@pure01fx/dsh-openai-codex-auth 0.11.1 → 0.12.0

package/CHANGELOG.md

@@ -2,6 +2,32 @@
 
 ## Unreleased
 
+## 0.12.0
+
+### Migration
+
+- Migrated to DSH 0.1.7-rc.2 SDK
+- Updated tool message format: tool results are now role:'tool' with top-level toolCallId/isError
+- Migrated shell.run to shell.execute with result() method
+- Updated peer dependencies to 0.1.7-rc.2
+- Updated cordis to ~4.0.4
+- Preserved developer messages in HTTP/WS wire requests
+- Restored successful-result-only matched provenance in cloud context
+- Added mismatch/failed-result regressions in cloud context tests
+
+### Key Changes
+
+- Tool messages: role:'tool' with top-level toolCallId/isError (no longer tool-result content blocks)
+- Shell: execute() then .result() instead of run()
+- Developer messages: preserved in Responses API requests
+- Cloud context: requires toolCallId === source.callId and !isError
+
+## 0.11.2
+
+- Requests the live model catalog as Codex 0.155.0 so version-gated GPT-6 Sol can be advertised; preserves dynamic account-specific discovery and routes exact Sol requests through Responses Lite when catalog metadata is unavailable. Fast remains catalog-authorized.
+- Classifies `slow_down` as a rate limit with bounded provider wait hints, recognizes credit/organization/project quota exhaustion, and distinguishes `INVALID_PROMPT` across HTTP, SSE, and WebSocket errors without retrying permanent failures or switching transports.
+- Preserves validated image request IDs and HTTP status on rejected responses and malformed JSON/UTF-8, without exposing credential echoes or retrying ambiguous image POSTs.
+
 ## 0.11.1
 
 - Adds live network status and process-wide request, retry, and received-payload statistics to the quota panel, distinguishing network activity from model output.
@@ -125,4 +151,4 @@
 - Rejects redirects for credential-bearing OAuth, Catalog, Responses, and subscription-usage requests.
 - Keeps OAuth credentials and account authority host-side, validates credential-bearing endpoint authorities, bounds request/response/replay/WebSocket state, and refuses managed/external credential mismatches.
 - Pins Host/runtime contracts to DSH rc.6 and includes synchronized JavaScript/declaration artifacts plus the complete native transport modules.
-- Keeps the package name `@pure01fx/dsh-openai-codex-auth`; a rename is intentionally deferred.
+- Keeps the package name `@pure01fx/dsh-openai-codex-auth`; a rename is intentionally deferred.

package/CODEX-COMPATIBILITY.md

@@ -1,5 +1,59 @@
 # Codex 云端能力兼容性跟踪
 
+## 已实现：Sol 与错误协议增量适配（0.11.2）
+
+参考上游 `c44deff7b1083e9660ac55d02122481f1cdf139b`，在既有 0.11.1 工作树上完成以下必要改动。此节取代下方历史审查中相应的“待适配”结论；不代表完整上游运行时对齐。
+
+- 模型目录仍动态获取，查询 `client_version` 升至 `0.155.0`，满足上游 Sol 条目的最低版本。仅为缺失目录 metadata 的精确 `gpt-6-sol` 增加 Lite 路由后备；正常目录的 `use_responses_lite` 仍是依据，Fast 仍须目录授权，未知模型不按名称前缀猜测协议。
+- HTTP 错误正文、SSE `response.failed` 与 WS `error` / `response.failed` 共用安全分类。`slow_down` 按限流处理并保留有界等待提示；新增三类 credit/organization/project 额度错误映射宿主 `QUOTA`；`invalid_prompt` 映射独立的 `INVALID_PROMPT`。这些失败不会在传输内重试、重连或回退 HTTP；原有 WS error 包装的限流代码 `RATE_LIMITED` 保持兼容。
+- 图片生成/编辑 HTTP 失败及 JSON/UTF-8 解码失败，通过正式 LlmError failure 元数据保留状态码和校验后的 requestId。图片专用响应头优先；拒绝非法 ID 与 token/account 回显，不增加图片 POST 重试。
+- 新增来自上游的精简 Sol fixture（含全部六档 reasoning effort），覆盖动态目录、最低版本过滤的模拟服务、metadata 优先、缺目录后备、Fast 权限；HTTP/SSE/WS 回归覆盖上述错误的终止行为和隐私边界。模拟版本过滤验证客户端请求条件，不证明真实账号一定获得 Sol。
+
+验证：`pnpm build`、`pnpm test`（26 文件 / 467 项）、`pnpm typecheck`、`git diff --check` 全部通过；生成产物与新鲜编译一致。已更新相关 `lib/` 产物和 CHANGELOG，保留全部已有工作树改动。没有运行真实账号推理、查询线上目录或发布/替换常驻 GUI 插件；当前进程是否使用新构建未验证。Provider/cloud 原始跟踪提交仍分别保持 `ddf04ad` / `b348fc26`，为集合 0.3.7 准备插件版本 0.11.2；尚未发布。
+
+## 上游增量审查：064e701b（插件 0.11.1 工作树）
+
+本次 `../codex` 从 `b348fc26674189f758d5941cdab3f78f258b2aa7` fast-forward 到 `064e701b0fa4f5c0dcaf195a565d088f99bb6af8`，新增 596 个提交；更新后 HEAD 与本次 fetch 的 origin/main 一致，工作树干净。插件以当前已有未提交改动的 0.11.1 工作树为审查对象（Git HEAD 为 `188abd13430b3fff7782c1ee87e5809c4a2e8e96`），不是仅审查已提交版本。
+
+结论：建议做针对性的兼容适配；目前证据不支持“必须整体重写”或“已完整对齐最新上游”。本次只更新上游并记录审查，没有修改插件运行时代码、跟踪常量或发布版本。
+
+### 优先适配：失败响应分类
+
+上游 `codex-rs/codex-api/src/sse/responses.rs` 的变化直接影响现有推理路径：
+
+- `slow_down` 现在与 `rate_limit_exceeded` 一样表示限流，并解析服务端建议等待时间。插件 `src/responses.ts:333` 的 `responsesFailure` 仅对后者返回 `RATE_LIMIT`，前者落入 `SERVER`。
+- `credit_balance_exhausted`、`organization_spend_limit_exceeded`、`project_spend_limit_exceeded` 现在明确归为额度耗尽。插件未按这些 code 分类；若 message 未命中宿主的额度文本识别，会落入 `SERVER`，影响不可重试额度错误的判断。
+- 已用现有 `lib/responses.js` 直接调用复现：上述四个 code 搭配中性 message 均返回 `SERVER`，对照 `rate_limit_exceeded` 返回 `RATE_LIMIT`。这是本地确定性复现，不是线上故障记录。
+- `bio_policy` 在上游已独立分类；插件仍返回不可重试的 `INVALID_REQUEST`。这是诊断粒度差异，优先级低于限流和额度分类。
+
+建议首个适配单元覆盖错误分类、等待时间，以及 HTTP/SSE、WebSocket 两条传输中的终止/重试语义；仅修改版本号不能解决差异。
+
+### 可按需要扩展的能力
+
+- **WebSocket 握手稳定性**：本次上游将已有 HTTP 基础设施 cookie store 接入 WSS 握手，并从成功或拒绝的 upgrade 响应回收 Set-Cookie（`websocket-client/src/lib.rs:145`）。插件没有对应 cookie store。可作为下一优先级的稳定性适配，但未验证当前账号存在握手故障。HTTP cookie store 是上次基线已有能力，不能将整个缺口归为本次新增；实现时应遵循 host/TLS 和基础设施 cookie allowlist，不共享账号/会话 cookie。
+- **WebSocket 空增量续传**：上游 `core/src/client.rs:1377` 放宽为允许与已完成上下文相等的 input；插件 `src/native-websocket-session.ts:83` 默认仍要求严格增长，仅预热例外。可减少无新增 input 时的全量重发，当前没有数据丢失证据。
+- **目录 access programs**：新增可选 `available_access_programs`，插件未建模；需要对应模型访问计划选择时再接入，不能单凭字段新增判定目录解析失败。
+
+- **工作区路由与 gateway OAuth**：上游增加账号工作区路由发现以及独立 gateway 登录。插件仍限制凭据端点为 chatgpt.com 或 loopback（`src/endpoint.ts`），没有对应发现流程。需要面向这类工作区或网关时再设计端点信任、账号绑定及切换后的缓存/连接隔离。上游 `login/src/auth/manager/workspace_routing.rs:63` 明确保留无 resolver 调用者的旧行为，不能据此认定普通 ChatGPT 登录已失效。
+- **推理强度增量更新**：目录新增 `supports_reasoning_effort_updates`；上游以 `configuration_update` 历史项保留请求基线、减少切换 effort 的缓存损失。该路径受功能开关、provider 和模型能力共同控制（`core/src/client.rs:548`）。插件继续使用请求级参数；属于缓存与功能对齐项目，尚无证据表明普通请求因此不兼容。
+- **图片文件引用**：`protocol/src/models.rs:897` 新增 `ImageReference::File { file_id }`，原有 `image_url` wire 形状仍保留。插件继续发送内联图片；搜索结果只提取 `image_url` 图片，不能将不认识的 file_id 视为已下载附件。可在明确上传、下载和账号作用域后单独接入。
+
+### 深入评估：推理增量更新与 file_id
+
+推理增量更新建议先做有开关的实验，暂不默认启用。插件当前 `responses.ts:245` 将 effort 放在请求属性中，`native-websocket-session.ts:57` 对请求属性做复用哈希，因此 effort 改变会使增量链不满足复用条件。保持请求基线并追加 configuration_update 有明确的客户端复用收益机制，但服务端缓存收益尚未实测。真正成本是持久化：现有 `replay.ts:7` 只记录 message/reasoning/function_call，不能直接表示宿主生成的配置更新及其历史位置。需先确认宿主能否提供稳定的轮次、分叉、压缩边界和持久化扩展点，再决定插件内扩展是否足够，不能只做进程内 sessionId→effort 缓存。验收应覆盖 medium→high→medium、断线/重启、分叉、失败重试、成功/失败压缩、模型切换以及能力缺失时保留请求级 effort。收益验证比较匹配历史的缓存输入量、请求字节数和首 token 延迟，不预设改善比例。
+
+file_id 建议暂缓产品化，先验证后端契约。补充检查发现上游 `attachment-store/src/lib.rs:161` 提供的 InlineAttachmentStore 直接返回原字节；`core/src/thread_manager.rs:498` 默认图片存储仍使用它。`core/src/image_preparation.rs:363` 接受可注入存储返回 File，并在上传失败时回退内联。因而协议存在 file_id 不等于默认 Codex 已经提供现成可复制的图片上传闭环。`codex-api/src/files.rs` 的通用/连接器文件上传也不能未经验证就当成 Responses 图片上传。下一步应以一张合成小图验证上传→finalize→同账号 Responses 识图→引用失效/解析契约；若不能确认接受性与生命周期，继续内联。后续实现优先将远端 ID 作为插件内部、按账号与内容隔离的缓存，保留 DSH 本地附件作恢复依据；应对重新上传、账号切换、并发、取消、签名 URL 凭据隔离，并避免中途改变历史图片表示破坏 WS 前缀。现有 WS 增量已避免重复发送不变历史，因此收益主要在 HTTP、多次新连接或跨请求复用，不能直接宣称节省视觉 token。
+
+本轮为源码评估，无线上请求和运行时代码修改；为读取上游接口追加了 `codex-rs/attachment-store` 稀疏检出目录。
+
+### 当前无需随上游机械改动的接口
+
+本次区间 `codex-api/src/search.rs`、`images.rs`、`endpoint/images.rs` 无变化；`endpoint/search.rs` 保留 `alpha/search`，主要是内部常量和测试类型调整。普通 Responses HTTP/WebSocket 仍使用 `/responses`，移除的是 Guardian 专用分支，不要求插件改路由。模型目录支持自定义 URL 和更多可选元数据，不等于默认目录的旧字段全部失效。
+
+### 验证与边界
+
+`pnpm test`：26 个文件、384 项测试全部通过；`pnpm typecheck` 通过；测试包括现有生成产物与新鲜 TypeScript 编译的一致性校验。未执行真实 OAuth、刷新或付费推理；现有回归通过不覆盖上游所有新增行为。原 Provider 基线 `ddf04ad` 与云能力基线 `b348fc26` 保持不变。下方历史章节中的旧缺口、旧测试数量和账号验证不能自动视为当前工作树结论。
+
 ## DSH 0.1.5-rc.1 版本适配
 
 `0.10.1` 迁移 ToolCallId、精确 SDK 依赖和客户端服务提供包，隐藏策略在目录发布投影之前执行。`tests/target-directory.spec.ts` 执行官方 shell 的真实 store/Zustand/Immer 与目标目录/catalog 类，使用合成 Remote 响应和投影输入；这不是浏览器联合验收。`tests/target-services.spec.ts` 使用真实 Cordis/LlmRuntime/ToolRuntime、真实 loopback HTTP/SSE/WebSocket，验证工具 ID、工具执行、含加密 reasoning 的历史回放、永久错误和取消，并确认 WS 没有偷偷回退 HTTP。实际 socket 测试也修复了 ws 成功回调为 null 时的误判。

package/README.md

@@ -17,7 +17,7 @@
 
 ## 快速开始
 
-版本 `0.11.1` 明确适配 DeepSeek Harness `0.1.5-rc.1`，SDK peers/runtime helpers 精确锁定该版本；不声明兼容旧 SDK。旧 DSH `0.1.1-rc.2` profile 应保留原插件 `0.10.0`，升级时在独立 profile 安装新版插件并保留原 profile/锁文件以便回滚。无需迁移账号文件或浏览器隐藏列表；`dsh.engines` 仅作兼容性说明，不当作宿主硬门禁。
+版本 `0.12.0` 明确适配 DeepSeek Harness `0.1.7-rc.2`，SDK peers/runtime helpers 精确锁定该版本；不声明兼容旧 SDK。旧 DSH `0.1.5-rc.1` profile 应保留原插件 `0.11.2`，升级时在独立 profile 安装新版插件并保留原 profile/锁文件以便回滚。无需迁移账号文件或浏览器隐藏列表；`dsh.engines` 仅作兼容性说明，不当作宿主硬门禁。
 
 本版本通过真实目标 LLM/tools 服务与本地 HTTP/SSE/WebSocket 回放、错误/取消回归；模型隐藏覆盖真实目录/catalog/store 的投影更新、重连和重新加载。验证命令为 `pnpm typecheck && pnpm build && pnpm test`；当前版本未执行线上 OAuth/付费模型验收。
 
@@ -356,4 +356,4 @@ node --check client.js
 
 ## License
 
-[MIT](./LICENSE)
+[MIT](./LICENSE)

package/lib/cloud-context.js

@@ -14,16 +14,6 @@ export function chooseCloudModel(params, models) {
 export function visibleCloudImages(messages) {
     const images = [];
     const calls = new Set();
-    const visit = (blocks, depth = 0) => {
-        if (depth > 16)
-            return;
-        for (const block of blocks) {
-            if (block.type === 'image')
-                images.push(block.attachment);
-            else if (block.type === 'tool-result' && !block.isError && calls.has(block.toolCallId))
-                visit(block.content, depth + 1);
-        }
-    };
     for (const message of messages) {
         if (message.role === 'assistant') {
             for (const block of message.content)
@@ -36,10 +26,16 @@ export function visibleCloudImages(messages) {
                 if (block.type === 'image')
                     images.push(block.attachment);
         }
-        else if (message.role === 'user' && message.source.kind === 'tool' && calls.has(message.source.callId)) {
+        else if (message.role === 'tool' && message.source.kind === 'tool') {
+            // Require successful result with matching callId and earlier matching call
+            if (message.isError || message.toolCallId !== message.source.callId)
+                continue;
+            if (!calls.has(message.source.callId))
+                continue;
+            // Tool messages with images in their content
             for (const block of message.content)
-                if (block.type === 'tool-result' && block.toolCallId === message.source.callId && !block.isError)
-                    visit(block.content);
+                if (block.type === 'image')
+                    images.push(block.attachment);
         }
     }
     return images;

package/lib/cloud-http.js

@@ -1,5 +1,5 @@
 /** Nonstreaming Codex cloud requests. Never retry ambiguous POST failures. */
-import { attributionHeaders, LlmError } from '@deepseek-ai/dsh-llm';
+import { attributionHeaders, LlmError, ProviderRequestId } from '@deepseek-ai/dsh-llm';
 import { nativeCodexEndpoint } from './endpoint.js';
 const fail = (message, code = 'CODEX_CLOUD_FAILED') => new LlmError(message, code);
 function positive(value, fallback) {
@@ -93,6 +93,14 @@ export class NativeCodexCloudClient {
                 catch {
                     throw fail(signal.aborted ? 'Codex cloud request cancelled or timed out' : 'Codex cloud request failed; it was not retried', signal.aborted ? 'ABORTED' : 'CODEX_CLOUD_FAILED');
                 }
+                const rawRequestId = path.startsWith('images/')
+                    ? response.headers.get('x-codex-imagegen-request-id') ?? response.headers.get('x-request-id')
+                    : response.headers.get('x-request-id');
+                // Validate before branding: response headers are untrusted, including credential echoes.
+                const requestId = rawRequestId && /^[a-zA-Z0-9_.:-]{1,256}$/.test(rawRequestId)
+                    && ![options.credential.accessToken, credential.accessToken, accountId].some(secret => rawRequestId.includes(secret))
+                    ? ProviderRequestId(rawRequestId) : undefined;
+                const responseFail = (message, code = 'CODEX_CLOUD_FAILED') => new LlmError(message, code, { status: response.status, ...(requestId === undefined ? {} : { requestId }) });
                 if (!response.ok) {
                     void response.body?.cancel().catch(() => { });
                     if (response.status === 401 && attempt === 0 && this.options.recoverCredential) {
@@ -110,7 +118,7 @@ export class NativeCodexCloudClient {
                             continue;
                         }
                     }
-                    throw fail('Codex cloud service returned HTTP ' + response.status, response.status === 401 ? 'AUTH_REQUIRED' : 'CODEX_CLOUD_HTTP_ERROR');
+                    throw responseFail('Codex cloud service returned HTTP ' + response.status, response.status === 401 ? 'AUTH_REQUIRED' : 'CODEX_CLOUD_HTTP_ERROR');
                 }
                 const reader = response.body?.getReader();
                 if (!reader)
@@ -144,12 +152,9 @@ export class NativeCodexCloudClient {
                     parsed = JSON.parse(new TextDecoder('utf-8', { fatal: true }).decode(Buffer.concat(chunks)));
                 }
                 catch {
-                    throw fail('Codex cloud service returned invalid JSON');
+                    throw responseFail('Codex cloud service returned invalid JSON');
                 }
-                const requestId = path.startsWith('images/')
-                    ? response.headers.get('x-codex-imagegen-request-id') ?? response.headers.get('x-request-id')
-                    : response.headers.get('x-request-id');
-                return { body: parsed, ...(requestId && /^[a-zA-Z0-9_.:-]{1,256}$/.test(requestId) && !requestId.includes(credential.accessToken) ? { requestId } : {}) };
+                return { body: parsed, ...(requestId ? { requestId } : {}) };
             }
             throw fail('Codex cloud authentication failed', 'AUTH_REQUIRED');
         }

package/lib/cloud-media.d.ts

@@ -24,7 +24,7 @@ export interface ResolvedCloudImage {
 export declare function resolveImageSources(value: ImageSelection, deps: MediaSourceDependencies): Promise<ResolvedCloudImage[]>;
 export interface MediaOutputDependencies {
     fs: Pick<FileSystem, 'resolve' | 'processPath' | 'contains'>;
-    shell: Pick<ShellExecutor, 'resolve' | 'run' | 'sandboxMode'>;
+    shell: Pick<ShellExecutor, 'resolve' | 'execute' | 'sandboxMode'>;
     /** Resolve from the calling session with the public sandboxPolicy service. */
     policy: SandboxExecutionPolicy;
     attachments: Pick<AttachmentStore, 'validateImage' | 'saveImage'>;

package/lib/cloud-media.js

@@ -92,11 +92,12 @@ export async function saveImageOutputs(response, deps) {
         if (!deps.fs.contains(workspace, target))
             throw new Error('Original image target escapes caller workspace');
         const path = deps.fs.processPath(target);
-        const outcome = await deps.shell.run(deps.shell.resolve({
+        const execution = await deps.shell.execute(deps.shell.resolve({
             command: 'node -e ' + quote(WRITE_ORIGINAL) + ' -- ' + quote(path),
             stdin: Buffer.from(image.data).toString('base64'), workdir: deps.fs.processPath(workspace),
             sandboxPolicy: deps.policy, signal: deps.signal, timeoutMs: 30_000, stdoutMaxBytes: 1024,
         }));
+        const outcome = await execution.result();
         if (outcome.sandbox?.denied || outcome.sandbox?.runnerFailed || outcome.exitCode !== 0 || outcome.aborted || outcome.timedOut)
             throw new Error('Original image write failed under caller filesystem policy');
         const attachment = await deps.attachments.saveImage({ data: image.data, mediaType: image.mediaType, name });

package/lib/cloud-web-tool.js

@@ -124,12 +124,11 @@ function checkReferences(args, messages, id, recent) {
             for (const block of message.content)
                 if (block.type === 'tool-call')
                     calls.set(block.id, block.name);
-        if (message.source.kind !== 'tool' || !['codex_web', 'run_code'].includes(calls.get(message.source.callId) ?? ''))
+        if (message.role !== 'tool' || message.isError || message.source.kind !== 'tool'
+            || message.toolCallId !== message.source.callId
+            || !['codex_web', 'run_code'].includes(calls.get(message.toolCallId) ?? ''))
             continue;
-        for (const block of message.content)
-            if (block.type === 'tool-result' && block.toolCallId === message.source.callId && !block.isError) {
-                outputs.push(block.content.filter(part => part.type === 'text').map(part => part.text).join('\n'));
-            }
+        outputs.push(message.content.filter(block => block.type === 'text').map(block => block.text).join('\n'));
     }
     outputs.push(...recent);
     for (const ref of refs) {

package/lib/native-adapter.js

@@ -56,8 +56,8 @@ function supportsFast(model) {
 function responsesLiteMode(model, requestedModel = model?.slug) {
     if (model !== undefined && !model.useResponsesLite)
         return {};
-    // Never downgrade the tracked Lite-only Astra slug to Standard Responses during catalog outages.
-    if (model === undefined && requestedModel !== 'gpt-6-astra')
+    // Never downgrade known Lite-only slugs to Standard Responses during catalog outages.
+    if (model === undefined && requestedModel !== 'gpt-6-astra' && requestedModel !== 'gpt-6-sol')
         return {};
     return {
         responsesLite: {

package/lib/native-http.js

@@ -1,11 +1,11 @@
 /** Native ChatGPT Codex HTTP/SSE transport with safe pre-output retries. */
 import { createHash, randomUUID } from 'node:crypto';
 import { networkTelemetry, trackNetwork, observeNetworkOutput } from './network-telemetry.js';
-import { CONTEXT_WINDOW_EXCEEDED_CODE, EMPTY_RESPONSE_CODE, isContextWindowExceededError, isQuotaExceededError, LlmError, ProviderRequestId, QUOTA_EXCEEDED_CODE, attributionHeaders, } from '@deepseek-ai/dsh-llm';
+import { EMPTY_RESPONSE_CODE, LlmError, ProviderRequestId, attributionHeaders, } from '@deepseek-ai/dsh-llm';
 import { nativeCodexAuthorityHash } from './catalog.js';
 import { nativeCodexEndpoint } from './endpoint.js';
 import { NATIVE_CODEX_CONNECTION_FAILED_CODE, NATIVE_CODEX_STREAM_INTERRUPTED_CODE, isNativeCodexConnectionFailure, } from './native-adapter.js';
-import { boundedCodexTurnState, codexRequestBody, codexResponseTurnState, streamResponses, } from './responses.js';
+import { boundedCodexTurnState, codexRequestBody, codexResponseTurnState, streamResponses, responsesFailure, } from './responses.js';
 import { hasNativeCodexReplayKind } from './replay.js';
 import { parseCodexResponseUsageMetadata, publishCodexResponseUsage, } from './response-usage.js';
 import { parseCodexRateLimitEvent, parseCodexRateLimitHeaders, publishCodexRateLimits, } from './rate-limits.js';
@@ -122,28 +122,29 @@ async function resolveImage(block, options, signal) {
         ...stored.detail === undefined ? {} : { detail: stored.detail },
     };
 }
-async function resolveToolResult(block, options, signal) {
-    const content = [];
-    for (const part of block.content) {
-        throwIfAborted(signal);
-        if (part.type === 'text')
-            content.push(part);
-        else if (part.type === 'image')
-            content.push(await resolveImage(part, options, signal));
-        else {
-            throw fixedFailure('native Codex tool output contains an unsupported content block', 'UNSUPPORTED');
-        }
-    }
-    return {
-        type: 'tool-result',
-        toolCallId: block.toolCallId,
-        content,
-        ...block.isError === undefined ? {} : { isError: block.isError },
-    };
-}
 async function resolveMessages(generation, options) {
     const messages = [];
     for (const message of generation.messages) {
+        if (message.role === 'tool') {
+            const content = [];
+            for (const block of message.content) {
+                throwIfAborted(generation.signal);
+                if (block.type === 'text')
+                    content.push(block);
+                else if (block.type === 'image')
+                    content.push(await resolveImage(block, options, generation.signal));
+                else {
+                    throw fixedFailure('native Codex tool output contains an unsupported content block', 'UNSUPPORTED');
+                }
+            }
+            messages.push({
+                role: 'tool',
+                content,
+                toolCallId: message.toolCallId,
+                ...message.isError !== undefined ? { isError: message.isError } : {},
+            });
+            continue;
+        }
         const content = [];
         const source = message.source;
         const sourceKind = source?.kind;
@@ -177,12 +178,6 @@ async function resolveMessages(generation, options) {
                     // A settlement copies the child's assistant output into user-role context.
                     // Its tool calls belong to the child and must not become parent wire calls.
                     break;
-                case 'tool-result':
-                    if (message.role !== 'user') {
-                        throw fixedFailure('native Codex tool results require user messages', 'INVALID_ARGS');
-                    }
-                    content.push(await resolveToolResult(block, options, generation.signal));
-                    break;
                 default:
                     throw fixedFailure('native Codex request contains an unsupported content block', 'UNSUPPORTED');
             }
@@ -191,6 +186,7 @@ async function resolveMessages(generation, options) {
             && source.replayState !== undefined && hasNativeCodexReplayKind(source.replayState)
             ? { provider: source.provider, model: source.model, replayState: source.replayState }
             : undefined;
+        // Developer messages are preserved for the Responses API
         messages.push({
             role: message.role,
             content,
@@ -280,24 +276,20 @@ function errorFacts(response, maxDelayMs) {
 async function httpFailure(response, maxDelayMs, signal, telemetry) {
     const detail = await boundedError(response, signal, telemetry);
     const facts = errorFacts(response, maxDelayMs);
-    const classification = `${detail.code ?? ''} ${detail.message ?? ''}`;
+    const classified = responsesFailure(detail.code, detail.message);
     if (response.status === 401 || response.status === 403) {
         return fixedFailure('native Codex rejected the configured credential', 'AUTH', facts);
     }
+    if (classified.code !== 'SERVER') {
+        const delay = facts?.providerRetryAfterMs ?? classified.failure.providerRetryAfterMs;
+        return fixedFailure(classified.message, classified.code, {
+            ...facts,
+            ...(delay === undefined ? {} : { providerRetryAfterMs: Math.min(delay, maxDelayMs) }),
+        });
+    }
     if (response.status === 429) {
-        if (isQuotaExceededError(classification)) {
-            return fixedFailure('native Codex account quota is exhausted', QUOTA_EXCEEDED_CODE, facts);
-        }
         return fixedFailure('native Codex request was rate limited', 'RATE_LIMIT', facts);
     }
-    if (detail.code === 'context_length_exceeded'
-        || detail.code === 'context_window_exceeded'
-        || isContextWindowExceededError(classification)) {
-        return fixedFailure('native Codex request exceeded the model context window', CONTEXT_WINDOW_EXCEEDED_CODE, facts);
-    }
-    if (isQuotaExceededError(classification)) {
-        return fixedFailure('native Codex account quota is exhausted', QUOTA_EXCEEDED_CODE, facts);
-    }
     if (response.status === 408 || response.status === 504) {
         return fixedFailure('native Codex request timed out', 'TIMEOUT', facts);
     }

package/lib/native-websocket.js

@@ -8,7 +8,7 @@ import { NativeCodexHttpTransport, } from './native-http.js';
 import { replayableItemId } from './replay.js';
 import { parseCodexResponseUsageMetadata, publishCodexResponseUsage, } from './response-usage.js';
 import { parseCodexRateLimitEvent, parseCodexRateLimitHeaders, publishCodexRateLimits, } from './rate-limits.js';
-import { ResponsesStreamTranslator, codexResponseTurnState, } from './responses.js';
+import { ResponsesStreamTranslator, responsesFailure, codexResponseTurnState, } from './responses.js';
 import { NodeNativeCodexWebSocketFactory, } from './native-websocket-socket.js';
 import { NativeCodexWebSocketSessionState } from './native-websocket-session.js';
 const WS_BETA = 'responses_websockets=2026-02-06';
@@ -98,7 +98,8 @@ function sessionKey(generation, routingId) {
 function turnKey(generation) {
     for (let index = generation.messages.length - 1; index >= 0; index--) {
         const message = generation.messages[index];
-        if (message?.source.kind === 'user') {
+        // RequestUserInput doesn't have source or id
+        if (message && 'source' in message && message.source?.kind === 'user' && 'id' in message) {
             return createHash('sha256').update(String(message.id)).digest('base64url');
         }
     }
@@ -193,8 +194,15 @@ function eventFailure(event) {
     }
     if (status === 403)
         return new LlmError('native Codex WebSocket authorization failed', 'AUTH', facts);
-    if (providerCode === 'rate_limit_exceeded' || providerCode === 'usage_limit_reached'
-        || status === 429) {
+    const classified = responsesFailure(providerCode, typeof error.message === 'string' ? error.message : undefined);
+    if (classified.code !== 'SERVER') {
+        const delay = facts?.providerRetryAfterMs ?? classified.failure.providerRetryAfterMs;
+        return new LlmError(classified.message, classified.code === 'RATE_LIMIT' ? 'RATE_LIMITED' : classified.code, {
+            ...facts,
+            ...(delay === undefined ? {} : { providerRetryAfterMs: Math.min(delay, 120_000) }),
+        });
+    }
+    if (providerCode === 'usage_limit_reached' || status === 429) {
         return new LlmError('native Codex WebSocket rate limit was reached', 'RATE_LIMITED', facts);
     }
     if (status !== undefined && status >= 500) {

package/lib/responses.d.ts

@@ -9,18 +9,14 @@ export interface ResolvedImagePart {
     dataBase64: string;
     detail?: CodexImageDetail;
 }
-export interface ResolvedToolResultPart {
-    type: 'tool-result';
-    toolCallId: ToolCallId;
-    content: readonly ResolvedContentPart[];
-    isError?: boolean;
-}
 export type ResolvedContentPart = Exclude<ContentBlock, {
-    type: 'image' | 'tool-result';
-}> | ResolvedImagePart | ResolvedToolResultPart;
+    type: 'image';
+}> | ResolvedImagePart;
 export interface ResolvedMessage {
-    role: 'system' | 'user' | 'assistant';
+    role: 'system' | 'user' | 'assistant' | 'tool' | 'developer';
     content: readonly ResolvedContentPart[];
+    toolCallId?: ToolCallId;
+    isError?: boolean;
     replaySource?: NativeCodexReplaySource;
 }
 export interface ResponsesRequestInput {

package/lib/responses.js

@@ -26,12 +26,12 @@ function imageItem(image) {
     return { type: 'input_image', image_url: `data:${image.mediaType};base64,${image.dataBase64}`,
         ...image.detail === undefined ? {} : { detail: image.detail } };
 }
-function toolOutput(block) {
-    const images = block.content.some(part => part.type === 'image');
+function toolOutput(content) {
+    const images = content.some(part => part.type === 'image');
     if (!images) {
-        return block.content.map(part => part.type === 'text' ? part.text : '').join('');
+        return content.map(part => part.type === 'text' ? part.text : '').join('');
     }
-    return block.content.map((part) => {
+    return content.map((part) => {
         if (part.type === 'text')
             return { type: 'input_text', text: part.text };
         if (part.type === 'image' && 'dataBase64' in part)
@@ -67,6 +67,29 @@ export function toResponsesInput(messages, system) {
             input.push({ type: 'message', role: message.role, content });
             content = [];
         };
+        // Handle tool messages separately - they have no content blocks, just top-level fields
+        if (message.role === 'tool') {
+            if (message.toolCallId === undefined) {
+                throw fixedError('native Codex tool message requires toolCallId', 'INVALID_ARGS');
+            }
+            input.push({
+                type: 'function_call_output',
+                call_id: String(message.toolCallId),
+                output: toolOutput(message.content),
+                ...message.isError === true ? { is_error: true } : {},
+            });
+            continue;
+        }
+        // Handle developer messages - map to developer role in Responses API
+        if (message.role === 'developer') {
+            for (const block of message.content) {
+                if (block.type === 'text') {
+                    content.push({ type: 'input_text', text: block.text });
+                }
+            }
+            flush();
+            continue;
+        }
         for (const block of message.content) {
             if (message.role === 'system') {
                 if (block.type === 'text')
@@ -89,18 +112,15 @@ export function toResponsesInput(messages, system) {
                     content.push(imageItem(block));
                     break;
                 case 'tool-call':
+                    if (message.role !== 'assistant') {
+                        throw fixedError('native Codex tool calls require assistant messages', 'INVALID_ARGS');
+                    }
                     flush();
                     input.push({
                         type: 'function_call', call_id: String(block.id),
                         name: block.name, arguments: block.arguments,
                     });
                     break;
-                case 'tool-result':
-                    flush();
-                    input.push({
-                        type: 'function_call_output', call_id: String(block.toolCallId), output: toolOutput(block),
-                    });
-                    break;
                 case 'reasoning':
                     break; // Visible reasoning is never replayed; M4 may add encrypted items.
                 default:
@@ -282,16 +302,21 @@ export function responsesFailure(code, message) {
         || isContextWindowExceededError(detail)) {
         return fixedError('native Codex request exceeded the model context window', CONTEXT_WINDOW_EXCEEDED_CODE);
     }
-    if (code === 'insufficient_quota' || isQuotaExceededError(detail)) {
+    if (code === 'insufficient_quota' || code === 'credit_balance_exhausted'
+        || code === 'organization_spend_limit_exceeded' || code === 'project_spend_limit_exceeded'
+        || isQuotaExceededError(detail)) {
         return fixedError('native Codex account quota is exhausted', QUOTA_EXCEEDED_CODE);
     }
-    if (code === 'rate_limit_exceeded') {
+    if (code === 'rate_limit_exceeded' || code === 'slow_down') {
         const delay = retryDelay(message);
         return new LlmError('native Codex request was rate limited', 'RATE_LIMIT', {
             ...delay === undefined ? {} : { providerRetryAfterMs: delay },
         });
     }
-    if (code === 'invalid_prompt' || code === 'bio_policy') {
+    if (code === 'invalid_prompt') {
+        return fixedError('native Codex rejected the prompt', 'INVALID_PROMPT');
+    }
+    if (code === 'bio_policy') {
         return fixedError('native Codex rejected the request', 'INVALID_REQUEST');
     }
     return fixedError('native Codex reported a failed response', 'SERVER');

package/lib/upstream.d.ts

@@ -5,4 +5,4 @@ export declare const TRACKED_CODEX_RELEASE = "main@ddf04ad";
 /** Independently audited cloud search/image contracts; not local Codex executor parity. */
 export declare const TRACKED_CODEX_CLOUD_COMMIT = "b348fc26674189f758d5941cdab3f78f258b2aa7";
 /** Whole stable release version sent to the Codex model-catalog endpoint. */
-export declare const CODEX_CLIENT_VERSION = "0.153.4";
+export declare const CODEX_CLIENT_VERSION = "0.155.0";

package/lib/upstream.js

@@ -5,4 +5,4 @@ export const TRACKED_CODEX_RELEASE = 'main@ddf04ad';
 /** Independently audited cloud search/image contracts; not local Codex executor parity. */
 export const TRACKED_CODEX_CLOUD_COMMIT = 'b348fc26674189f758d5941cdab3f78f258b2aa7';
 /** Whole stable release version sent to the Codex model-catalog endpoint. */
-export const CODEX_CLIENT_VERSION = '0.153.4';
+export const CODEX_CLIENT_VERSION = '0.155.0';

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@pure01fx/dsh-openai-codex-auth",
-  "version": "0.11.1",
+  "version": "0.12.0",
   "description": "Native ChatGPT Codex provider, device-code-first login, and same-origin Web integration for DeepSeek Harness",
   "license": "MIT",
   "publishConfig": {
@@ -96,7 +96,7 @@
   },
   "dsh": {
     "engines": {
-      "dsh": "0.1.5-rc.1"
+      "dsh": "0.1.7-rc.2"
     },
     "bundle": {
       "patch": "./cordis.patch.yml"
@@ -112,20 +112,20 @@
     }
   },
   "peerDependencies": {
-    "@deepseek-ai/cordis": "^4.0.1",
-    "@deepseek-ai/dsh-attachment": "0.1.5-rc.1",
-    "@deepseek-ai/dsh-credentials": "0.1.5-rc.1",
-    "@deepseek-ai/dsh-fs": "0.1.5-rc.1",
-    "@deepseek-ai/dsh-host-webserver": "0.1.5-rc.1",
-    "@deepseek-ai/dsh-llm": "0.1.5-rc.1",
-    "@deepseek-ai/dsh-sandbox": "0.1.5-rc.1",
-    "@deepseek-ai/dsh-sandbox-policy": "0.1.5-rc.1",
-    "@deepseek-ai/dsh-shell": "0.1.5-rc.1",
-    "@deepseek-ai/dsh-tools": "0.1.5-rc.1"
+    "@deepseek-ai/cordis": "~4.0.4",
+    "@deepseek-ai/dsh-attachment": "0.1.7-rc.2",
+    "@deepseek-ai/dsh-credentials": "0.1.7-rc.2",
+    "@deepseek-ai/dsh-fs": "0.1.7-rc.2",
+    "@deepseek-ai/dsh-host-webserver": "0.1.7-rc.2",
+    "@deepseek-ai/dsh-llm": "0.1.7-rc.2",
+    "@deepseek-ai/dsh-sandbox": "0.1.7-rc.2",
+    "@deepseek-ai/dsh-sandbox-policy": "0.1.7-rc.2",
+    "@deepseek-ai/dsh-shell": "0.1.7-rc.2",
+    "@deepseek-ai/dsh-tools": "0.1.7-rc.2"
   },
   "dependencies": {
-    "@deepseek-ai/dsh-atomic-write": "0.1.5-rc.1",
-    "@deepseek-ai/dsh-home-paths": "0.1.5-rc.1",
+    "@deepseek-ai/dsh-atomic-write": "0.1.7-rc.2",
+    "@deepseek-ai/dsh-home-paths": "0.1.7-rc.2",
     "@deepseek-ai/schemastery": "^3.18.1",
     "https-proxy-agent": "7.0.6",
     "image-size": "^2.0.2",
@@ -133,29 +133,32 @@
     "ws": "8.21.3"
   },
   "devDependencies": {
-    "@deepseek-ai/dsh-client-ui-model-selection": "0.1.5-rc.1",
-    "@deepseek-ai/dsh-web-frontend": "0.1.5-rc.1",
-    "@deepseek-ai/cordis": "^4.0.1",
-    "@deepseek-ai/dsh-agent": "0.1.5-rc.1",
-    "@deepseek-ai/dsh-attachment": "0.1.5-rc.1",
-    "@deepseek-ai/dsh-attachment-local": "0.1.5-rc.1",
-    "@deepseek-ai/dsh-credentials": "0.1.5-rc.1",
-    "@deepseek-ai/dsh-fs": "0.1.5-rc.1",
-    "@deepseek-ai/dsh-host-webserver": "0.1.5-rc.1",
-    "@deepseek-ai/dsh-llm": "0.1.5-rc.1",
-    "@deepseek-ai/dsh-sandbox": "0.1.5-rc.1",
-    "@deepseek-ai/dsh-sandbox-policy": "0.1.5-rc.1",
-    "@deepseek-ai/dsh-scope": "0.1.5-rc.1",
-    "@deepseek-ai/dsh-session": "0.1.5-rc.1",
-    "@deepseek-ai/dsh-shell": "0.1.5-rc.1",
-    "@deepseek-ai/dsh-system-prompt": "0.1.5-rc.1",
-    "@deepseek-ai/dsh-tools": "0.1.5-rc.1",
+    "@deepseek-ai/cordis": "~4.0.4",
+    "@deepseek-ai/dsh-agent": "0.1.7-rc.2",
+    "@deepseek-ai/dsh-attachment": "0.1.7-rc.2",
+    "@deepseek-ai/dsh-attachment-local": "0.1.7-rc.2",
+    "@deepseek-ai/dsh-client-store": "0.1.7-rc.2",
+    "@deepseek-ai/dsh-client-ui-model-selection": "0.1.7-rc.2",
+    "@deepseek-ai/dsh-credentials": "0.1.7-rc.2",
+    "@deepseek-ai/dsh-fs": "0.1.7-rc.2",
+    "@deepseek-ai/dsh-host-webserver": "0.1.7-rc.2",
+    "@deepseek-ai/dsh-llm": "0.1.7-rc.2",
+    "@deepseek-ai/dsh-sandbox": "0.1.7-rc.2",
+    "@deepseek-ai/dsh-sandbox-policy": "0.1.7-rc.2",
+    "@deepseek-ai/dsh-scope": "0.1.7-rc.2",
+    "@deepseek-ai/dsh-session": "0.1.7-rc.2",
+    "@deepseek-ai/dsh-shell": "0.1.7-rc.2",
+    "@deepseek-ai/dsh-system-prompt": "0.1.7-rc.2",
+    "@deepseek-ai/dsh-tools": "0.1.7-rc.2",
+    "@deepseek-ai/dsh-web-frontend": "0.1.7-rc.2",
     "@types/node": "^22.20.0",
     "@types/proxy-from-env": "1.0.4",
     "@types/ws": "8.18.1",
+    "immer": "10.1.1",
     "sharp": "^0.35.4",
     "typescript": "^6.0.3",
-    "vitest": "^4.1.8"
+    "vitest": "^4.1.8",
+    "zustand": "4.4.7"
   },
   "peerDependenciesMeta": {
     "@deepseek-ai/dsh-tools": {
@@ -177,4 +180,4 @@
       "optional": true
     }
   }
-}
+}