@wlv-zedd/dsh-chatgpt-web 1.0.0 → 1.0.2

package/src/responses/reasoning-envelope.ts

@@ -1,49 +1,49 @@
-/**
- * Opaque signed-reasoning metadata round-trip through Codex's `encrypted_content` slot.
- *
- * Some Responses histories contain signed or redacted reasoning metadata that must be replayed
- * verbatim. Codex round-trips `encrypted_content`, so the bridge preserves that metadata inside
- * the inherited `ocxr1:` + base64(JSON) envelope format.
- *
- * Native OpenAI-encrypted blobs (no ocxr1 prefix) are left untouched by the decoder, and the
- * passthrough scrub strips ocxr1 envelopes before native forwarding.
- */
-
-export const BRIDGE_REASONING_PREFIX = "ocxr1:";
-
-export interface ReasoningEnvelope {
-  /** Opaque reasoning-block signature, if captured. */
-  sig?: string;
-  /** Raw redacted_thinking block data payloads, order preserved. */
-  red?: string[];
-  /**
-   * Hidden thinking text (hideThinkingSummary providers): the signature signs this exact text,
-   * so replay needs it even though the visible summary was suppressed.
-   */
-  txt?: string;
-}
-
-export function encodeReasoningEnvelope(envelope: ReasoningEnvelope): string {
-  return BRIDGE_REASONING_PREFIX + Buffer.from(JSON.stringify(envelope), "utf-8").toString("base64");
-}
-
-/** Decode an ocxr1 envelope; returns null for native (OpenAI-encrypted) blobs or garbage. */
-export function decodeReasoningEnvelope(encryptedContent: string): ReasoningEnvelope | null {
-  if (!encryptedContent.startsWith(BRIDGE_REASONING_PREFIX)) return null;
-  try {
-    const parsed: unknown = JSON.parse(Buffer.from(encryptedContent.slice(BRIDGE_REASONING_PREFIX.length), "base64").toString("utf-8"));
-    if (!parsed || typeof parsed !== "object" || Array.isArray(parsed)) return null;
-    const obj = parsed as { sig?: unknown; red?: unknown };
-    const envelope: ReasoningEnvelope = {};
-    if (typeof obj.sig === "string") envelope.sig = obj.sig;
-    if (Array.isArray(obj.red)) {
-      const red = obj.red.filter((r): r is string => typeof r === "string");
-      if (red.length > 0) envelope.red = red;
-    }
-    const txt = (parsed as { txt?: unknown }).txt;
-    if (typeof txt === "string" && txt.length > 0) envelope.txt = txt;
-    return envelope.sig || envelope.red || envelope.txt ? envelope : null;
-  } catch {
-    return null;
-  }
-}
+/**
+ * Opaque signed-reasoning metadata round-trip through Codex's `encrypted_content` slot.
+ *
+ * Some Responses histories contain signed or redacted reasoning metadata that must be replayed
+ * verbatim. Codex round-trips `encrypted_content`, so the bridge preserves that metadata inside
+ * the inherited `ocxr1:` + base64(JSON) envelope format.
+ *
+ * Native OpenAI-encrypted blobs (no ocxr1 prefix) are left untouched by the decoder, and the
+ * passthrough scrub strips ocxr1 envelopes before native forwarding.
+ */
+
+export const BRIDGE_REASONING_PREFIX = "ocxr1:";
+
+export interface ReasoningEnvelope {
+  /** Opaque reasoning-block signature, if captured. */
+  sig?: string;
+  /** Raw redacted_thinking block data payloads, order preserved. */
+  red?: string[];
+  /**
+   * Hidden thinking text (hideThinkingSummary providers): the signature signs this exact text,
+   * so replay needs it even though the visible summary was suppressed.
+   */
+  txt?: string;
+}
+
+export function encodeReasoningEnvelope(envelope: ReasoningEnvelope): string {
+  return BRIDGE_REASONING_PREFIX + Buffer.from(JSON.stringify(envelope), "utf-8").toString("base64");
+}
+
+/** Decode an ocxr1 envelope; returns null for native (OpenAI-encrypted) blobs or garbage. */
+export function decodeReasoningEnvelope(encryptedContent: string): ReasoningEnvelope | null {
+  if (!encryptedContent.startsWith(BRIDGE_REASONING_PREFIX)) return null;
+  try {
+    const parsed: unknown = JSON.parse(Buffer.from(encryptedContent.slice(BRIDGE_REASONING_PREFIX.length), "base64").toString("utf-8"));
+    if (!parsed || typeof parsed !== "object" || Array.isArray(parsed)) return null;
+    const obj = parsed as { sig?: unknown; red?: unknown };
+    const envelope: ReasoningEnvelope = {};
+    if (typeof obj.sig === "string") envelope.sig = obj.sig;
+    if (Array.isArray(obj.red)) {
+      const red = obj.red.filter((r): r is string => typeof r === "string");
+      if (red.length > 0) envelope.red = red;
+    }
+    const txt = (parsed as { txt?: unknown }).txt;
+    if (typeof txt === "string" && txt.length > 0) envelope.txt = txt;
+    return envelope.sig || envelope.red || envelope.txt ? envelope : null;
+  } catch {
+    return null;
+  }
+}

package/src/responses/schema.ts

@@ -1,172 +1,172 @@
-import * as z from "zod/v4";
-
-const inputTextSchema = z.object({ type: z.literal("input_text"), text: z.string() });
-const plainTextSchema = z.object({ type: z.literal("text"), text: z.string() });
-const inputImageBlockSchema = z.object({
-  type: z.literal("input_image"),
-  // codex-rs ImageDetail: auto|low|high|original (view_image --detail original).
-  detail: z.enum(["auto", "low", "high", "original"]).optional(),
-  image_url: z.string().optional(),
-  file_id: z.string().optional(),
-}).refine(v => typeof v.image_url === "string" || typeof v.file_id === "string", {
-  message: "input_image requires at least one of image_url or file_id",
-});
-const inputFileBlockSchema = z.object({
-  type: z.literal("input_file"),
-  file_id: z.string().optional(),
-  filename: z.string().optional(),
-  file_data: z.string().optional(),
-});
-const outputTextSchema = z.object({ type: z.literal("output_text"), text: z.string() });
-const outputRefusalSchema = z.object({ type: z.literal("refusal"), refusal: z.string() });
-const summaryTextSchema = z.object({ type: z.literal("summary_text"), text: z.string() });
-const reasoningTextSchema = z.object({ type: z.literal("reasoning_text"), text: z.string() });
-// codex-rs FunctionCallOutputContentItem (protocol/src/models.rs): input_text | input_image | encrypted_content.
-const encryptedContentBlockSchema = z.object({ type: z.literal("encrypted_content"), encrypted_content: z.string() });
-
-const inputContentBlockSchema = z.union([inputTextSchema, plainTextSchema, inputImageBlockSchema, inputFileBlockSchema]);
-const outputContentBlockSchema = z.union([outputTextSchema, plainTextSchema, outputRefusalSchema]);
-// Codex tool outputs can contain both input-shaped and output-shaped content blocks.
-const toolOutputContentBlockSchema = z.union([
-  outputTextSchema, plainTextSchema, outputRefusalSchema,
-  inputTextSchema, inputImageBlockSchema, encryptedContentBlockSchema,
-]);
-const toolOutputSchema = z.union([z.string(), z.array(toolOutputContentBlockSchema)]);
-
-const userMessageItemSchema = z.object({
-  type: z.literal("message").optional(),
-  role: z.union([z.literal("user"), z.literal("developer")]),
-  content: z.union([z.string(), z.array(inputContentBlockSchema)]).optional(),
-});
-const systemMessageItemSchema = z.object({
-  type: z.literal("message").optional(),
-  role: z.literal("system"),
-  content: z.union([z.string(), z.array(inputContentBlockSchema)]).optional(),
-});
-const assistantMessageItemSchema = z.object({
-  type: z.literal("message").optional(),
-  role: z.literal("assistant"),
-  content: z.union([z.string(), z.array(outputContentBlockSchema)]).optional(),
-  phase: z.enum(["commentary", "final_answer"]).optional(),
-});
-const agentMessageItemSchema = z.object({
-  type: z.literal("agent_message"),
-  author: z.string().optional(),
-  recipient: z.string().optional(),
-  // MultiAgent V1 sends normal input content. V2 may send only encrypted_content; accept that
-  // shape so the HTTP boundary can reject it before constructing a browser adapter instead of
-  // silently manufacturing an empty task or starting a retryable SSE stream.
-  content: z.union([
-    z.string(),
-    z.array(z.union([inputContentBlockSchema, encryptedContentBlockSchema])),
-  ]).optional(),
-}).loose();
-const reasoningItemSchema = z.object({
-  type: z.literal("reasoning"),
-  id: z.string().optional(),
-  summary: z.array(summaryTextSchema).optional(),
-  content: z.array(reasoningTextSchema).optional(),
-  // Round-tripped opaque payload (native OpenAI encryption OR the proxy's ocxr1 envelope).
-  encrypted_content: z.string().optional(),
-});
-const functionCallItemSchema = z.object({
-  type: z.literal("function_call"),
-  id: z.string().optional(),
-  call_id: z.string().min(1),
-  name: z.string().min(1),
-  namespace: z.string().optional(),
-  arguments: z.string().optional(),
-});
-const functionCallOutputItemSchema = z.object({
-  type: z.literal("function_call_output"),
-  call_id: z.string().min(1),
-  output: toolOutputSchema.optional(),
-});
-const customToolCallItemSchema = z.object({
-  type: z.literal("custom_tool_call"),
-  id: z.string().optional(),
-  call_id: z.string().min(1),
-  name: z.string().min(1),
-  input: z.string(),
-});
-const customToolCallOutputItemSchema = z.object({
-  type: z.literal("custom_tool_call_output"),
-  call_id: z.string().min(1),
-  // codex-rs CustomToolCallOutput carries FunctionCallOutputPayload: string OR content items.
-  output: toolOutputSchema,
-});
-
-export const inputItemSchema = z.union([
-  userMessageItemSchema,
-  systemMessageItemSchema,
-  assistantMessageItemSchema,
-  agentMessageItemSchema,
-  reasoningItemSchema,
-  functionCallItemSchema,
-  functionCallOutputItemSchema,
-  customToolCallItemSchema,
-  customToolCallOutputItemSchema,
-  z.object({ type: z.string() }).loose(),
-]);
-
-export const toolSchema = z.object({
-  type: z.literal("function"),
-  name: z.string().min(1),
-  description: z.string().optional(),
-  parameters: z.record(z.string(), z.unknown()).optional(),
-  strict: z.boolean().optional(),
-});
-
-const builtinToolSchema = z.object({ type: z.string() }).loose();
-
-const hostedToolType = z.enum([
-  "web_search_preview", "file_search", "computer_use_preview",
-  "code_interpreter", "image_generation", "mcp",
-]);
-
-const allowedToolEntrySchema = z.object({ type: z.string(), name: z.string().optional() });
-
-export const toolChoiceSchema = z.union([
-  z.literal("auto"),
-  z.literal("none"),
-  z.literal("required"),
-  z.object({ type: z.literal("function"), name: z.string().min(1) }),
-  z.object({ type: z.literal("custom"), name: z.string().min(1) }),
-  z.object({ type: hostedToolType }),
-  z.object({ type: z.literal("allowed_tools"), mode: z.enum(["auto", "required"]), tools: z.array(allowedToolEntrySchema) }),
-]);
-
-export const reasoningConfigSchema = z.object({
-  effort: z.string().optional(),
-  summary: z.enum(["auto", "concise", "detailed", "none"]).optional(),
-});
-
-export const stopSchema = z.union([z.string(), z.array(z.string()), z.null()]);
-
-export const responsesRequestSchema = z.object({
-  model: z.string().min(1),
-  input: z.union([z.string(), z.array(inputItemSchema)]).optional(),
-  instructions: z.union([z.string(), z.null()]).optional(),
-  tools: z.array(z.union([toolSchema, builtinToolSchema])).optional(),
-  tool_choice: toolChoiceSchema.optional(),
-  max_output_tokens: z.number().optional(),
-  temperature: z.number().optional(),
-  top_p: z.number().optional(),
-  stop: stopSchema.optional(),
-  stream: z.boolean().optional(),
-  reasoning: reasoningConfigSchema.nullable().optional(),
-  store: z.boolean().optional(),
-  previous_response_id: z.string().optional(),
-  parallel_tool_calls: z.boolean().optional(),
-  prompt_cache_key: z.string().optional(),
-  metadata: z.unknown().optional(),
-  user: z.string().optional(),
-  service_tier: z.string().optional(),
-  presence_penalty: z.number().optional(),
-  frequency_penalty: z.number().optional(),
-  background: z.unknown().optional(),
-  include: z.unknown().optional(),
-  prompt: z.unknown().optional(),
-  text: z.unknown().optional(),
-  truncation: z.unknown().optional(),
-});
+import * as z from "zod/v4";
+
+const inputTextSchema = z.object({ type: z.literal("input_text"), text: z.string() });
+const plainTextSchema = z.object({ type: z.literal("text"), text: z.string() });
+const inputImageBlockSchema = z.object({
+  type: z.literal("input_image"),
+  // codex-rs ImageDetail: auto|low|high|original (view_image --detail original).
+  detail: z.enum(["auto", "low", "high", "original"]).optional(),
+  image_url: z.string().optional(),
+  file_id: z.string().optional(),
+}).refine(v => typeof v.image_url === "string" || typeof v.file_id === "string", {
+  message: "input_image requires at least one of image_url or file_id",
+});
+const inputFileBlockSchema = z.object({
+  type: z.literal("input_file"),
+  file_id: z.string().optional(),
+  filename: z.string().optional(),
+  file_data: z.string().optional(),
+});
+const outputTextSchema = z.object({ type: z.literal("output_text"), text: z.string() });
+const outputRefusalSchema = z.object({ type: z.literal("refusal"), refusal: z.string() });
+const summaryTextSchema = z.object({ type: z.literal("summary_text"), text: z.string() });
+const reasoningTextSchema = z.object({ type: z.literal("reasoning_text"), text: z.string() });
+// codex-rs FunctionCallOutputContentItem (protocol/src/models.rs): input_text | input_image | encrypted_content.
+const encryptedContentBlockSchema = z.object({ type: z.literal("encrypted_content"), encrypted_content: z.string() });
+
+const inputContentBlockSchema = z.union([inputTextSchema, plainTextSchema, inputImageBlockSchema, inputFileBlockSchema]);
+const outputContentBlockSchema = z.union([outputTextSchema, plainTextSchema, outputRefusalSchema]);
+// Codex tool outputs can contain both input-shaped and output-shaped content blocks.
+const toolOutputContentBlockSchema = z.union([
+  outputTextSchema, plainTextSchema, outputRefusalSchema,
+  inputTextSchema, inputImageBlockSchema, encryptedContentBlockSchema,
+]);
+const toolOutputSchema = z.union([z.string(), z.array(toolOutputContentBlockSchema)]);
+
+const userMessageItemSchema = z.object({
+  type: z.literal("message").optional(),
+  role: z.union([z.literal("user"), z.literal("developer")]),
+  content: z.union([z.string(), z.array(inputContentBlockSchema)]).optional(),
+});
+const systemMessageItemSchema = z.object({
+  type: z.literal("message").optional(),
+  role: z.literal("system"),
+  content: z.union([z.string(), z.array(inputContentBlockSchema)]).optional(),
+});
+const assistantMessageItemSchema = z.object({
+  type: z.literal("message").optional(),
+  role: z.literal("assistant"),
+  content: z.union([z.string(), z.array(outputContentBlockSchema)]).optional(),
+  phase: z.enum(["commentary", "final_answer"]).optional(),
+});
+const agentMessageItemSchema = z.object({
+  type: z.literal("agent_message"),
+  author: z.string().optional(),
+  recipient: z.string().optional(),
+  // MultiAgent V1 sends normal input content. V2 may send only encrypted_content; accept that
+  // shape so the HTTP boundary can reject it before constructing a browser adapter instead of
+  // silently manufacturing an empty task or starting a retryable SSE stream.
+  content: z.union([
+    z.string(),
+    z.array(z.union([inputContentBlockSchema, encryptedContentBlockSchema])),
+  ]).optional(),
+}).loose();
+const reasoningItemSchema = z.object({
+  type: z.literal("reasoning"),
+  id: z.string().optional(),
+  summary: z.array(summaryTextSchema).optional(),
+  content: z.array(reasoningTextSchema).optional(),
+  // Round-tripped opaque payload (native OpenAI encryption OR the proxy's ocxr1 envelope).
+  encrypted_content: z.string().optional(),
+});
+const functionCallItemSchema = z.object({
+  type: z.literal("function_call"),
+  id: z.string().optional(),
+  call_id: z.string().min(1),
+  name: z.string().min(1),
+  namespace: z.string().optional(),
+  arguments: z.string().optional(),
+});
+const functionCallOutputItemSchema = z.object({
+  type: z.literal("function_call_output"),
+  call_id: z.string().min(1),
+  output: toolOutputSchema.optional(),
+});
+const customToolCallItemSchema = z.object({
+  type: z.literal("custom_tool_call"),
+  id: z.string().optional(),
+  call_id: z.string().min(1),
+  name: z.string().min(1),
+  input: z.string(),
+});
+const customToolCallOutputItemSchema = z.object({
+  type: z.literal("custom_tool_call_output"),
+  call_id: z.string().min(1),
+  // codex-rs CustomToolCallOutput carries FunctionCallOutputPayload: string OR content items.
+  output: toolOutputSchema,
+});
+
+export const inputItemSchema = z.union([
+  userMessageItemSchema,
+  systemMessageItemSchema,
+  assistantMessageItemSchema,
+  agentMessageItemSchema,
+  reasoningItemSchema,
+  functionCallItemSchema,
+  functionCallOutputItemSchema,
+  customToolCallItemSchema,
+  customToolCallOutputItemSchema,
+  z.object({ type: z.string() }).loose(),
+]);
+
+export const toolSchema = z.object({
+  type: z.literal("function"),
+  name: z.string().min(1),
+  description: z.string().optional(),
+  parameters: z.record(z.string(), z.unknown()).optional(),
+  strict: z.boolean().optional(),
+});
+
+const builtinToolSchema = z.object({ type: z.string() }).loose();
+
+const hostedToolType = z.enum([
+  "web_search_preview", "file_search", "computer_use_preview",
+  "code_interpreter", "image_generation", "mcp",
+]);
+
+const allowedToolEntrySchema = z.object({ type: z.string(), name: z.string().optional() });
+
+export const toolChoiceSchema = z.union([
+  z.literal("auto"),
+  z.literal("none"),
+  z.literal("required"),
+  z.object({ type: z.literal("function"), name: z.string().min(1) }),
+  z.object({ type: z.literal("custom"), name: z.string().min(1) }),
+  z.object({ type: hostedToolType }),
+  z.object({ type: z.literal("allowed_tools"), mode: z.enum(["auto", "required"]), tools: z.array(allowedToolEntrySchema) }),
+]);
+
+export const reasoningConfigSchema = z.object({
+  effort: z.string().optional(),
+  summary: z.enum(["auto", "concise", "detailed", "none"]).optional(),
+});
+
+export const stopSchema = z.union([z.string(), z.array(z.string()), z.null()]);
+
+export const responsesRequestSchema = z.object({
+  model: z.string().min(1),
+  input: z.union([z.string(), z.array(inputItemSchema)]).optional(),
+  instructions: z.union([z.string(), z.null()]).optional(),
+  tools: z.array(z.union([toolSchema, builtinToolSchema])).optional(),
+  tool_choice: toolChoiceSchema.optional(),
+  max_output_tokens: z.number().optional(),
+  temperature: z.number().optional(),
+  top_p: z.number().optional(),
+  stop: stopSchema.optional(),
+  stream: z.boolean().optional(),
+  reasoning: reasoningConfigSchema.nullable().optional(),
+  store: z.boolean().optional(),
+  previous_response_id: z.string().optional(),
+  parallel_tool_calls: z.boolean().optional(),
+  prompt_cache_key: z.string().optional(),
+  metadata: z.unknown().optional(),
+  user: z.string().optional(),
+  service_tier: z.string().optional(),
+  presence_penalty: z.number().optional(),
+  frequency_penalty: z.number().optional(),
+  background: z.unknown().optional(),
+  include: z.unknown().optional(),
+  prompt: z.unknown().optional(),
+  text: z.unknown().optional(),
+  truncation: z.unknown().optional(),
+});