":\"assistant\",\"content\":[],\"stop_reason\":null,\"stop_sequence\":null,\"stop_details\":null,\"usage\":{\"input_tokens\":677,\"cache_creation_input_tokens\":0,\"cache_read_input_tokens\":0,\"cache_creation\":{\"ephemeral_5m_input_tokens\":0,\"ephemeral_1h_input_tokens\":0},\"output_tokens\":16,\"service_tier\":\"standard\",\"inference_geo\":\"not_available\"}}}\n\nevent: content_block_start\ndata: {\"type\":\"content_block_start\",\"index\":0,\"content_block\":{\"type\":\"tool_use\",\"id\":\"toolu_017Dqk9SAAsHHfiLKsUyitaQ\",\"name\":\"get_weather\",\"input\":{},\"caller\":{\"type\":\"direct\"}}   }\n\nevent: ping\ndata: {\"type\": \"ping\"}\n\nevent: content_block_delta\ndata: {\"type\":\"content_block_delta\",\"index\":0,\"delta\":{\"type\":\"input_json_delta\",\"partial_json\":\"\"}    }\n\nevent: content_block_delta\ndata: {\"type\":\"content_block_delta\",\"index\":0,\"delta\":{\"type\":\"input_json_delta\",\"partial_json\":\"{\\\"cit\"}          }\n\nevent: content_block_delta\ndata: {\"type\":\"content_block_delta\",\"index\":0,\"delta\":{\"type\":\"input_json_delta\",\"partial_json\":\"y\\\": \\\"Paris\\\"}\"}           }\n\nevent: content_block_stop\ndata: {\"type\":\"content_block_stop\",\"index\":0}\n\nevent: message_delta\ndata: {\"type\":\"message_delta\",\"delta\":{\"stop_reason\":\"tool_use\",\"stop_sequence\":null,\"stop_details\":null},\"usage\":{\"input_tokens\":677,\"cache_creation_input_tokens\":0,\"cache_read_input_tokens\":0,\"output_tokens\":33}      }\n\nevent: message_stop\ndata: {\"type\":\"message_stop\"             }\n\n"
+      }
+    }
+  ]
+}
@@ -0,0 +1,59 @@
+{
+  "version": 1,
+  "metadata": {
+    "name": "anthropic-messages/anthropic-opus-4-7-tool-loop",
+    "recordedAt": "2026-05-11T02:02:07.788Z",
+    "provider": "anthropic",
+    "route": "anthropic-messages",
+    "transport": "http",
+    "model": "claude-opus-4-7",
+    "tags": [
+      "prefix:anthropic-messages",
+      "provider:anthropic",
+      "flagship",
+      "tool",
+      "tool-loop",
+      "golden"
+    ]
+  },
+  "interactions": [
+    {
+      "transport": "http",
+      "request": {
+        "method": "POST",
+        "url": "https://api.anthropic.com/v1/messages",
+        "headers": {
+          "anthropic-version": "2023-06-01",
+          "content-type": "application/json"
+        },
+        "body": "{\"model\":\"claude-opus-4-7\",\"system\":[{\"type\":\"text\",\"text\":\"Use the get_weather tool exactly once. After the tool result, reply exactly: Paris is sunny.\"}],\"messages\":[{\"role\":\"user\",\"content\":[{\"type\":\"text\",\"text\":\"What is the weather in Paris?\"}]}],\"tools\":[{\"name\":\"get_weather\",\"description\":\"Get current weather for a city.\",\"input_schema\":{\"type\":\"object\",\"properties\":{\"city\":{\"type\":\"string\"}},\"required\":[\"city\"],\"additionalProperties\":false}}],\"stream\":true,\"max_tokens\":80}"
+      },
+      "response": {
+        "status": 200,
+        "headers": {
+          "content-type": "text/event-stream; charset=utf-8"
+        },
+        "body": "event: message_start\ndata: {\"type\":\"message_start\",\"message\":{\"model\":\"claude-opus-4-7\",\"id\":\"msg_01GK5kgi8AuVfRCnQFcXEfV8\",\"type\":\"message\",\"role\":\"assistant\",\"content\":[],\"stop_reason\":null,\"stop_sequence\":null,\"stop_details\":null,\"usage\":{\"input_tokens\":812,\"cache_creation_input_tokens\":0,\"cache_read_input_tokens\":0,\"cache_creation\":{\"ephemeral_5m_input_tokens\":0,\"ephemeral_1h_input_tokens\":0},\"output_tokens\":0,\"service_tier\":\"standard\",\"inference_geo\":\"global\"}} }\n\nevent: content_block_start\ndata: {\"type\":\"content_block_start\",\"index\":0,\"content_block\":{\"type\":\"tool_use\",\"id\":\"toolu_01BnVDAp13NU8ZdJ9JeJ7byF\",\"name\":\"get_weather\",\"input\":{},\"caller\":{\"type\":\"direct\"}}}\n\nevent: ping\ndata: {\"type\": \"ping\"}\n\nevent: content_block_delta\ndata: {\"type\":\"content_block_delta\",\"index\":0,\"delta\":{\"type\":\"input_json_delta\",\"partial_json\":\"\"}        }\n\nevent: content_block_delta\ndata: {\"type\":\"content_block_delta\",\"index\":0,\"delta\":{\"type\":\"input_json_delta\",\"partial_json\":\"{\\\"c\"}      }\n\nevent: content_block_delta\ndata: {\"type\":\"content_block_delta\",\"index\":0,\"delta\":{\"type\":\"input_json_delta\",\"partial_json\":\"ity\\\": \\\"\"}        }\n\nevent: content_block_delta\ndata: {\"type\":\"content_block_delta\",\"index\":0,\"delta\":{\"type\":\"input_json_delta\",\"partial_json\":\"Paris\"} }\n\nevent: content_block_delta\ndata: {\"type\":\"content_block_delta\",\"index\":0,\"delta\":{\"type\":\"input_json_delta\",\"partial_json\":\"\\\"}\"}         }\n\nevent: content_block_stop\ndata: {\"type\":\"content_block_stop\",\"index\":0       }\n\nevent: message_delta\ndata: {\"type\":\"message_delta\",\"delta\":{\"stop_reason\":\"tool_use\",\"stop_sequence\":null,\"stop_details\":null},\"usage\":{\"input_tokens\":812,\"cache_creation_input_tokens\":0,\"cache_read_input_tokens\":0,\"output_tokens\":66}   }\n\nevent: message_stop\ndata: {\"type\":\"message_stop\"          }\n\n"
+      }
+    },
+    {
+      "transport": "http",
+      "request": {
+        "method": "POST",
+        "url": "https://api.anthropic.com/v1/messages",
+        "headers": {
+          "anthropic-version": "2023-06-01",
+          "content-type": "application/json"
+        },
+        "body": "{\"model\":\"claude-opus-4-7\",\"system\":[{\"type\":\"text\",\"text\":\"Use the get_weather tool exactly once. After the tool result, reply exactly: Paris is sunny.\"}],\"messages\":[{\"role\":\"user\",\"content\":[{\"type\":\"text\",\"text\":\"What is the weather in Paris?\"}]},{\"role\":\"assistant\",\"content\":[{\"type\":\"tool_use\",\"id\":\"toolu_01BnVDAp13NU8ZdJ9JeJ7byF\",\"name\":\"get_weather\",\"input\":{\"city\":\"Paris\"}}]},{\"role\":\"user\",\"content\":[{\"type\":\"tool_result\",\"tool_use_id\":\"toolu_01BnVDAp13NU8ZdJ9JeJ7byF\",\"content\":\"{\\\"temperature\\\":22,\\\"condition\\\":\\\"sunny\\\"}\"}]}],\"tools\":[{\"name\":\"get_weather\",\"description\":\"Get current weather for a city.\",\"input_schema\":{\"type\":\"object\",\"properties\":{\"city\":{\"type\":\"string\"}},\"required\":[\"city\"],\"additionalProperties\":false}}],\"stream\":true,\"max_tokens\":80}"
+      },
+      "response": {
+        "status": 200,
+        "headers": {
+          "content-type": "text/event-stream; charset=utf-8"
+        },
+        "body": "event: message_start\ndata: {\"type\":\"message_start\",\"message\":{\"model\":\"claude-opus-4-7\",\"id\":\"msg_01237VTnjPeYSRh31UjXWaEa\",\"type\":\"message\",\"role\":\"assistant\",\"content\":[],\"stop_reason\":null,\"stop_sequence\":null,\"stop_details\":null,\"usage\":{\"input_tokens\":909,\"cache_creation_input_tokens\":0,\"cache_read_input_tokens\":0,\"cache_creation\":{\"ephemeral_5m_input_tokens\":0,\"ephemeral_1h_input_tokens\":0},\"output_tokens\":8,\"service_tier\":\"standard\",\"inference_geo\":\"global\"}}       }\n\nevent: content_block_start\ndata: {\"type\":\"content_block_start\",\"index\":0,\"content_block\":{\"type\":\"text\",\"text\":\"\"}               }\n\nevent: ping\ndata: {\"type\": \"ping\"}\n\nevent: content_block_delta\ndata: {\"type\":\"content_block_delta\",\"index\":0,\"delta\":{\"type\":\"text_delta\",\"text\":\"Paris is sunny.\"}     }\n\nevent: content_block_stop\ndata: {\"type\":\"content_block_stop\",\"index\":0      }\n\nevent: message_delta\ndata: {\"type\":\"message_delta\",\"delta\":{\"stop_reason\":\"end_turn\",\"stop_sequence\":null,\"stop_details\":null},\"usage\":{\"input_tokens\":909,\"cache_creation_input_tokens\":0,\"cache_read_input_tokens\":0,\"output_tokens\":12}      }\n\nevent: message_stop\ndata: {\"type\":\"message_stop\"       }\n\n"
+      }
+    }
+  ]
+}
@@ -0,0 +1,35 @@
+{
+  "version": 1,
+  "metadata": {
+    "name": "anthropic-messages/rejects-malformed-assistant-tool-order",
+    "recordedAt": "2026-05-11T02:01:44.544Z",
+    "tags": [
+      "prefix:anthropic-messages",
+      "provider:anthropic",
+      "protocol:anthropic-messages",
+      "tool",
+      "sad-path"
+    ]
+  },
+  "interactions": [
+    {
+      "transport": "http",
+      "request": {
+        "method": "POST",
+        "url": "https://api.anthropic.com/v1/messages",
+        "headers": {
+          "anthropic-version": "2023-06-01",
+          "content-type": "application/json"
+        },
+        "body": "{\"model\":\"claude-haiku-4-5-20251001\",\"messages\":[{\"role\":\"assistant\",\"content\":[{\"type\":\"tool_use\",\"id\":\"call_1\",\"name\":\"get_weather\",\"input\":{\"city\":\"Paris\"}},{\"type\":\"text\",\"text\":\"I will check the weather.\"}]},{\"role\":\"user\",\"content\":[{\"type\":\"tool_result\",\"tool_use_id\":\"call_1\",\"content\":\"{\\\"temperature\\\":\\\"72F\\\"}\"}]},{\"role\":\"user\",\"content\":[{\"type\":\"text\",\"text\":\"Use that result to answer briefly.\"}]}],\"tools\":[{\"name\":\"get_weather\",\"description\":\"Get weather\",\"input_schema\":{\"type\":\"object\",\"properties\":{}}}],\"stream\":true,\"max_tokens\":4096}"
+      },
+      "response": {
+        "status": 400,
+        "headers": {
+          "content-type": "application/json"
+        },
+        "body": "{\"type\":\"error\",\"error\":{\"type\":\"invalid_request_error\",\"message\":\"messages.1: `tool_use` ids were found without `tool_result` blocks immediately after: call_1. Each `tool_use` block must have a corresponding `tool_result` block in the next message.\"},\"request_id\":\"req_011CauxVdQf3N2PPFJ5aH8Bh\"}"
+      }
+    }
+  ]
+}
@@ -0,0 +1,37 @@
+{
+  "version": 1,
+  "metadata": {
+    "name": "gemini/gemini-2-5-flash-text",
+    "recordedAt": "2026-05-11T02:02:08.410Z",
+    "provider": "google",
+    "route": "gemini",
+    "transport": "http",
+    "model": "gemini-2.5-flash",
+    "tags": [
+      "prefix:gemini",
+      "provider:google",
+      "text",
+      "golden"
+    ]
+  },
+  "interactions": [
+    {
+      "transport": "http",
+      "request": {
+        "method": "POST",
+        "url": "https://generativelanguage.googleapis.com/v1beta/models/gemini-2.5-flash:streamGenerateContent?alt=sse",
+        "headers": {
+          "content-type": "application/json"
+        },
+        "body": "{\"contents\":[{\"role\":\"user\",\"parts\":[{\"text\":\"Reply exactly with: Hello!\"}]}],\"systemInstruction\":{\"parts\":[{\"text\":\"You are concise.\"}]},\"generationConfig\":{\"maxOutputTokens\":80,\"temperature\":0}}"
+      },
+      "response": {
+        "status": 200,
+        "headers": {
+          "content-type": "text/event-stream"
+        },
+        "body": "data: {\"candidates\": [{\"content\": {\"parts\": [{\"text\": \"Hello!\"}],\"role\": \"model\"},\"finishReason\": \"STOP\",\"index\": 0}],\"usageMetadata\": {\"promptTokenCount\": 11,\"candidatesTokenCount\": 2,\"totalTokenCount\": 13,\"promptTokensDetails\": [{\"modality\": \"TEXT\",\"tokenCount\": 11}],\"serviceTier\": \"standard\"},\"modelVersion\": \"gemini-2.5-flash\",\"responseId\": \"nzgBatP3OZmW-8YP567bqQs\"}\r\n\r\n"
+      }
+    }
+  ]
+}
@@ -0,0 +1,38 @@
+{
+  "version": 1,
+  "metadata": {
+    "name": "gemini/gemini-2-5-flash-tool-call",
+    "recordedAt": "2026-05-11T02:02:09.308Z",
+    "provider": "google",
+    "route": "gemini",
+    "transport": "http",
+    "model": "gemini-2.5-flash",
+    "tags": [
+      "prefix:gemini",
+      "provider:google",
+      "tool",
+      "tool-call",
+      "golden"
+    ]
+  },
+  "interactions": [
+    {
+      "transport": "http",
+      "request": {
+        "method": "POST",
+        "url": "https://generativelanguage.googleapis.com/v1beta/models/gemini-2.5-flash:streamGenerateContent?alt=sse",
+        "headers": {
+          "content-type": "application/json"
+        },
+        "body": "{\"contents\":[{\"role\":\"user\",\"parts\":[{\"text\":\"Call get_weather with city exactly Paris.\"}]}],\"systemInstruction\":{\"parts\":[{\"text\":\"Call tools exactly as requested.\"}]},\"tools\":[{\"functionDeclarations\":[{\"name\":\"get_weather\",\"description\":\"Get current weather for a city.\",\"parameters\":{\"required\":[\"city\"],\"type\":\"object\",\"properties\":{\"city\":{\"type\":\"string\"}}}}]}],\"toolConfig\":{\"functionCallingConfig\":{\"mode\":\"ANY\",\"allowedFunctionNames\":[\"get_weather\"]}},\"generationConfig\":{\"maxOutputTokens\":80,\"temperature\":0}}"
+      },
+      "response": {
+        "status": 200,
+        "headers": {
+          "content-type": "text/event-stream"
+        },
+        "body": "data: {\"candidates\": [{\"content\": {\"parts\": [{\"functionCall\": {\"name\": \"get_weather\",\"args\": {\"city\": \"Paris\"}},\"thoughtSignature\": \"CiQBDDnWx/X6sWeX2joSugyWO3L/lt0AgIPCvhpqf3845fj+H70KXwEMOdbHB/cnaqYCro0pU+yLWoA55jhuwoLmTcnYm4Qzcm5DuW/v0NUyz8RDx6DFh61juENveUztly6yc6/XiWJHtsgncd9YgcZhuQKqtp5KZTkGYpT3g6v3yP9GK4AoCoUBAQw51sePh3WuWovHnwIotKLVZiU9pwh34k4FY7ugPOxyDAG9j69cy7BYYzSchI10LEjLoLlCMZuNIPootBgI02QWY/4h2PIv33BAADrFPM2T3aE4cAuMoa3GCu2nztJ/95junDhIhuXZSQ/Mh9EVxpx7ml99Z7Hxb7OtDsSZZLeCBuGmSw==\"}],\"role\": \"model\"},\"finishReason\": \"STOP\",\"index\": 0,\"finishMessage\": \"Model generated function call(s).\"}],\"usageMetadata\": {\"promptTokenCount\": 55,\"candidatesTokenCount\": 15,\"totalTokenCount\": 115,\"promptTokensDetails\": [{\"modality\": \"TEXT\",\"tokenCount\": 55}],\"thoughtsTokenCount\": 45,\"serviceTier\": \"standard\"},\"modelVersion\": \"gemini-2.5-flash\",\"responseId\": \"oDgBauj6HZ3B-8YPpaOc6QU\"}\r\n\r\n"
+      }
+    }
+  ]
+}
@@ -0,0 +1,37 @@
+{
+  "version": 1,
+  "metadata": {
+    "name": "openai-chat/openai-chat-gpt-4o-mini-text",
+    "recordedAt": "2026-05-11T02:01:46.536Z",
+    "provider": "openai",
+    "route": "openai-chat",
+    "transport": "http",
+    "model": "gpt-4o-mini",
+    "tags": [
+      "prefix:openai-chat",
+      "provider:openai",
+      "text",
+      "golden"
+    ]
+  },
+  "interactions": [
+    {
+      "transport": "http",
+      "request": {
+        "method": "POST",
+        "url": "https://api.openai.com/v1/chat/completions",
+        "headers": {
+          "content-type": "application/json"
+        },
+        "body": "{\"model\":\"gpt-4o-mini\",\"messages\":[{\"role\":\"system\",\"content\":\"You are concise.\"},{\"role\":\"user\",\"content\":\"Reply exactly with: Hello!\"}],\"stream\":true,\"stream_options\":{\"include_usage\":true},\"max_tokens\":40,\"temperature\":0}"
+      },
+      "response": {
+        "status": 200,
+        "headers": {
+          "content-type": "text/event-stream; charset=utf-8"
+        },
+        "body": "data: {\"id\":\"chatcmpl-DeAFmC9UiBnVu6dAjBvWVBVypm2gc\",\"object\":\"chat.completion.chunk\",\"created\":1778464906,\"model\":\"gpt-4o-mini-2024-07-18\",\"service_tier\":\"default\",\"system_fingerprint\":\"fp_3ef558a83f\",\"choices\":[{\"index\":0,\"delta\":{\"role\":\"assistant\",\"content\":\"\",\"refusal\":null},\"logprobs\":null,\"finish_reason\":null}],\"usage\":null,\"obfuscation\":\"wHajUz1js\"}\n\ndata: {\"id\":\"chatcmpl-DeAFmC9UiBnVu6dAjBvWVBVypm2gc\",\"object\":\"chat.completion.chunk\",\"created\":1778464906,\"model\":\"gpt-4o-mini-2024-07-18\",\"service_tier\":\"default\",\"system_fingerprint\":\"fp_3ef558a83f\",\"choices\":[{\"index\":0,\"delta\":{\"content\":\"Hello\"},\"logprobs\":null,\"finish_reason\":null}],\"usage\":null,\"obfuscation\":\"lVHzGq\"}\n\ndata: {\"id\":\"chatcmpl-DeAFmC9UiBnVu6dAjBvWVBVypm2gc\",\"object\":\"chat.completion.chunk\",\"created\":1778464906,\"model\":\"gpt-4o-mini-2024-07-18\",\"service_tier\":\"default\",\"system_fingerprint\":\"fp_3ef558a83f\",\"choices\":[{\"index\":0,\"delta\":{\"content\":\"!\"},\"logprobs\":null,\"finish_reason\":null}],\"usage\":null,\"obfuscation\":\"3gaQRWEjE7\"}\n\ndata: {\"id\":\"chatcmpl-DeAFmC9UiBnVu6dAjBvWVBVypm2gc\",\"object\":\"chat.completion.chunk\",\"created\":1778464906,\"model\":\"gpt-4o-mini-2024-07-18\",\"service_tier\":\"default\",\"system_fingerprint\":\"fp_3ef558a83f\",\"choices\":[{\"index\":0,\"delta\":{},\"logprobs\":null,\"finish_reason\":\"stop\"}],\"usage\":null,\"obfuscation\":\"vuKPZ\"}\n\ndata: {\"id\":\"chatcmpl-DeAFmC9UiBnVu6dAjBvWVBVypm2gc\",\"object\":\"chat.completion.chunk\",\"created\":1778464906,\"model\":\"gpt-4o-mini-2024-07-18\",\"service_tier\":\"default\",\"system_fingerprint\":\"fp_3ef558a83f\",\"choices\":[],\"usage\":{\"prompt_tokens\":21,\"completion_tokens\":2,\"total_tokens\":23,\"prompt_tokens_details\":{\"cached_tokens\":0,\"audio_tokens\":0},\"completion_tokens_details\":{\"reasoning_tokens\":0,\"audio_tokens\":0,\"accepted_prediction_tokens\":0,\"rejected_prediction_tokens\":0}},\"obfuscation\":\"kfFsssdujmM\"}\n\ndata: [DONE]\n\n"
+      }
+    }
+  ]
+}
@@ -0,0 +1,38 @@
+{
+  "version": 1,
+  "metadata": {
+    "name": "openai-chat/openai-chat-gpt-4o-mini-tool-call",
+    "recordedAt": "2026-05-11T02:01:47.484Z",
+    "provider": "openai",
+    "route": "openai-chat",
+    "transport": "http",
+    "model": "gpt-4o-mini",
+    "tags": [
+      "prefix:openai-chat",
+      "provider:openai",
+      "tool",
+      "tool-call",
+      "golden"
+    ]
+  },
+  "interactions": [
+    {
+      "transport": "http",
+      "request": {
+        "method": "POST",
+        "url": "https://api.openai.com/v1/chat/completions",
+        "headers": {
+          "content-type": "application/json"
+        },
+        "body": "{\"model\":\"gpt-4o-mini\",\"messages\":[{\"role\":\"system\",\"content\":\"Call tools exactly as requested.\"},{\"role\":\"user\",\"content\":\"Call get_weather with city exactly Paris.\"}],\"tools\":[{\"type\":\"function\",\"function\":{\"name\":\"get_weather\",\"description\":\"Get current weather for a city.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"city\":{\"type\":\"string\"}},\"required\":[\"city\"],\"additionalProperties\":false}}}],\"tool_choice\":{\"type\":\"function\",\"function\":{\"name\":\"get_weather\"}},\"stream\":true,\"stream_options\":{\"include_usage\":true},\"max_tokens\":80,\"temperature\":0}"
+      },
+      "response": {
+        "status": 200,
+        "headers": {
+          "content-type": "text/event-stream; charset=utf-8"
+        },
+        "body": "data: {\"id\":\"chatcmpl-DeAFnVMdELyuymMRlTqN5dBdM7eGd\",\"object\":\"chat.completion.chunk\",\"created\":1778464907,\"model\":\"gpt-4o-mini-2024-07-18\",\"service_tier\":\"default\",\"system_fingerprint\":\"fp_d88f6e55bb\",\"choices\":[{\"index\":0,\"delta\":{\"role\":\"assistant\",\"content\":null,\"tool_calls\":[{\"index\":0,\"id\":\"call_sH8T7MPdJXS5KJginKrzexL5\",\"type\":\"function\",\"function\":{\"name\":\"get_weather\",\"arguments\":\"\"}}],\"refusal\":null},\"logprobs\":null,\"finish_reason\":null}],\"usage\":null,\"obfuscation\":\"d0rJ\"}\n\ndata: {\"id\":\"chatcmpl-DeAFnVMdELyuymMRlTqN5dBdM7eGd\",\"object\":\"chat.completion.chunk\",\"created\":1778464907,\"model\":\"gpt-4o-mini-2024-07-18\",\"service_tier\":\"default\",\"system_fingerprint\":\"fp_d88f6e55bb\",\"choices\":[{\"index\":0,\"delta\":{\"tool_calls\":[{\"index\":0,\"function\":{\"arguments\":\"{\\\"\"}}]},\"logprobs\":null,\"finish_reason\":null}],\"usage\":null,\"obfuscation\":\"hL82erd6bBoy91\"}\n\ndata: {\"id\":\"chatcmpl-DeAFnVMdELyuymMRlTqN5dBdM7eGd\",\"object\":\"chat.completion.chunk\",\"created\":1778464907,\"model\":\"gpt-4o-mini-2024-07-18\",\"service_tier\":\"default\",\"system_fingerprint\":\"fp_d88f6e55bb\",\"choices\":[{\"index\":0,\"delta\":{\"tool_calls\":[{\"index\":0,\"function\":{\"arguments\":\"city\"}}]},\"logprobs\":null,\"finish_reason\":null}],\"usage\":null,\"obfuscation\":\"e3aoKH4tvJlXW\"}\n\ndata: {\"id\":\"chatcmpl-DeAFnVMdELyuymMRlTqN5dBdM7eGd\",\"object\":\"chat.completion.chunk\",\"created\":1778464907,\"model\":\"gpt-4o-mini-2024-07-18\",\"service_tier\":\"default\",\"system_fingerprint\":\"fp_d88f6e55bb\",\"choices\":[{\"index\":0,\"delta\":{\"tool_calls\":[{\"index\":0,\"function\":{\"arguments\":\"\\\":\\\"\"}}]},\"logprobs\":null,\"finish_reason\":null}],\"usage\":null,\"obfuscation\":\"T8EzFVNaUCLE\"}\n\ndata: {\"id\":\"chatcmpl-DeAFnVMdELyuymMRlTqN5dBdM7eGd\",\"object\":\"chat.completion.chunk\",\"created\":1778464907,\"model\":\"gpt-4o-mini-2024-07-18\",\"service_tier\":\"default\",\"system_fingerprint\":\"fp_d88f6e55bb\",\"choices\":[{\"index\":0,\"delta\":{\"tool_calls\":[{\"index\":0,\"function\":{\"arguments\":\"Paris\"}}]},\"logprobs\":null,\"finish_reason\":null}],\"usage\":null,\"obfuscation\":\"lelRBxF08Zes\"}\n\ndata: {\"id\":\"chatcmpl-DeAFnVMdELyuymMRlTqN5dBdM7eGd\",\"object\":\"chat.completion.chunk\",\"created\":1778464907,\"model\":\"gpt-4o-mini-2024-07-18\",\"service_tier\":\"default\",\"system_fingerprint\":\"fp_d88f6e55bb\",\"choices\":[{\"index\":0,\"delta\":{\"tool_calls\":[{\"index\":0,\"function\":{\"arguments\":\"\\\"}\"}}]},\"logprobs\":null,\"finish_reason\":null}],\"usage\":null,\"obfuscation\":\"F1xO8sVMyZt4BU\"}\n\ndata: {\"id\":\"chatcmpl-DeAFnVMdELyuymMRlTqN5dBdM7eGd\",\"object\":\"chat.completion.chunk\",\"created\":1778464907,\"model\":\"gpt-4o-mini-2024-07-18\",\"service_tier\":\"default\",\"system_fingerprint\":\"fp_d88f6e55bb\",\"choices\":[{\"index\":0,\"delta\":{},\"logprobs\":null,\"finish_reason\":\"stop\"}],\"usage\":null,\"obfuscation\":\"YL3pl\"}\n\ndata: {\"id\":\"chatcmpl-DeAFnVMdELyuymMRlTqN5dBdM7eGd\",\"object\":\"chat.completion.chunk\",\"created\":1778464907,\"model\":\"gpt-4o-mini-2024-07-18\",\"service_tier\":\"default\",\"system_fingerprint\":\"fp_d88f6e55bb\",\"choices\":[],\"usage\":{\"prompt_tokens\":67,\"completion_tokens\":5,\"total_tokens\":72,\"prompt_tokens_details\":{\"cached_tokens\":0,\"audio_tokens\":0},\"completion_tokens_details\":{\"reasoning_tokens\":0,\"audio_tokens\":0,\"accepted_prediction_tokens\":0,\"rejected_prediction_tokens\":0}},\"obfuscation\":\"7cov9qkofwo\"}\n\ndata: [DONE]\n\n"
+      }
+    }
+  ]
+}
@@ -0,0 +1,56 @@
+{
+  "version": 1,
+  "metadata": {
+    "name": "openai-chat/openai-chat-gpt-4o-mini-tool-loop",
+    "recordedAt": "2026-05-11T02:01:50.433Z",
+    "provider": "openai",
+    "route": "openai-chat",
+    "transport": "http",
+    "model": "gpt-4o-mini",
+    "tags": [
+      "prefix:openai-chat",
+      "provider:openai",
+      "tool",
+      "tool-loop",
+      "golden"
+    ]
+  },
+  "interactions": [
+    {
+      "transport": "http",
+      "request": {
+        "method": "POST",
+        "url": "https://api.openai.com/v1/chat/completions",
+        "headers": {
+          "content-type": "application/json"
+        },
+        "body": "{\"model\":\"gpt-4o-mini\",\"messages\":[{\"role\":\"system\",\"content\":\"Use the get_weather tool exactly once. After the tool result, reply exactly: Paris is sunny.\"},{\"role\":\"user\",\"content\":\"What is the weather in Paris?\"}],\"tools\":[{\"type\":\"function\",\"function\":{\"name\":\"get_weather\",\"description\":\"Get current weather for a city.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"city\":{\"type\":\"string\"}},\"required\":[\"city\"],\"additionalProperties\":false}}}],\"stream\":true,\"stream_options\":{\"include_usage\":true},\"max_tokens\":80,\"temperature\":0}"
+      },
+      "response": {
+        "status": 200,
+        "headers": {
+          "content-type": "text/event-stream; charset=utf-8"
+        },
+        "body": "data: {\"id\":\"chatcmpl-DeAFo2rbsuuMn0ftzxwfDpqnYANrT\",\"object\":\"chat.completion.chunk\",\"created\":1778464908,\"model\":\"gpt-4o-mini-2024-07-18\",\"service_tier\":\"default\",\"system_fingerprint\":\"fp_c83829e916\",\"choices\":[{\"index\":0,\"delta\":{\"role\":\"assistant\",\"content\":null,\"tool_calls\":[{\"index\":0,\"id\":\"call_HrDZhrMUauvVddKWzVQFJ69Q\",\"type\":\"function\",\"function\":{\"name\":\"get_weather\",\"arguments\":\"\"}}],\"refusal\":null},\"logprobs\":null,\"finish_reason\":null}],\"usage\":null,\"obfuscation\":\"aAqC\"}\n\ndata: {\"id\":\"chatcmpl-DeAFo2rbsuuMn0ftzxwfDpqnYANrT\",\"object\":\"chat.completion.chunk\",\"created\":1778464908,\"model\":\"gpt-4o-mini-2024-07-18\",\"service_tier\":\"default\",\"system_fingerprint\":\"fp_c83829e916\",\"choices\":[{\"index\":0,\"delta\":{\"tool_calls\":[{\"index\":0,\"function\":{\"arguments\":\"{\\\"\"}}]},\"logprobs\":null,\"finish_reason\":null}],\"usage\":null,\"obfuscation\":\"haZ3pKk14oDay0\"}\n\ndata: {\"id\":\"chatcmpl-DeAFo2rbsuuMn0ftzxwfDpqnYANrT\",\"object\":\"chat.completion.chunk\",\"created\":1778464908,\"model\":\"gpt-4o-mini-2024-07-18\",\"service_tier\":\"default\",\"system_fingerprint\":\"fp_c83829e916\",\"choices\":[{\"index\":0,\"delta\":{\"tool_calls\":[{\"index\":0,\"function\":{\"arguments\":\"city\"}}]},\"logprobs\":null,\"finish_reason\":null}],\"usage\":null,\"obfuscation\":\"ZUJHQytFyDezp\"}\n\ndata: {\"id\":\"chatcmpl-DeAFo2rbsuuMn0ftzxwfDpqnYANrT\",\"object\":\"chat.completion.chunk\",\"created\":1778464908,\"model\":\"gpt-4o-mini-2024-07-18\",\"service_tier\":\"default\",\"system_fingerprint\":\"fp_c83829e916\",\"choices\":[{\"index\":0,\"delta\":{\"tool_calls\":[{\"index\":0,\"function\":{\"arguments\":\"\\\":\\\"\"}}]},\"logprobs\":null,\"finish_reason\":null}],\"usage\":null,\"obfuscation\":\"3buqnStPteyG\"}\n\ndata: {\"id\":\"chatcmpl-DeAFo2rbsuuMn0ftzxwfDpqnYANrT\",\"object\":\"chat.completion.chunk\",\"created\":1778464908,\"model\":\"gpt-4o-mini-2024-07-18\",\"service_tier\":\"default\",\"system_fingerprint\":\"fp_c83829e916\",\"choices\":[{\"index\":0,\"delta\":{\"tool_calls\":[{\"index\":0,\"function\":{\"arguments\":\"Paris\"}}]},\"logprobs\":null,\"finish_reason\":null}],\"usage\":null,\"obfuscation\":\"7epPqHEU3OeA\"}\n\ndata: {\"id\":\"chatcmpl-DeAFo2rbsuuMn0ftzxwfDpqnYANrT\",\"object\":\"chat.completion.chunk\",\"created\":1778464908,\"model\":\"gpt-4o-mini-2024-07-18\",\"service_tier\":\"default\",\"system_fingerprint\":\"fp_c83829e916\",\"choices\":[{\"index\":0,\"delta\":{\"tool_calls\":[{\"index\":0,\"function\":{\"arguments\":\"\\\"}\"}}]},\"logprobs\":null,\"finish_reason\":null}],\"usage\":null,\"obfuscation\":\"gSU5gyl9K7sUCv\"}\n\ndata: {\"id\":\"chatcmpl-DeAFo2rbsuuMn0ftzxwfDpqnYANrT\",\"object\":\"chat.completion.chunk\",\"created\":1778464908,\"model\":\"gpt-4o-mini-2024-07-18\",\"service_tier\":\"default\",\"system_fingerprint\":\"fp_c83829e916\",\"choices\":[{\"index\":0,\"delta\":{},\"logprobs\":null,\"finish_reason\":\"tool_calls\"}],\"usage\":null,\"obfuscation\":\"P3oSvByfy60JCsz\"}\n\ndata: {\"id\":\"chatcmpl-DeAFo2rbsuuMn0ftzxwfDpqnYANrT\",\"object\":\"chat.completion.chunk\",\"created\":1778464908,\"model\":\"gpt-4o-mini-2024-07-18\",\"service_tier\":\"default\",\"system_fingerprint\":\"fp_c83829e916\",\"choices\":[],\"usage\":{\"prompt_tokens\":71,\"completion_tokens\":14,\"total_tokens\":85,\"prompt_tokens_details\":{\"cached_tokens\":0,\"audio_tokens\":0},\"completion_tokens_details\":{\"reasoning_tokens\":0,\"audio_tokens\":0,\"accepted_prediction_tokens\":0,\"rejected_prediction_tokens\":0}},\"obfuscation\":\"5AhQ5ToYra\"}\n\ndata: [DONE]\n\n"
+      }
+    },
+    {
+      "transport": "http",
+      "request": {
+        "method": "POST",
+        "url": "https://api.openai.com/v1/chat/completions",
+        "headers": {
+          "content-type": "application/json"
+        },
+        "body": "{\"model\":\"gpt-4o-mini\",\"messages\":[{\"role\":\"system\",\"content\":\"Use the get_weather tool exactly once. After the tool result, reply exactly: Paris is sunny.\"},{\"role\":\"user\",\"content\":\"What is the weather in Paris?\"},{\"role\":\"assistant\",\"content\":null,\"tool_calls\":[{\"id\":\"call_HrDZhrMUauvVddKWzVQFJ69Q\",\"type\":\"function\",\"function\":{\"name\":\"get_weather\",\"arguments\":\"{\\\"city\\\":\\\"Paris\\\"}\"}}]},{\"role\":\"tool\",\"tool_call_id\":\"call_HrDZhrMUauvVddKWzVQFJ69Q\",\"content\":\"{\\\"temperature\\\":22,\\\"condition\\\":\\\"sunny\\\"}\"}],\"tools\":[{\"type\":\"function\",\"function\":{\"name\":\"get_weather\",\"description\":\"Get current weather for a city.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"city\":{\"type\":\"string\"}},\"required\":[\"city\"],\"additionalProperties\":false}}}],\"stream\":true,\"stream_options\":{\"include_usage\":true},\"max_tokens\":80,\"temperature\":0}"
+      },
+      "response": {
+        "status": 200,
+        "headers": {
+          "content-type": "text/event-stream; charset=utf-8"
+        },
+        "body": "data: {\"id\":\"chatcmpl-DeAFpaPsUwxt5hjwGgC4BBmlALxwR\",\"object\":\"chat.completion.chunk\",\"created\":1778464909,\"model\":\"gpt-4o-mini-2024-07-18\",\"service_tier\":\"default\",\"system_fingerprint\":\"fp_c83829e916\",\"choices\":[{\"index\":0,\"delta\":{\"role\":\"assistant\",\"content\":\"\",\"refusal\":null},\"logprobs\":null,\"finish_reason\":null}],\"usage\":null,\"obfuscation\":\"2QWjdWPSe\"}\n\ndata: {\"id\":\"chatcmpl-DeAFpaPsUwxt5hjwGgC4BBmlALxwR\",\"object\":\"chat.completion.chunk\",\"created\":1778464909,\"model\":\"gpt-4o-mini-2024-07-18\",\"service_tier\":\"default\",\"system_fingerprint\":\"fp_c83829e916\",\"choices\":[{\"index\":0,\"delta\":{\"content\":\"Paris\"},\"logprobs\":null,\"finish_reason\":null}],\"usage\":null,\"obfuscation\":\"sm5IpQ\"}\n\ndata: {\"id\":\"chatcmpl-DeAFpaPsUwxt5hjwGgC4BBmlALxwR\",\"object\":\"chat.completion.chunk\",\"created\":1778464909,\"model\":\"gpt-4o-mini-2024-07-18\",\"service_tier\":\"default\",\"system_fingerprint\":\"fp_c83829e916\",\"choices\":[{\"index\":0,\"delta\":{\"content\":\" is\"},\"logprobs\":null,\"finish_reason\":null}],\"usage\":null,\"obfuscation\":\"cWsGxqHw\"}\n\ndata: {\"id\":\"chatcmpl-DeAFpaPsUwxt5hjwGgC4BBmlALxwR\",\"object\":\"chat.completion.chunk\",\"created\":1778464909,\"model\":\"gpt-4o-mini-2024-07-18\",\"service_tier\":\"default\",\"system_fingerprint\":\"fp_c83829e916\",\"choices\":[{\"index\":0,\"delta\":{\"content\":\" sunny\"},\"logprobs\":null,\"finish_reason\":null}],\"usage\":null,\"obfuscation\":\"8hNCi\"}\n\ndata: {\"id\":\"chatcmpl-DeAFpaPsUwxt5hjwGgC4BBmlALxwR\",\"object\":\"chat.completion.chunk\",\"created\":1778464909,\"model\":\"gpt-4o-mini-2024-07-18\",\"service_tier\":\"default\",\"system_fingerprint\":\"fp_c83829e916\",\"choices\":[{\"index\":0,\"delta\":{\"content\":\".\"},\"logprobs\":null,\"finish_reason\":null}],\"usage\":null,\"obfuscation\":\"4yKCNDRcwq\"}\n\ndata: {\"id\":\"chatcmpl-DeAFpaPsUwxt5hjwGgC4BBmlALxwR\",\"object\":\"chat.completion.chunk\",\"created\":1778464909,\"model\":\"gpt-4o-mini-2024-07-18\",\"service_tier\":\"default\",\"system_fingerprint\":\"fp_c83829e916\",\"choices\":[{\"index\":0,\"delta\":{},\"logprobs\":null,\"finish_reason\":\"stop\"}],\"usage\":null,\"obfuscation\":\"NvRgg\"}\n\ndata: {\"id\":\"chatcmpl-DeAFpaPsUwxt5hjwGgC4BBmlALxwR\",\"object\":\"chat.completion.chunk\",\"created\":1778464909,\"model\":\"gpt-4o-mini-2024-07-18\",\"service_tier\":\"default\",\"system_fingerprint\":\"fp_c83829e916\",\"choices\":[],\"usage\":{\"prompt_tokens\":103,\"completion_tokens\":5,\"total_tokens\":108,\"prompt_tokens_details\":{\"cached_tokens\":0,\"audio_tokens\":0},\"completion_tokens_details\":{\"reasoning_tokens\":0,\"audio_tokens\":0,\"accepted_prediction_tokens\":0,\"rejected_prediction_tokens\":0}},\"obfuscation\":\"4s7mLtNpZ\"}\n\ndata: [DONE]\n\n"
+      }
+    }
+  ]
+}
@@ -0,0 +1,37 @@
+{
+  "version": 1,
+  "metadata": {
+    "name": "openai-compatible-chat/deepseek-chat-text",
+    "recordedAt": "2026-05-11T02:02:10.220Z",
+    "provider": "deepseek",
+    "route": "openai-compatible-chat",
+    "transport": "http",
+    "model": "deepseek-chat",
+    "tags": [
+      "prefix:openai-compatible-chat",
+      "provider:deepseek",
+      "text",
+      "golden"
+    ]
+  },
+  "interactions": [
+    {
+      "transport": "http",
+      "request": {
+        "method": "POST",
+        "url": "https://api.deepseek.com/v1/chat/completions",
+        "headers": {
+          "content-type": "application/json"
+        },
+        "body": "{\"model\":\"deepseek-chat\",\"messages\":[{\"role\":\"system\",\"content\":\"You are concise.\"},{\"role\":\"user\",\"content\":\"Reply exactly with: Hello!\"}],\"stream\":true,\"stream_options\":{\"include_usage\":true},\"max_tokens\":40,\"temperature\":0}"
+      },
+      "response": {
+        "status": 200,
+        "headers": {
+          "content-type": "text/event-stream; charset=utf-8"
+        },
+        "body": "data: {\"id\":\"159019a5-f981-4103-8b3c-2be37cefc505\",\"object\":\"chat.completion.chunk\",\"created\":1778464929,\"model\":\"deepseek-v4-flash\",\"system_fingerprint\":\"fp_058df29938_prod0820_fp8_kvcache_20260402\",\"choices\":[{\"index\":0,\"delta\":{\"role\":\"assistant\",\"content\":\"\"},\"logprobs\":null,\"finish_reason\":null}],\"usage\":null}\n\ndata: {\"id\":\"159019a5-f981-4103-8b3c-2be37cefc505\",\"object\":\"chat.completion.chunk\",\"created\":1778464929,\"model\":\"deepseek-v4-flash\",\"system_fingerprint\":\"fp_058df29938_prod0820_fp8_kvcache_20260402\",\"choices\":[{\"index\":0,\"delta\":{\"content\":\"Hello\"},\"logprobs\":null,\"finish_reason\":null}],\"usage\":null}\n\ndata: {\"id\":\"159019a5-f981-4103-8b3c-2be37cefc505\",\"object\":\"chat.completion.chunk\",\"created\":1778464929,\"model\":\"deepseek-v4-flash\",\"system_fingerprint\":\"fp_058df29938_prod0820_fp8_kvcache_20260402\",\"choices\":[{\"index\":0,\"delta\":{\"content\":\"!\"},\"logprobs\":null,\"finish_reason\":null}],\"usage\":null}\n\ndata: {\"id\":\"159019a5-f981-4103-8b3c-2be37cefc505\",\"object\":\"chat.completion.chunk\",\"created\":1778464929,\"model\":\"deepseek-v4-flash\",\"system_fingerprint\":\"fp_058df29938_prod0820_fp8_kvcache_20260402\",\"choices\":[{\"index\":0,\"delta\":{\"content\":\"\"},\"logprobs\":null,\"finish_reason\":\"stop\"}],\"usage\":{\"prompt_tokens\":14,\"completion_tokens\":2,\"total_tokens\":16,\"prompt_tokens_details\":{\"cached_tokens\":0},\"prompt_cache_hit_tokens\":0,\"prompt_cache_miss_tokens\":14}}\n\ndata: [DONE]\n\n"
+      }
+    }
+  ]
+}
@@ -0,0 +1,37 @@
+{
+  "version": 1,
+  "metadata": {
+    "name": "openai-compatible-chat/togetherai-llama-3-3-70b-text",
+    "recordedAt": "2026-05-11T02:02:13.341Z",
+    "provider": "togetherai",
+    "route": "openai-compatible-chat",
+    "transport": "http",
+    "model": "meta-llama/Llama-3.3-70B-Instruct-Turbo",
+    "tags": [
+      "prefix:openai-compatible-chat",
+      "provider:togetherai",
+      "text",
+      "golden"
+    ]
+  },
+  "interactions": [
+    {
+      "transport": "http",
+      "request": {
+        "method": "POST",
+        "url": "https://api.together.xyz/v1/chat/completions",
+        "headers": {
+          "content-type": "application/json"
+        },
+        "body": "{\"model\":\"meta-llama/Llama-3.3-70B-Instruct-Turbo\",\"messages\":[{\"role\":\"system\",\"content\":\"You are concise.\"},{\"role\":\"user\",\"content\":\"Reply exactly with: Hello!\"}],\"stream\":true,\"stream_options\":{\"include_usage\":true},\"max_tokens\":40,\"temperature\":0}"
+      },
+      "response": {
+        "status": 200,
+        "headers": {
+          "content-type": "text/event-stream;charset=utf-8"
+        },
+        "body": "data: {\"id\":\"oibreET-3pDw3Z-9f9d9996ce37066b\",\"object\":\"chat.completion.chunk\",\"created\":1778464931,\"choices\":[{\"index\":0,\"text\":\"Hello\",\"logprobs\":null,\"finish_reason\":null,\"seed\":null,\"delta\":{\"token_id\":9906,\"role\":\"assistant\",\"content\":\"Hello\"}}],\"model\":\"meta-llama/Llama-3.3-70B-Instruct-Turbo\",\"usage\":null}\n\ndata: {\"id\":\"oibreET-3pDw3Z-9f9d9996ce37066b\",\"object\":\"chat.completion.chunk\",\"created\":1778464931,\"choices\":[{\"index\":0,\"text\":\"!\",\"logprobs\":null,\"finish_reason\":null,\"seed\":null,\"delta\":{\"token_id\":null,\"role\":\"assistant\",\"content\":\"!\"}}],\"model\":\"meta-llama/Llama-3.3-70B-Instruct-Turbo\",\"usage\":null}\n\ndata: {\"id\":\"oibreET-3pDw3Z-9f9d9996ce37066b\",\"object\":\"chat.completion.chunk\",\"created\":1778464931,\"choices\":[{\"index\":0,\"text\":\"\",\"logprobs\":null,\"finish_reason\":\"stop\",\"seed\":12144769634208630000,\"delta\":{\"token_id\":128009,\"role\":\"assistant\",\"content\":\"\"}}],\"model\":\"meta-llama/Llama-3.3-70B-Instruct-Turbo\",\"usage\":{\"prompt_tokens\":45,\"completion_tokens\":3,\"total_tokens\":48,\"cached_tokens\":0}}\n\ndata: [DONE]\n\n"
+      }
+    }
+  ]
+}
@@ -0,0 +1,38 @@
+{
+  "version": 1,
+  "metadata": {
+    "name": "openai-compatible-chat/togetherai-llama-3-3-70b-tool-call",
+    "recordedAt": "2026-05-11T02:02:14.453Z",
+    "provider": "togetherai",
+    "route": "openai-compatible-chat",
+    "transport": "http",
+    "model": "meta-llama/Llama-3.3-70B-Instruct-Turbo",
+    "tags": [
+      "prefix:openai-compatible-chat",
+      "provider:togetherai",
+      "tool",
+      "tool-call",
+      "golden"
+    ]
+  },
+  "interactions": [
+    {
+      "transport": "http",
+      "request": {
+        "method": "POST",
+        "url": "https://api.together.xyz/v1/chat/completions",
+        "headers": {
+          "content-type": "application/json"
+        },
+        "body": "{\"model\":\"meta-llama/Llama-3.3-70B-Instruct-Turbo\",\"messages\":[{\"role\":\"system\",\"content\":\"Call tools exactly as requested.\"},{\"role\":\"user\",\"content\":\"Call get_weather with city exactly Paris.\"}],\"tools\":[{\"type\":\"function\",\"function\":{\"name\":\"get_weather\",\"description\":\"Get current weather for a city.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"city\":{\"type\":\"string\"}},\"required\":[\"city\"],\"additionalProperties\":false}}}],\"tool_choice\":{\"type\":\"function\",\"function\":{\"name\":\"get_weather\"}},\"stream\":true,\"stream_options\":{\"include_usage\":true},\"max_tokens\":80,\"temperature\":0}"
+      },
+      "response": {
+        "status": 200,
+        "headers": {
+          "content-type": "text/event-stream;charset=utf-8"
+        },
+        "body": "data: {\"id\":\"oibreu3-6Ng1vN-9f9d99a9ba73066b\",\"object\":\"chat.completion.chunk\",\"created\":1778464933,\"choices\":[{\"index\":0,\"role\":\"assistant\",\"text\":\"\",\"logprobs\":null,\"finish_reason\":null,\"delta\":{\"token_id\":null,\"role\":\"assistant\",\"content\":\"\"}}],\"model\":\"meta-llama/Llama-3.3-70B-Instruct-Turbo\"}\n\ndata: {\"id\":\"oibreu3-6Ng1vN-9f9d99a9ba73066b\",\"object\":\"chat.completion.chunk\",\"created\":1778464933,\"choices\":[{\"index\":0,\"text\":\"\",\"logprobs\":null,\"finish_reason\":null,\"delta\":{\"token_id\":null,\"role\":\"assistant\",\"content\":\"\",\"tool_calls\":[{\"index\":0,\"id\":\"call_jue52dtu6iozr0ny9rq357u5\",\"type\":\"function\",\"function\":{\"name\":\"get_weather\",\"arguments\":\"\"}}]}}],\"model\":\"meta-llama/Llama-3.3-70B-Instruct-Turbo\"}\n\ndata: {\"id\":\"oibreu3-6Ng1vN-9f9d99a9ba73066b\",\"object\":\"chat.completion.chunk\",\"created\":1778464933,\"choices\":[{\"index\":0,\"text\":\"\",\"logprobs\":null,\"finish_reason\":\"tool_calls\",\"delta\":{\"token_id\":null,\"role\":\"assistant\",\"content\":\"\",\"tool_calls\":[{\"index\":0,\"function\":{\"arguments\":\"{\\\"city\\\":\\\"Paris\\\"}\"}}]}}],\"model\":\"meta-llama/Llama-3.3-70B-Instruct-Turbo\"}\n\ndata: {\"id\":\"oibreu3-6Ng1vN-9f9d99a9ba73066b\",\"object\":\"chat.completion.chunk\",\"created\":1778464933,\"choices\":[{\"index\":0,\"text\":\"\",\"logprobs\":null,\"finish_reason\":\"tool_calls\",\"seed\":17440360718047570000,\"delta\":{\"token_id\":128009,\"role\":\"assistant\",\"content\":\"\"}}],\"model\":\"meta-llama/Llama-3.3-70B-Instruct-Turbo\",\"usage\":{\"prompt_tokens\":194,\"completion_tokens\":19,\"total_tokens\":213,\"cached_tokens\":0}}\n\ndata: [DONE]\n\n"
+      }
+    }
+  ]
+}
@@ -0,0 +1,143 @@
+{
+  "version": 1,
+  "metadata": {
+    "name": "openai-responses-websocket/openai-responses-websocket-gpt-4-1-mini-tool-loop",
+    "recordedAt": "2026-05-11T02:02:03.284Z",
+    "provider": "openai",
+    "route": "openai-responses-websocket",
+    "transport": "websocket",
+    "model": "gpt-4.1-mini",
+    "tags": [
+      "prefix:openai-responses-websocket",
+      "provider:openai",
+      "transport:websocket",
+      "tool",
+      "tool-loop",
+      "golden"
+    ]
+  },
+  "interactions": [
+    {
+      "transport": "websocket",
+      "open": {
+        "url": "wss://api.openai.com/v1/responses",
+        "headers": {}
+      },
+      "client": [
+        {
+          "kind": "text",
+          "body": "{\"type\":\"response.create\",\"model\":\"gpt-4.1-mini\",\"input\":[{\"role\":\"system\",\"content\":\"Use the get_weather tool exactly once. After the tool result, reply exactly: Paris is sunny.\"},{\"role\":\"user\",\"content\":[{\"type\":\"input_text\",\"text\":\"What is the weather in Paris?\"}]}],\"tools\":[{\"type\":\"function\",\"name\":\"get_weather\",\"description\":\"Get current weather for a city.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"city\":{\"type\":\"string\"}},\"required\":[\"city\"],\"additionalProperties\":false}}],\"store\":false,\"max_output_tokens\":80,\"temperature\":0}"
+        }
+      ],
+      "server": [
+        {
+          "kind": "text",
+          "body": "{\"type\":\"response.created\",\"response\":{\"id\":\"resp_0f8d81b5d3287513016a013897415481a28585571d6710366f\",\"object\":\"response\",\"created_at\":1778464919,\"status\":\"in_progress\",\"background\":false,\"completed_at\":null,\"error\":null,\"frequency_penalty\":0.0,\"incomplete_details\":null,\"instructions\":null,\"max_output_tokens\":80,\"max_tool_calls\":null,\"model\":\"gpt-4.1-mini-2025-04-14\",\"moderation\":null,\"output\":[],\"parallel_tool_calls\":true,\"presence_penalty\":0.0,\"previous_response_id\":null,\"prompt_cache_key\":null,\"prompt_cache_retention\":\"in_memory\",\"reasoning\":{\"effort\":null,\"summary\":null},\"safety_identifier\":null,\"service_tier\":\"auto\",\"store\":false,\"temperature\":0.0,\"text\":{\"format\":{\"type\":\"text\"},\"verbosity\":\"medium\"},\"tool_choice\":\"auto\",\"tools\":[{\"type\":\"function\",\"description\":\"Get current weather for a city.\",\"name\":\"get_weather\",\"parameters\":{\"type\":\"object\",\"properties\":{\"city\":{\"type\":\"string\"}},\"required\":[\"city\"],\"additionalProperties\":false},\"strict\":true}],\"top_logprobs\":0,\"top_p\":1.0,\"truncation\":\"disabled\",\"usage\":null,\"user\":null,\"metadata\":{}},\"sequence_number\":0}"
+        },
+        {
+          "kind": "text",
+          "body": "{\"type\":\"response.in_progress\",\"response\":{\"id\":\"resp_0f8d81b5d3287513016a013897415481a28585571d6710366f\",\"object\":\"response\",\"created_at\":1778464919,\"status\":\"in_progress\",\"background\":false,\"completed_at\":null,\"error\":null,\"frequency_penalty\":0.0,\"incomplete_details\":null,\"instructions\":null,\"max_output_tokens\":80,\"max_tool_calls\":null,\"model\":\"gpt-4.1-mini-2025-04-14\",\"moderation\":null,\"output\":[],\"parallel_tool_calls\":true,\"presence_penalty\":0.0,\"previous_response_id\":null,\"prompt_cache_key\":null,\"prompt_cache_retention\":\"in_memory\",\"reasoning\":{\"effort\":null,\"summary\":null},\"safety_identifier\":null,\"service_tier\":\"auto\",\"store\":false,\"temperature\":0.0,\"text\":{\"format\":{\"type\":\"text\"},\"verbosity\":\"medium\"},\"tool_choice\":\"auto\",\"tools\":[{\"type\":\"function\",\"description\":\"Get current weather for a city.\",\"name\":\"get_weather\",\"parameters\":{\"type\":\"object\",\"properties\":{\"city\":{\"type\":\"string\"}},\"required\":[\"city\"],\"additionalProperties\":false},\"strict\":true}],\"top_logprobs\":0,\"top_p\":1.0,\"truncation\":\"disabled\",\"usage\":null,\"user\":null,\"metadata\":{}},\"sequence_number\":1}"
+        },
+        {
+          "kind": "text",
+          "body": "{\"type\":\"response.output_item.added\",\"item\":{\"id\":\"fc_0f8d81b5d3287513016a01389815f081a2a68583ed9b19c405\",\"type\":\"function_call\",\"status\":\"in_progress\",\"arguments\":\"\",\"call_id\":\"call_wRJoUdcb6w5SIpieBFOKxF6z\",\"name\":\"get_weather\"},\"output_index\":0,\"sequence_number\":2}"
+        },
+        {
+          "kind": "text",
+          "body": "{\"type\":\"response.function_call_arguments.delta\",\"delta\":\"{\\\"\",\"item_id\":\"fc_0f8d81b5d3287513016a01389815f081a2a68583ed9b19c405\",\"obfuscation\":\"EAdvUfYaysnNd3\",\"output_index\":0,\"sequence_number\":3}"
+        },
+        {
+          "kind": "text",
+          "body": "{\"type\":\"response.function_call_arguments.delta\",\"delta\":\"city\",\"item_id\":\"fc_0f8d81b5d3287513016a01389815f081a2a68583ed9b19c405\",\"obfuscation\":\"b0QWJJlscLyl\",\"output_index\":0,\"sequence_number\":4}"
+        },
+        {
+          "kind": "text",
+          "body": "{\"type\":\"response.function_call_arguments.delta\",\"delta\":\"\\\":\\\"\",\"item_id\":\"fc_0f8d81b5d3287513016a01389815f081a2a68583ed9b19c405\",\"obfuscation\":\"m5EV1wzceCkjb\",\"output_index\":0,\"sequence_number\":5}"
+        },
+        {
+          "kind": "text",
+          "body": "{\"type\":\"response.function_call_arguments.delta\",\"delta\":\"Paris\",\"item_id\":\"fc_0f8d81b5d3287513016a01389815f081a2a68583ed9b19c405\",\"obfuscation\":\"QM3FjPoN9Gi\",\"output_index\":0,\"sequence_number\":6}"
+        },
+        {
+          "kind": "text",
+          "body": "{\"type\":\"response.function_call_arguments.delta\",\"delta\":\"\\\"}\",\"item_id\":\"fc_0f8d81b5d3287513016a01389815f081a2a68583ed9b19c405\",\"obfuscation\":\"lug74orAYXtg0e\",\"output_index\":0,\"sequence_number\":7}"
+        },
+        {
+          "kind": "text",
+          "body": "{\"type\":\"response.function_call_arguments.done\",\"arguments\":\"{\\\"city\\\":\\\"Paris\\\"}\",\"item_id\":\"fc_0f8d81b5d3287513016a01389815f081a2a68583ed9b19c405\",\"output_index\":0,\"sequence_number\":8}"
+        },
+        {
+          "kind": "text",
+          "body": "{\"type\":\"response.output_item.done\",\"item\":{\"id\":\"fc_0f8d81b5d3287513016a01389815f081a2a68583ed9b19c405\",\"type\":\"function_call\",\"status\":\"completed\",\"arguments\":\"{\\\"city\\\":\\\"Paris\\\"}\",\"call_id\":\"call_wRJoUdcb6w5SIpieBFOKxF6z\",\"name\":\"get_weather\"},\"output_index\":0,\"sequence_number\":9}"
+        },
+        {
+          "kind": "text",
+          "body": "{\"type\":\"response.completed\",\"response\":{\"id\":\"resp_0f8d81b5d3287513016a013897415481a28585571d6710366f\",\"object\":\"response\",\"created_at\":1778464919,\"status\":\"completed\",\"background\":false,\"completed_at\":1778464920,\"error\":null,\"frequency_penalty\":0.0,\"incomplete_details\":null,\"instructions\":null,\"max_output_tokens\":80,\"max_tool_calls\":null,\"model\":\"gpt-4.1-mini-2025-04-14\",\"moderation\":null,\"output\":[{\"id\":\"fc_0f8d81b5d3287513016a01389815f081a2a68583ed9b19c405\",\"type\":\"function_call\",\"status\":\"completed\",\"arguments\":\"{\\\"city\\\":\\\"Paris\\\"}\",\"call_id\":\"call_wRJoUdcb6w5SIpieBFOKxF6z\",\"name\":\"get_weather\"}],\"parallel_tool_calls\":true,\"presence_penalty\":0.0,\"previous_response_id\":null,\"prompt_cache_key\":null,\"prompt_cache_retention\":\"in_memory\",\"reasoning\":{\"effort\":null,\"summary\":null},\"safety_identifier\":null,\"service_tier\":\"default\",\"store\":false,\"temperature\":0.0,\"text\":{\"format\":{\"type\":\"text\"},\"verbosity\":\"medium\"},\"tool_choice\":\"auto\",\"tools\":[{\"type\":\"function\",\"description\":\"Get current weather for a city.\",\"name\":\"get_weather\",\"parameters\":{\"type\":\"object\",\"properties\":{\"city\":{\"type\":\"string\"}},\"required\":[\"city\"],\"additionalProperties\":false},\"strict\":true}],\"top_logprobs\":0,\"top_p\":1.0,\"truncation\":\"disabled\",\"usage\":{\"input_tokens\":69,\"input_tokens_details\":{\"cached_tokens\":0},\"output_tokens\":15,\"output_tokens_details\":{\"reasoning_tokens\":0},\"total_tokens\":84},\"user\":null,\"metadata\":{}},\"sequence_number\":10}"
+        }
+      ]
+    },
+    {
+      "transport": "websocket",
+      "open": {
+        "url": "wss://api.openai.com/v1/responses",
+        "headers": {}
+      },
+      "client": [
+        {
+          "kind": "text",
+          "body": "{\"type\":\"response.create\",\"model\":\"gpt-4.1-mini\",\"input\":[{\"role\":\"system\",\"content\":\"Use the get_weather tool exactly once. After the tool result, reply exactly: Paris is sunny.\"},{\"role\":\"user\",\"content\":[{\"type\":\"input_text\",\"text\":\"What is the weather in Paris?\"}]},{\"type\":\"function_call\",\"call_id\":\"call_wRJoUdcb6w5SIpieBFOKxF6z\",\"name\":\"get_weather\",\"arguments\":\"{\\\"city\\\":\\\"Paris\\\"}\"},{\"type\":\"function_call_output\",\"call_id\":\"call_wRJoUdcb6w5SIpieBFOKxF6z\",\"output\":\"{\\\"temperature\\\":22,\\\"condition\\\":\\\"sunny\\\"}\"}],\"tools\":[{\"type\":\"function\",\"name\":\"get_weather\",\"description\":\"Get current weather for a city.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"city\":{\"type\":\"string\"}},\"required\":[\"city\"],\"additionalProperties\":false}}],\"store\":false,\"max_output_tokens\":80,\"temperature\":0}"
+        }
+      ],
+      "server": [
+        {
+          "kind": "text",
+          "body": "{\"type\":\"response.created\",\"response\":{\"id\":\"resp_0767cfe3f5d98b2a016a0138994258819485d082e5c78849a4\",\"object\":\"response\",\"created_at\":1778464921,\"status\":\"in_progress\",\"background\":false,\"completed_at\":null,\"error\":null,\"frequency_penalty\":0.0,\"incomplete_details\":null,\"instructions\":null,\"max_output_tokens\":80,\"max_tool_calls\":null,\"model\":\"gpt-4.1-mini-2025-04-14\",\"moderation\":null,\"output\":[],\"parallel_tool_calls\":true,\"presence_penalty\":0.0,\"previous_response_id\":null,\"prompt_cache_key\":null,\"prompt_cache_retention\":\"in_memory\",\"reasoning\":{\"effort\":null,\"summary\":null},\"safety_identifier\":null,\"service_tier\":\"auto\",\"store\":false,\"temperature\":0.0,\"text\":{\"format\":{\"type\":\"text\"},\"verbosity\":\"medium\"},\"tool_choice\":\"auto\",\"tools\":[{\"type\":\"function\",\"description\":\"Get current weather for a city.\",\"name\":\"get_weather\",\"parameters\":{\"type\":\"object\",\"properties\":{\"city\":{\"type\":\"string\"}},\"required\":[\"city\"],\"additionalProperties\":false},\"strict\":true}],\"top_logprobs\":0,\"top_p\":1.0,\"truncation\":\"disabled\",\"usage\":null,\"user\":null,\"metadata\":{}},\"sequence_number\":0}"
+        },
+        {
+          "kind": "text",
+          "body": "{\"type\":\"response.in_progress\",\"response\":{\"id\":\"resp_0767cfe3f5d98b2a016a0138994258819485d082e5c78849a4\",\"object\":\"response\",\"created_at\":1778464921,\"status\":\"in_progress\",\"background\":false,\"completed_at\":null,\"error\":null,\"frequency_penalty\":0.0,\"incomplete_details\":null,\"instructions\":null,\"max_output_tokens\":80,\"max_tool_calls\":null,\"model\":\"gpt-4.1-mini-2025-04-14\",\"moderation\":null,\"output\":[],\"parallel_tool_calls\":true,\"presence_penalty\":0.0,\"previous_response_id\":null,\"prompt_cache_key\":null,\"prompt_cache_retention\":\"in_memory\",\"reasoning\":{\"effort\":null,\"summary\":null},\"safety_identifier\":null,\"service_tier\":\"auto\",\"store\":false,\"temperature\":0.0,\"text\":{\"format\":{\"type\":\"text\"},\"verbosity\":\"medium\"},\"tool_choice\":\"auto\",\"tools\":[{\"type\":\"function\",\"description\":\"Get current weather for a city.\",\"name\":\"get_weather\",\"parameters\":{\"type\":\"object\",\"properties\":{\"city\":{\"type\":\"string\"}},\"required\":[\"city\"],\"additionalProperties\":false},\"strict\":true}],\"top_logprobs\":0,\"top_p\":1.0,\"truncation\":\"disabled\",\"usage\":null,\"user\":null,\"metadata\":{}},\"sequence_number\":1}"
+        },
+        {
+          "kind": "text",
+          "body": "{\"type\":\"response.output_item.added\",\"item\":{\"id\":\"msg_0767cfe3f5d98b2a016a01389b1434819493f3d33e0ec6973c\",\"type\":\"message\",\"status\":\"in_progress\",\"content\":[],\"role\":\"assistant\"},\"output_index\":0,\"sequence_number\":2}"
+        },
+        {
+          "kind": "text",
+          "body": "{\"type\":\"response.content_part.added\",\"content_index\":0,\"item_id\":\"msg_0767cfe3f5d98b2a016a01389b1434819493f3d33e0ec6973c\",\"output_index\":0,\"part\":{\"type\":\"output_text\",\"annotations\":[],\"logprobs\":[],\"text\":\"\"},\"sequence_number\":3}"
+        },
+        {
+          "kind": "text",
+          "body": "{\"type\":\"response.output_text.delta\",\"content_index\":0,\"delta\":\"Paris\",\"item_id\":\"msg_0767cfe3f5d98b2a016a01389b1434819493f3d33e0ec6973c\",\"logprobs\":[],\"obfuscation\":\"fWbkBKTZ5oG\",\"output_index\":0,\"sequence_number\":4}"
+        },
+        {
+          "kind": "text",
+          "body": "{\"type\":\"response.output_text.delta\",\"content_index\":0,\"delta\":\" is\",\"item_id\":\"msg_0767cfe3f5d98b2a016a01389b1434819493f3d33e0ec6973c\",\"logprobs\":[],\"obfuscation\":\"BGuPlDrPchXwG\",\"output_index\":0,\"sequence_number\":5}"
+        },
+        {
+          "kind": "text",
+          "body": "{\"type\":\"response.output_text.delta\",\"content_index\":0,\"delta\":\" sunny\",\"item_id\":\"msg_0767cfe3f5d98b2a016a01389b1434819493f3d33e0ec6973c\",\"logprobs\":[],\"obfuscation\":\"7THzde2pni\",\"output_index\":0,\"sequence_number\":6}"
+        },
+        {
+          "kind": "text",
+          "body": "{\"type\":\"response.output_text.delta\",\"content_index\":0,\"delta\":\".\",\"item_id\":\"msg_0767cfe3f5d98b2a016a01389b1434819493f3d33e0ec6973c\",\"logprobs\":[],\"obfuscation\":\"Eo38bSElfyNmljj\",\"output_index\":0,\"sequence_number\":7}"
+        },
+        {
+          "kind": "text",
+          "body": "{\"type\":\"response.output_text.done\",\"content_index\":0,\"item_id\":\"msg_0767cfe3f5d98b2a016a01389b1434819493f3d33e0ec6973c\",\"logprobs\":[],\"output_index\":0,\"sequence_number\":8,\"text\":\"Paris is sunny.\"}"
+        },
+        {
+          "kind": "text",
+          "body": "{\"type\":\"response.content_part.done\",\"content_index\":0,\"item_id\":\"msg_0767cfe3f5d98b2a016a01389b1434819493f3d33e0ec6973c\",\"output_index\":0,\"part\":{\"type\":\"output_text\",\"annotations\":[],\"logprobs\":[],\"text\":\"Paris is sunny.\"},\"sequence_number\":9}"
+        },
+        {
+          "kind": "text",
+          "body": "{\"type\":\"response.output_item.done\",\"item\":{\"id\":\"msg_0767cfe3f5d98b2a016a01389b1434819493f3d33e0ec6973c\",\"type\":\"message\",\"status\":\"completed\",\"content\":[{\"type\":\"output_text\",\"annotations\":[],\"logprobs\":[],\"text\":\"Paris is sunny.\"}],\"role\":\"assistant\"},\"output_index\":0,\"sequence_number\":10}"
+        },
+        {
+          "kind": "text",
+          "body": "{\"type\":\"response.completed\",\"response\":{\"id\":\"resp_0767cfe3f5d98b2a016a0138994258819485d082e5c78849a4\",\"object\":\"response\",\"created_at\":1778464921,\"status\":\"completed\",\"background\":false,\"completed_at\":1778464923,\"error\":null,\"frequency_penalty\":0.0,\"incomplete_details\":null,\"instructions\":null,\"max_output_tokens\":80,\"max_tool_calls\":null,\"model\":\"gpt-4.1-mini-2025-04-14\",\"moderation\":null,\"output\":[{\"id\":\"msg_0767cfe3f5d98b2a016a01389b1434819493f3d33e0ec6973c\",\"type\":\"message\",\"status\":\"completed\",\"content\":[{\"type\":\"output_text\",\"annotations\":[],\"logprobs\":[],\"text\":\"Paris is sunny.\"}],\"role\":\"assistant\"}],\"parallel_tool_calls\":true,\"presence_penalty\":0.0,\"previous_response_id\":null,\"prompt_cache_key\":null,\"prompt_cache_retention\":\"in_memory\",\"reasoning\":{\"effort\":null,\"summary\":null},\"safety_identifier\":null,\"service_tier\":\"default\",\"store\":false,\"temperature\":0.0,\"text\":{\"format\":{\"type\":\"text\"},\"verbosity\":\"medium\"},\"tool_choice\":\"auto\",\"tools\":[{\"type\":\"function\",\"description\":\"Get current weather for a city.\",\"name\":\"get_weather\",\"parameters\":{\"type\":\"object\",\"properties\":{\"city\":{\"type\":\"string\"}},\"required\":[\"city\"],\"additionalProperties\":false},\"strict\":true}],\"top_logprobs\":0,\"top_p\":1.0,\"truncation\":\"disabled\",\"usage\":{\"input_tokens\":99,\"input_tokens_details\":{\"cached_tokens\":0},\"output_tokens\":6,\"output_tokens_details\":{\"reasoning_tokens\":0},\"total_tokens\":105},\"user\":null,\"metadata\":{}},\"sequence_number\":11}"
+        }
+      ]
+    }
+  ]
+}
@@ -1,6 +1,6 @@
 import { describe, expect } from "bun:test"
 import { Effect } from "effect"
-import { CacheHint, LLM, LLMError } from "../../src"
+import { CacheHint, LLM, LLMError, Usage } from "../../src"
 import { LLMClient } from "../../src/route"
 import * as AnthropicMessages from "../../src/protocols/anthropic-messages"
 import { it } from "../lib/effect"
@@ -110,10 +110,11 @@ describe("Anthropic Messages route", () => {
      expect(response.text).toBe("Hello!")
      expect(response.reasoning).toBe("thinking")
      expect(response.usage).toMatchObject({
-        inputTokens: 5,
+        inputTokens: 6,
        outputTokens: 2,
+        nonCachedInputTokens: 5,
        cacheReadInputTokens: 1,
-        totalTokens: 7,
+        totalTokens: 8,
      })
      expect(response.events.find((event) => event.type === "reasoning-end")).toMatchObject({
        providerMetadata: { anthropic: { signature: "sig_1" } },
@@ -152,7 +153,13 @@ describe("Anthropic Messages route", () => {
        {
          type: "request-finish",
          reason: "tool-calls",
-          usage: { inputTokens: 5, outputTokens: 1, totalTokens: 6, native: { input_tokens: 5, output_tokens: 1 } },
+          usage: new Usage({
+            inputTokens: 5,
+            outputTokens: 1,
+            nonCachedInputTokens: 5,
+            totalTokens: 6,
+            providerMetadata: { anthropic: { input_tokens: 5, output_tokens: 1 } },
+          }),
        },
      ])
    }),
@@ -1,6 +1,6 @@
 import { describe, expect } from "bun:test"
 import { Effect } from "effect"
-import { LLM, LLMError } from "../../src"
+import { LLM, LLMError, Usage } from "../../src"
 import { LLMClient } from "../../src/route"
 import * as Gemini from "../../src/protocols/gemini"
 import { it } from "../lib/effect"
@@ -198,9 +198,10 @@ describe("Gemini route", () => {
      expect(response.reasoning).toBe("thinking")
      expect(response.usage).toMatchObject({
        inputTokens: 5,
-        outputTokens: 2,
-        reasoningTokens: 1,
+        outputTokens: 3,
+        nonCachedInputTokens: 4,
        cacheReadInputTokens: 1,
+        reasoningTokens: 1,
        totalTokens: 7,
      })
      expect(response.events).toEqual([
@@ -210,20 +211,23 @@ describe("Gemini route", () => {
        {
          type: "request-finish",
          reason: "stop",
-          usage: {
+          usage: new Usage({
            inputTokens: 5,
-            outputTokens: 2,
-            reasoningTokens: 1,
+            outputTokens: 3,
+            nonCachedInputTokens: 4,
            cacheReadInputTokens: 1,
+            reasoningTokens: 1,
            totalTokens: 7,
-            native: {
-              promptTokenCount: 5,
-              candidatesTokenCount: 2,
-              totalTokenCount: 7,
-              thoughtsTokenCount: 1,
-              cachedContentTokenCount: 1,
+            providerMetadata: {
+              google: {
+                promptTokenCount: 5,
+                candidatesTokenCount: 2,
+                totalTokenCount: 7,
+                thoughtsTokenCount: 1,
+                cachedContentTokenCount: 1,
+              },
            },
-          },
+          }),
        },
      ])
    }),
@@ -257,12 +261,13 @@ describe("Gemini route", () => {
        {
          type: "request-finish",
          reason: "tool-calls",
-          usage: {
+          usage: new Usage({
            inputTokens: 5,
            outputTokens: 1,
+            nonCachedInputTokens: 5,
            totalTokens: 6,
-            native: { promptTokenCount: 5, candidatesTokenCount: 1 },
-          },
+            providerMetadata: { google: { promptTokenCount: 5, candidatesTokenCount: 1 } },
+          }),
        },
      ])
    }),
@@ -1,7 +1,7 @@
 import { describe, expect } from "bun:test"
 import { Effect, Schema, Stream } from "effect"
 import { HttpClientRequest } from "effect/unstable/http"
-import { LLM, LLMError } from "../../src"
+import { LLM, LLMError, Usage } from "../../src"
 import * as Azure from "../../src/providers/azure"
 import * as OpenAI from "../../src/providers/openai"
 import * as OpenAIChat from "../../src/protocols/openai-chat"
@@ -230,20 +230,23 @@ describe("OpenAI Chat route", () => {
        {
          type: "request-finish",
          reason: "stop",
-          usage: {
+          usage: new Usage({
            inputTokens: 5,
            outputTokens: 2,
-            reasoningTokens: 0,
+            nonCachedInputTokens: 4,
            cacheReadInputTokens: 1,
+            reasoningTokens: 0,
            totalTokens: 7,
-            native: {
-              prompt_tokens: 5,
-              completion_tokens: 2,
-              total_tokens: 7,
-              prompt_tokens_details: { cached_tokens: 1 },
-              completion_tokens_details: { reasoning_tokens: 0 },
+            providerMetadata: {
+              openai: {
+                prompt_tokens: 5,
+                completion_tokens: 2,
+                total_tokens: 7,
+                prompt_tokens_details: { cached_tokens: 1 },
+                completion_tokens_details: { reasoning_tokens: 0 },
+              },
            },
-          },
+          }),
        },
      ])
    }),
@@ -1,7 +1,7 @@
 import { describe, expect } from "bun:test"
 import { ConfigProvider, Effect, Layer, Stream } from "effect"
 import { Headers, HttpClientRequest } from "effect/unstable/http"
-import { LLM, LLMError } from "../../src"
+import { LLM, LLMError, Usage } from "../../src"
 import { Auth, LLMClient, RequestExecutor, WebSocketExecutor } from "../../src/route"
 import * as Azure from "../../src/providers/azure"
 import * as OpenAI from "../../src/providers/openai"
@@ -342,20 +342,23 @@ describe("OpenAI Responses route", () => {
          type: "request-finish",
          reason: "stop",
          providerMetadata: { openai: { responseId: "resp_1", serviceTier: "default" } },
-          usage: {
+          usage: new Usage({
            inputTokens: 5,
            outputTokens: 2,
-            reasoningTokens: 0,
+            nonCachedInputTokens: 4,
            cacheReadInputTokens: 1,
+            reasoningTokens: 0,
            totalTokens: 7,
-            native: {
-              input_tokens: 5,
-              output_tokens: 2,
-              total_tokens: 7,
-              input_tokens_details: { cached_tokens: 1 },
-              output_tokens_details: { reasoning_tokens: 0 },
+            providerMetadata: {
+              openai: {
+                input_tokens: 5,
+                output_tokens: 2,
+                total_tokens: 7,
+                input_tokens_details: { cached_tokens: 1 },
+                output_tokens_details: { reasoning_tokens: 0 },
+              },
            },
-          },
+          }),
        },
      ])
    }),
@@ -411,7 +414,13 @@ describe("OpenAI Responses route", () => {
        {
          type: "request-finish",
          reason: "tool-calls",
-          usage: { inputTokens: 5, outputTokens: 1, totalTokens: 6, native: { input_tokens: 5, output_tokens: 1 } },
+          usage: new Usage({
+            inputTokens: 5,
+            outputTokens: 1,
+            nonCachedInputTokens: 5,
+            totalTokens: 6,
+            providerMetadata: { openai: { input_tokens: 5, output_tokens: 1 } },
+          }),
        },
      ])
    }),
@@ -1,6 +1,7 @@
 import { describe, expect, test } from "bun:test"
 import { Schema } from "effect"
-import { ContentPart, LLMEvent, LLMRequest, ModelID, ModelLimits, ModelRef, ProviderID } from "../src/schema"
+import { ContentPart, LLMEvent, LLMRequest, ModelID, ModelLimits, ModelRef, ProviderID, Usage } from "../src/schema"
+import { ProviderShared } from "../src/protocols/shared"

 const model = new ModelRef({
  id: ModelID.make("fake-model"),
@@ -48,3 +49,30 @@ describe("llm schema", () => {
    expect(ContentPart.guards.media({ type: "text", text: "hi" })).toBe(false)
  })
 })
+
+describe("LLM.Usage", () => {
+  test("subtractTokens clamps non-sensical breakdowns to zero", () => {
+    // Defense against a provider reporting cached_tokens > prompt_tokens or
+    // reasoning_tokens > completion_tokens — the negative would otherwise
+    // round-trip through the pipeline and crash strict downstream schemas.
+    expect(ProviderShared.subtractTokens(5, 3)).toBe(2)
+    expect(ProviderShared.subtractTokens(5, 10)).toBe(0)
+    expect(ProviderShared.subtractTokens(5, undefined)).toBe(5)
+    expect(ProviderShared.subtractTokens(undefined, 3)).toBeUndefined()
+    expect(ProviderShared.subtractTokens(undefined, undefined)).toBeUndefined()
+  })
+
+  test("sumTokens returns undefined only when every input is undefined", () => {
+    expect(ProviderShared.sumTokens(1, 2, 3)).toBe(6)
+    expect(ProviderShared.sumTokens(1, undefined, 3)).toBe(4)
+    expect(ProviderShared.sumTokens(undefined, undefined, undefined)).toBeUndefined()
+    expect(ProviderShared.sumTokens()).toBeUndefined()
+  })
+
+  test("visibleOutputTokens clamps reasoning > output to zero", () => {
+    expect(new Usage({ outputTokens: 10, reasoningTokens: 4 }).visibleOutputTokens).toBe(6)
+    expect(new Usage({ outputTokens: 10 }).visibleOutputTokens).toBe(10)
+    expect(new Usage({ outputTokens: 4, reasoningTokens: 10 }).visibleOutputTokens).toBe(0)
+    expect(new Usage({}).visibleOutputTokens).toBe(0)
+  })
+})
@@ -9,13 +9,6 @@
 - **Output**: creates `migration/<timestamp>_<slug>/migration.sql` and `snapshot.json`.
 - **Tests**: migration tests should read the per-folder layout (no `_journal.json`).

-## Development server
-
- Running `bun dev` from `packages/opencode` starts the live interactive TUI. Do not run it as a blocking foreground command when you need to inspect the result.
- Start it in `tmux` instead: `tmux new-session -d -s opencode-dev 'bun dev'`.
- Capture the current TUI output with: `tmux capture-pane -pt opencode-dev`.
- Stop the session explicitly when done: `tmux kill-session -t opencode-dev`.
-
 # Module shape

 Do not use `export namespace Foo { ... }` for module organization. It is not
@@ -93,6 +93,7 @@ const appBindingCommands = [
  "theme.mode.lock",
  "help.show",
  "docs.open",
+  "app.exit",
  "app.debug",
  "app.console",
  "app.heap_snapshot",
@@ -647,6 +648,11 @@ function App(props: { onSnapshot?: () => Promise<string[]> }) {
        title: "Exit the app",
        slashName: "exit",
        slashAliases: ["quit", "q"],
+        enabled: () => {
+          const current = promptRef.current
+          if (!current?.focused) return true
+          return current.current.input === ""
+        },
        run: () => exit(),
        category: "System",
      },
@@ -779,17 +785,6 @@ function App(props: { onSnapshot?: () => Promise<string[]> }) {
    bindings: tuiConfig.keybinds.gather("app", appBindingCommands),
  }))

-  useBindings(() => ({
-    enabled: () => {
-      const ok = command.matcher.get()
-      if (!ok) return false
-      const current = promptRef.current
-      if (!current?.focused) return true
-      return current.current.input === ""
-    },
-    bindings: tuiConfig.keybinds.gather("app_exit", ["app.exit"]),
-  }))
-
  event.on(TuiEvent.CommandExecute.type, (evt) => {
    command.run(evt.properties.command)
  })
@@ -712,6 +712,7 @@ export function Prompt(props: PromptProps) {
      ...input.traits,
      ...computePromptTraits({
        mode: store.mode,
+        disabled: !!props.disabled,
        autocompleteVisible: !!auto()?.visible,
      }),
    }
@@ -4,6 +4,7 @@ export type PromptMode = "normal" | "shell"

 export interface PromptTraitsInput {
  mode: PromptMode
+  disabled: boolean
  autocompleteVisible: boolean
 }

@@ -15,9 +16,10 @@ export type PromptTraits = EditorTraits & {
 /**
 * Compute the textarea editor traits for the prompt.
 *
- * The OpenTUI managed textarea keymap owns `traits.suspend`. Prompt traits
- * only expose capture/status metadata so focus changes cannot unsuspend the
- * keymap-managed editor mappings.
+ * `traits.suspend` gates the textarea's keybinding actions (backspace,
+ * delete-word, arrow movement, undo/redo, etc.). Shell mode is an active
+ * editing mode — only `disabled` should suspend the textarea, otherwise
+ * users can type in shell mode but cannot delete or move the cursor.
 */
 export function computePromptTraits(input: PromptTraitsInput): PromptTraits {
  const capture =
@@ -28,6 +30,7 @@ export function computePromptTraits(input: PromptTraitsInput): PromptTraits {
      : undefined
  return {
    capture,
+    suspend: input.disabled,
    status: input.mode === "shell" ? "SHELL" : undefined,
    owner: "opencode",
    role: "prompt",
@@ -8,9 +8,9 @@ import {
 import {
  KeymapProvider,
  reactiveMatcherFromSignal,
+  useBindings,
  useKeymap,
  useKeymapSelector,
-  useBindings,
 } from "@opentui/keymap/solid"
 import type { Accessor } from "solid-js"
 import type { TuiConfig } from "./config/tui"
@@ -26,28 +26,6 @@ export { reactiveMatcherFromSignal, useBindings, useKeymapSelector }

 export type OpenTuiKeymap = ReturnType<typeof useKeymap>

-const KEY_ALIASES = {
-  enter: "return",
-  esc: "escape",
-} as const
-
-function expandKeyAliases(input: string) {
-  const result = Object.entries(KEY_ALIASES).reduce(
-    (acc, [alias, key]) => acc.replace(new RegExp(`(^|[+,\\s>])${alias}(?=$|[+,\\s<])`, "gi"), `$1${key}`),
-    input,
-  )
-  if (result === input) return
-  return result
-}
-
-function registerKeyAliases(keymap: OpenTuiKeymap) {
-  return keymap.appendBindingExpander((ctx) => {
-    const key = expandKeyAliases(ctx.input)
-    if (!key) return
-    return [{ key, displays: ctx.displays }]
-  })
-}
-
 const inputCommands = [
  "input.move.left",
  "input.move.right",
@@ -120,13 +98,8 @@ export function formatKeyBindings(
  return formatCommandBindingsExtra(bindings, formatOptions(config))
 }

-export function registerOpencodeKeymap(
-  keymap: OpenTuiKeymap,
-  renderer: CliRenderer,
-  config: Pick<TuiConfig.Resolved, "keybinds" | "leader_timeout">,
-) {
+export function registerOpencodeKeymap(keymap: OpenTuiKeymap, renderer: CliRenderer, config: TuiConfig.Resolved) {
  const offCommaBindings = addons.registerCommaBindings(keymap)
-  const offAliasExpander = registerKeyAliases(keymap)
  const offBaseLayout = addons.registerBaseLayoutFallback(keymap)
  const offLeader = addons.registerTimedLeader(keymap, {
    trigger: config.keybinds.get(LEADER_TOKEN),
@@ -135,17 +108,20 @@ export function registerOpencodeKeymap(
  })
  const offEscape = addons.registerEscapeClearsPendingSequence(keymap)
  const offBackspace = addons.registerBackspacePopsPendingSequence(keymap)
-  const offInputBindings = addons.registerManagedTextareaLayer(keymap, renderer, {
+  const offInputCommands = addons.registerEditBufferCommands(keymap, renderer)
+  const offInputSuspension = addons.registerTextareaMappingSuspension(keymap, renderer)
+  const offInputBindings = keymap.registerLayer({
    enabled: () => renderer.currentFocusedEditor !== null,
    bindings: config.keybinds.gather("input", inputCommands),
  })

  return () => {
    offInputBindings()
+    offInputSuspension()
+    offInputCommands()
    offBackspace()
    offEscape()
    offLeader()
-    offAliasExpander()
    offBaseLayout()
    offCommaBindings()
  }
@@ -1,8 +1,5 @@
 import { createOpencodeClient } from "@opencode-ai/sdk/v2"
 import { SessionID } from "@/session/schema"
-import { Schema } from "effect"
-
-const decodeSessionID = Schema.decodeUnknownSync(SessionID)

 export async function validateSession(input: {
  url: string
@@ -13,11 +10,9 @@ export async function validateSession(input: {
 }) {
  if (!input.sessionID) return

-  let sessionID: SessionID
-  try {
-    sessionID = decodeSessionID(input.sessionID)
-  } catch (error) {
-    throw new Error(`Invalid session ID: ${error instanceof Error ? error.message : "unknown error"}`, { cause: error })
+  const result = SessionID.zod.safeParse(input.sessionID)
+  if (!result.success) {
+    throw new Error(`Invalid session ID: ${result.error.issues.at(0)?.message ?? "unknown error"}`)
  }

  await createOpencodeClient({
@@ -25,5 +20,5 @@ export async function validateSession(input: {
    directory: input.directory,
    fetch: input.fetch,
    headers: input.headers,
-  }).session.get({ sessionID }, { throwOnError: true })
+  }).session.get({ sessionID: result.data }, { throwOnError: true })
 }
@@ -1,6 +1,7 @@
 import { Schema } from "effect"

 import { Identifier } from "@/id/id"
+import { zod } from "@opencode-ai/core/effect-zod"
 import { withStatics } from "@opencode-ai/core/schema"

 const workspaceIdSchema = Schema.String.check(Schema.isStartsWith("wrk")).pipe(Schema.brand("WorkspaceID"))
@@ -10,5 +11,6 @@ export type WorkspaceID = typeof workspaceIdSchema.Type
 export const WorkspaceID = workspaceIdSchema.pipe(
  withStatics((schema: typeof workspaceIdSchema) => ({
    ascending: (id?: string) => schema.make(Identifier.ascending("workspace", id)),
+    zod: zod(schema),
  })),
 )
@@ -4,6 +4,7 @@ import * as Session from "./session"
 import { SessionID, MessageID, PartID } from "./schema"
 import { Provider } from "@/provider/provider"
 import { MessageV2 } from "./message-v2"
+import z from "zod"
 import { Token } from "@/util/token"
 import * as Log from "@opencode-ai/core/util/log"
 import { SessionProcessor } from "./processor"
@@ -18,6 +19,7 @@ import { InstanceState } from "@/effect/instance-state"
 import { isOverflow as overflow, usable } from "./overflow"
 import { makeRuntime } from "@/effect/run-service"
 import { serviceUse } from "@/effect/service-use"
+import { fn } from "@/util/fn"
 import { EventV2 } from "@/v2/event"
 import { SessionEvent } from "@/v2/session-event"

@@ -639,4 +641,15 @@ export async function prune(input: { sessionID: SessionID }) {
  return runPromise((svc) => svc.prune(input))
 }

+export const create = fn(
+  z.object({
+    sessionID: SessionID.zod,
+    agent: z.string(),
+    model: z.object({ providerID: ProviderID.zod, modelID: ModelID.zod }),
+    auto: z.boolean(),
+    overflow: z.boolean().optional(),
+  }),
+  (input) => runPromise((svc) => svc.create(input)),
+)
+
 export * as SessionCompaction from "./compaction"
@@ -758,7 +758,7 @@ NOTE: At any point in time through this workflow you should feel free to ask the
              yield* bus.publish(Session.Event.Error, { sessionID: input.sessionID, error: error.toObject() })
              throw error
            }
-            const model = input.model ?? agent.model ?? (yield* currentModel(input.sessionID))
+            const model = input.model ?? agent.model ?? (yield* lastModel(input.sessionID))
            const userMsg: MessageV2.User = {
              id: input.messageID ?? MessageID.ascending(),
              sessionID: input.sessionID,
@@ -916,17 +916,7 @@ NOTE: At any point in time through this workflow you should feel free to ask the
      return yield* Effect.failCause(exit.cause)
    })

-    const currentModel = Effect.fnUntraced(function* (sessionID: SessionID) {
-      const current = Database.use((db) =>
-        db.select({ model: SessionTable.model }).from(SessionTable).where(eq(SessionTable.id, sessionID)).get(),
-      )
-      if (current?.model) {
-        return {
-          providerID: ProviderID.make(current.model.providerID),
-          modelID: ModelID.make(current.model.id),
-          ...(current.model.variant && current.model.variant !== "default" ? { variant: current.model.variant } : {}),
-        }
-      }
+    const lastModel = Effect.fnUntraced(function* (sessionID: SessionID) {
      const match = yield* sessions.findMessage(sessionID, (m) => m.info.role === "user" && !!m.info.model)
      if (Option.isSome(match) && match.value.info.role === "user") return match.value.info.model
      return yield* provider.defaultModel()
@@ -943,14 +933,7 @@ NOTE: At any point in time through this workflow you should feel free to ask the
        throw error
      }

-      const current = Database.use((db) =>
-        db
-          .select({ agent: SessionTable.agent, model: SessionTable.model })
-          .from(SessionTable)
-          .where(eq(SessionTable.id, input.sessionID))
-          .get(),
-      )
-      const model = input.model ?? ag.model ?? (yield* currentModel(input.sessionID))
+      const model = input.model ?? ag.model ?? (yield* lastModel(input.sessionID))
      const same = ag.model && model.providerID === ag.model.providerID && model.modelID === ag.model.modelID
      const full =
        !input.variant && ag.variant && same
@@ -974,35 +957,34 @@ NOTE: At any point in time through this workflow you should feel free to ask the
        format: input.format,
      }

+      const current = Database.use((db) =>
+        db
+          .select({ agent: SessionTable.agent, model: SessionTable.model })
+          .from(SessionTable)
+          .where(eq(SessionTable.id, input.sessionID))
+          .get(),
+      )
      if (current?.agent !== info.agent) {
-        EventV2.run(
-          SessionEvent.AgentSwitched.Sync,
-          {
-            sessionID: input.sessionID,
-            timestamp: DateTime.makeUnsafe(info.time.created),
-            agent: info.agent,
-          },
-          { bypassExperimentalEventSystem: true },
-        )
+        EventV2.run(SessionEvent.AgentSwitched.Sync, {
+          sessionID: input.sessionID,
+          timestamp: DateTime.makeUnsafe(info.time.created),
+          agent: info.agent,
+        })
      }
      if (
        current?.model?.providerID !== info.model.providerID ||
        current.model.id !== info.model.modelID ||
-        (current.model.variant === "default" ? undefined : current.model.variant) !== info.model.variant
+        current.model.variant !== info.model.variant
      ) {
-        EventV2.run(
-          SessionEvent.ModelSwitched.Sync,
-          {
-            sessionID: input.sessionID,
-            timestamp: DateTime.makeUnsafe(info.time.created),
-            model: {
-              id: Modelv2.ID.make(info.model.modelID),
-              providerID: Modelv2.ProviderID.make(info.model.providerID),
-              variant: Modelv2.VariantID.make(info.model.variant ?? "default"),
-            },
+        EventV2.run(SessionEvent.ModelSwitched.Sync, {
+          sessionID: input.sessionID,
+          timestamp: DateTime.makeUnsafe(info.time.created),
+          model: {
+            id: Modelv2.ID.make(info.model.modelID),
+            providerID: Modelv2.ProviderID.make(info.model.providerID),
+            variant: Modelv2.VariantID.make(info.model.variant ?? "default"),
          },
-          { bypassExperimentalEventSystem: true },
-        )
+        })
      }

      yield* Effect.addFinalizer(() => instruction.clear(info.id))
@@ -1722,7 +1704,7 @@ NOTE: At any point in time through this workflow you should feel free to ask the
          if (cmdAgent?.model) return cmdAgent.model
        }
        if (input.model) return Provider.parseModel(input.model)
-        return yield* currentModel(input.sessionID)
+        return yield* lastModel(input.sessionID)
      })

      yield* getModel(taskModel.providerID, taskModel.modelID, input.sessionID)
@@ -1755,7 +1737,7 @@ NOTE: At any point in time through this workflow you should feel free to ask the
      const userModel = isSubtask
        ? input.model
          ? Provider.parseModel(input.model)
-          : yield* currentModel(input.sessionID)
+          : yield* lastModel(input.sessionID)
        : taskModel

      yield* plugin.trigger(
@@ -1,12 +1,14 @@
 import { Schema } from "effect"

 import { Identifier } from "@/id/id"
+import { zod } from "@opencode-ai/core/effect-zod"
 import { withStatics } from "@opencode-ai/core/schema"

 export const SessionID = Schema.String.check(Schema.isStartsWith("ses")).pipe(
  Schema.brand("SessionID"),
  withStatics((s) => ({
    descending: (id?: string) => s.make(Identifier.descending("session", id)),
+    zod: zod(s),
  })),
 )

@@ -16,6 +18,7 @@ export const MessageID = Schema.String.check(Schema.isStartsWith("msg")).pipe(
  Schema.brand("MessageID"),
  withStatics((s) => ({
    ascending: (id?: string) => s.make(Identifier.ascending("message", id)),
+    zod: zod(s),
  })),
 )

@@ -25,6 +28,7 @@ export const PartID = Schema.String.check(Schema.isStartsWith("prt")).pipe(
  Schema.brand("PartID"),
  withStatics((s) => ({
    ascending: (id?: string) => s.make(Identifier.ascending("part", id)),
+    zod: zod(s),
  })),
 )

@@ -44,10 +44,9 @@ export function define<const Type extends string, Fields extends Schema.Struct.F
 export function run<Def extends SyncEvent.Definition>(
  def: Def,
  data: SyncEvent.Event<Def>["data"],
-  // Temporary escape hatch while the full v2 event system remains experimental.
-  options?: { publish?: boolean; bypassExperimentalEventSystem?: boolean },
+  options?: { publish?: boolean },
 ) {
-  if (!options?.bypassExperimentalEventSystem && !Flag.OPENCODE_EXPERIMENTAL_EVENT_SYSTEM) return
+  if (!Flag.OPENCODE_EXPERIMENTAL_EVENT_SYSTEM) return
  SyncEvent.run(def, data, options)
 }

@@ -269,26 +269,18 @@ export const layer = Layer.effect(
      shell: Effect.fn("V2Session.shell")(function* (_input) {}),
      skill: Effect.fn("V2Session.skill")(function* (_input) {}),
      switchAgent: Effect.fn("V2Session.switchAgent")(function* (input) {
-        EventV2.run(
-          SessionEvent.AgentSwitched.Sync,
-          {
-            sessionID: input.sessionID,
-            timestamp: DateTime.makeUnsafe(Date.now()),
-            agent: input.agent,
-          },
-          { bypassExperimentalEventSystem: true },
-        )
+        EventV2.run(SessionEvent.AgentSwitched.Sync, {
+          sessionID: input.sessionID,
+          timestamp: DateTime.makeUnsafe(Date.now()),
+          agent: input.agent,
+        })
      }),
      switchModel: Effect.fn("V2Session.switchModel")(function* (input) {
-        EventV2.run(
-          SessionEvent.ModelSwitched.Sync,
-          {
-            sessionID: input.sessionID,
-            timestamp: DateTime.makeUnsafe(Date.now()),
-            model: input.model,
-          },
-          { bypassExperimentalEventSystem: true },
-        )
+        EventV2.run(SessionEvent.ModelSwitched.Sync, {
+          sessionID: input.sessionID,
+          timestamp: DateTime.makeUnsafe(Date.now()),
+          model: input.model,
+        })
      }),
      subagent: Effect.fn("V2Session.subagent")(function* (input) {
        const parent = yield* result.get(input.parentID)
@@ -3,27 +3,36 @@ import { computePromptTraits } from "../../../../src/cli/cmd/tui/component/promp

 describe("computePromptTraits", () => {
  test("normal mode without autocomplete only captures tab", () => {
-    const traits = computePromptTraits({ mode: "normal", autocompleteVisible: false })
+    const traits = computePromptTraits({ mode: "normal", disabled: false, autocompleteVisible: false })
    expect(traits.capture).toEqual(["tab"])
-    expect(traits.suspend).toBeUndefined()
+    expect(traits.suspend).toBe(false)
    expect(traits.status).toBeUndefined()
  })

  test("normal mode with autocomplete captures navigation keys", () => {
-    const traits = computePromptTraits({ mode: "normal", autocompleteVisible: true })
+    const traits = computePromptTraits({ mode: "normal", disabled: false, autocompleteVisible: true })
    expect(traits.capture).toEqual(["escape", "navigate", "submit", "tab"])
-    expect(traits.suspend).toBeUndefined()
+    expect(traits.suspend).toBe(false)
    expect(traits.status).toBeUndefined()
  })

-  test("shell mode does not write the keymap-owned suspend trait", () => {
-    const traits = computePromptTraits({ mode: "shell", autocompleteVisible: false })
-    expect(traits.suspend).toBeUndefined()
+  test("shell mode does not suspend the textarea", () => {
+    // Suspending the textarea would gate every keybinding action
+    // (backspace, delete-word-backward, arrow movement, etc.) — see
+    // @opentui/core 0.2.x TextareaRenderable.handleKeyPress. Shell mode is
+    // an active editing mode, so suspend must stay off.
+    const traits = computePromptTraits({ mode: "shell", disabled: false, autocompleteVisible: false })
+    expect(traits.suspend).toBe(false)
  })

  test("shell mode disables capture and labels the prompt", () => {
-    const traits = computePromptTraits({ mode: "shell", autocompleteVisible: false })
+    const traits = computePromptTraits({ mode: "shell", disabled: false, autocompleteVisible: false })
    expect(traits.capture).toBeUndefined()
    expect(traits.status).toBe("SHELL")
  })
+
+  test("disabled suspends regardless of mode", () => {
+    expect(computePromptTraits({ mode: "normal", disabled: true, autocompleteVisible: false }).suspend).toBe(true)
+    expect(computePromptTraits({ mode: "shell", disabled: true, autocompleteVisible: false }).suspend).toBe(true)
+  })
 })
@@ -15,20 +15,20 @@ import { WorkspaceID } from "../../src/control-plane/schema"
 // schema we assert:
 //   1. The Effect decoder (`Schema.decodeUnknownSync`) accepts valid input.
 //   2. The derived Zod (`X.zod.parse`) accepts the same input and returns the
-//      same shape for schemas that still expose Zod statics.
-//   3. Clearly-invalid input is rejected by both paths where both exist.
+//      same shape.
+//   3. Clearly-invalid input is rejected by both paths.
 //
 // The point is to lock down the Schema <-> Zod bridge so a future edit to
 // any input schema can't silently drop or widen a field on one side.

 // Representative valid IDs — the branded schemas require the right prefix
 // (see src/id/id.ts).
-const sessionID = Schema.decodeUnknownSync(SessionID)("ses_01J5Y5H0AH4Q4NXJ6P4C3P5V2K")
-const sessionIDChild = Schema.decodeUnknownSync(SessionID)("ses_01J5Y5H0AH4Q4NXJ6P4C3P5V2L")
-const messageID = Schema.decodeUnknownSync(MessageID)("msg_01J5Y5H0AH4Q4NXJ6P4C3P5V2M")
-const partID = Schema.decodeUnknownSync(PartID)("prt_01J5Y5H0AH4Q4NXJ6P4C3P5V2N")
+const sessionID = SessionID.zod.parse("ses_01J5Y5H0AH4Q4NXJ6P4C3P5V2K")
+const sessionIDChild = SessionID.zod.parse("ses_01J5Y5H0AH4Q4NXJ6P4C3P5V2L")
+const messageID = MessageID.zod.parse("msg_01J5Y5H0AH4Q4NXJ6P4C3P5V2M")
+const partID = PartID.zod.parse("prt_01J5Y5H0AH4Q4NXJ6P4C3P5V2N")
 const projectID = ProjectID.zod.parse("proj-alpha")
-const workspaceID = Schema.decodeUnknownSync(WorkspaceID)("wrk-primary")
+const workspaceID = WorkspaceID.zod.parse("wrk-primary")

 function decodeUnknown<S extends Schema.Top>(schema: S) {
  const decode = Schema.decodeUnknownSync(schema as any)
Author	SHA1	Message	Date
Kit Langton	c20d070b9a	docs(llm): fix stale references in protocols/shared.ts and gemini.ts - subtractTokens JSDoc said the raw payload lives on Usage.native, but that field was renamed to providerMetadata earlier in this PR. - totalTokens JSDoc still described the abandoned "additive" first-pass contract where inputTokens/outputTokens were non-cached / visible only. We landed on inclusive totals; the fallback already covers cache and reasoning. - Removed a duplicate inline comment in Gemini's mapUsage — the function-level comment already explains the visible/reasoning sum and the undefined-when-incomplete rule.	2026-05-10 22:08:12 -04:00
Kit Langton	d048bd6f4b	test(llm): re-record golden scenarios against live providers Verifies the new Usage mapper code against live provider responses for OpenAI Chat, OpenAI Responses, Anthropic, Gemini, DeepSeek, and TogetherAI — 16 fresh recordings, all assertions pass. No existing cassettes were modified; these populate test slots that were previously skipped in replay mode. Recorded via: set -a; source .env.recorded.local; set +a RECORD=true bun test test/provider/*.recorded.test.ts Redactor stripped all auth headers; no secrets in the cassettes.	2026-05-10 22:03:02 -04:00
Kit Langton	ab9b79ef88	refactor(llm): rename Usage.native to providerMetadata Aligns the escape-hatch field name with `LLMEvent.providerMetadata` used elsewhere in this package (and with AI SDK / pydantic-ai / LangChain conventions for the same idea). Two parallel escape hatches having different names was a wart. The raw payload is now wrapped under the provider key — `{ openai: ... }`, `{ anthropic: ... }`, `{ google: ... }`, `{ bedrock: ... }` — using the existing `ProviderMetadata = Record<string, Record<string, unknown>>` schema rather than a flat record. Same shape as `LLMEvent.providerMetadata`, so consumers downstream can read both with the same code. Anthropic's `mergeUsage` merges the per-provider sub-record across `message_start` and `message_delta` instead of spreading at the top level.	2026-05-10 21:42:09 -04:00
Kit Langton	d4ff331052	refactor(llm): inclusive total + non-overlapping breakdown for Usage Final shape after considering ecosystem conventions: inputTokens — inclusive total (matches AI SDK / OpenAI / LangChain) outputTokens — inclusive total (includes reasoning) nonCachedInputTokens — breakdown: fresh prompt cacheReadInputTokens — breakdown: cache hit cacheWriteInputTokens — breakdown: cache write reasoningTokens — subset of outputTokens Invariant: nonCached + cacheRead + cacheWrite = inputTokens reasoningTokens <= outputTokens Why this shape: - `inputTokens` keeps its AI-SDK / OpenAI semantics, so a reader from any major ecosystem sees the number they expect. - The non-overlapping breakdown fields are populated alongside the inclusive totals — consumers read whichever they need without subtracting. This eliminates the underflow bug class (opencode#26620) structurally without diverging on naming. - Aligns with the AI SDK v3 spec proposal (vercel/ai#9921), which adds exactly this kind of non-overlapping breakdown to address the active ecosystem bugs around cache token double-counting and underflow (pydantic-ai#4364, langfuse#12306/#11979, vercel/ai#8349, langchain#32818, langchainjs#10249). Mappers: - OpenAI Chat / Responses / Bedrock: provider reports inclusive totals natively; mapper derives `nonCachedInputTokens` via `ProviderShared.subtractTokens`. - Gemini: `promptTokenCount` is inclusive; `candidatesTokenCount` is exclusive of `thoughtsTokenCount`, so mapper sums those to produce the inclusive `outputTokens`. Only computes the total when the visible component is reported (avoids fabricating an inclusive number from a partial breakdown). - Anthropic: `input_tokens` is non-cached natively; mapper sums it with cache reads/writes to produce the inclusive `inputTokens`. `output_tokens` is inclusive (Anthropic doesn't break thinking out, so `reasoningTokens` stays undefined). Added a `visibleOutputTokens` getter (clamped `outputTokens - reasoningTokens`) as the one safe escape hatch for consumers wanting the non-reasoning view. Added `ProviderShared.sumTokens` to derive an inclusive total from a non-overlapping breakdown, returning `undefined` when every input is undefined (so we don't fabricate a 0).	2026-05-10 20:39:22 -04:00
Kit Langton	f5d199db62	feat(llm): add Usage.totalInputTokens / totalOutputTokens getters Match the `LLMResponse.text` / `reasoning` / `toolCalls` getter pattern in the same file — `usage.totalInputTokens` reads naturally and lives where the Usage data does. Both sums are monotonic under the additive contract, so callers no longer need to remember which fields are non-overlapping. Test fixtures that previously asserted with `usage: { ... }` plain literals are now wrapped with `new Usage({...})` to match the runtime shape the mappers actually produce (an instance, not a struct).	2026-05-10 19:29:41 -04:00
Kit Langton	0d4f8d126f	refactor(llm): drop Usage.totalInput / totalOutput helpers The additive contract delivers value at the mapper boundary — every field is non-overlapping and non-negative, so any caller summing arbitrary subsets is correct by construction. Two-line helpers that just sum three or two known fields add API surface without paying for themselves, and there are no in-tree consumers today. If v2 wants them at integration time, the right place is a getter on the `Schema.Class` (matching the `LLMResponse.text` / `reasoning` / `toolCalls` pattern in the same file), not a static namespace helper.	2026-05-10 19:15:46 -04:00
Kit Langton	478f3ae50c	refactor(llm): trim Usage helpers + Bedrock subtraction Review pass: - Drop `Pick<>` type aliases on `Usage.totalInput` / `Usage.totalOutput` — the helpers can take `Usage` directly since every field is optional. - Collapse Bedrock's nested `subtractTokens(subtractTokens(...))` into a single subtraction against the summed cache subtotals. - Drop arithmetic-walkthrough comments in test fixtures (the raw fixture values are right next to the expected outputs). - Generalize the comment on `mapUsage` in `openai-chat.ts` so the rationale outlives the PR reference.	2026-05-10 13:22:49 -04:00
Kit Langton	b9451175a6	refactor(llm): make LLM.Usage a fully-additive contract Defines a single invariant for `LLM.Usage`: every field is non-negative and every meaningful aggregate is a sum, never a difference. Total billable input = inputTokens + cacheReadInputTokens + cacheWriteInputTokens. Total billable output = outputTokens + reasoningTokens. Adding two non-negatives cannot underflow, so consumers can no longer reproduce the underflow-then-clamp bug class fixed by #26620. Each protocol mapper now enforces the contract at the provider boundary via `ProviderShared.subtractTokens`, which clamps with `Math.max(0, …)` for defense against provider bugs: - OpenAI Chat / Responses: pull `cached_tokens` out of `prompt_tokens` / `input_tokens`; pull `reasoning_tokens` out of `completion_tokens` / `output_tokens`. The provider's `total_tokens` is preserved verbatim. - Gemini: pull `cachedContentTokenCount` out of `promptTokenCount`. Gemini already split visible candidates from thoughts. - Bedrock: pull `cacheReadInputTokens` and `cacheWriteInputTokens` out of `inputTokens`, matching AWS prompt-caching docs. - Anthropic: already non-overlapping per the Messages API; pass through. Adds `Usage.totalInput` / `Usage.totalOutput` helpers for callers that want the merged view, and a regression test covering the clamp behavior. The reasoning underflow fixed in #26620 was the most visible symptom of a broader semantic inconsistency in this package: providers also disagreed on whether `inputTokens` includes cache reads (Anthropic excluded; OpenAI/Gemini/Bedrock included), which would silently double-subtract the moment v2 wired LLM.Usage into Session.getUsage. Normalizing now, pre-integration, closes both holes in one move.	2026-05-10 13:07:58 -04:00