chore(skills): regenerate omni-webhooks after dependency changes (#11095 )

fix(install): make ONNX chain optional for Android/Termux installs (#11095 )
fix(m365): BizChat invocation shape drift + HAR-import UX + Antigravity alias note (#11069 )
2026-08-23 07:32:20 +03:00 · 2026-08-22 14:13:57 -03:00 · 2026-08-22 13:57:59 -03:00 · 2026-08-21 22:32:19 -03:00 · 2026-08-21 22:16:45 -03:00
33 changed files with 1313 additions and 2723 deletions
--- a/changelog.d/fixes/10265-command-code-provider-api.md
+++ b/changelog.d/fixes/10265-command-code-provider-api.md
@@ -0,0 +1 @@
+- fix(command-code): route chat to the documented /provider/v1/chat/completions endpoint instead of the CLI-only /alpha/generate, which Command Code gates/blocks for external callers (#10265)
--- a/changelog.d/fixes/11095-termux-onnx.md
+++ b/changelog.d/fixes/11095-termux-onnx.md
@@ -0,0 +1 @@
+- fix(install): make the ONNX dependency chain optional so Termux/Android installs succeed again (#11095)
--- a/config/quality/file-size-baseline.json
+++ b/config/quality/file-size-baseline.json
@@ -443,7 +443,7 @@
      "src/shared/components/ModelSelectModal.tsx": 1138,
      "src/shared/constants/providers/apikey/gateways.ts": 1250
    },
-    "src/app/(dashboard)/dashboard/providers/[id]/components/modals/AddApiKeyModal.tsx": 1073,
+    "src/app/(dashboard)/dashboard/providers/[id]/components/modals/AddApiKeyModal.tsx": 1080,
    "src/app/(dashboard)/dashboard/providers/[id]/hooks/useProviderConnections.ts": 1051,
    "src/shared/components/ModelSelectModal.tsx": 1138,
    "src/shared/constants/providers/apikey/gateways.ts": 1298,
@@ -459,7 +459,8 @@
    "open-sse/executors/commandCode.ts": 1059,
    "_rebaseline_2026_08_21_10859_vision_bridge_catalog": "#10859 own growth (Vision Bridge fixes #10808/#10809): src/lib/modelCapabilities.ts 1006->1016 (+10, cmd/gpt-5.3-codex* text-only capability resolution) and open-sse/executors/commandCode.ts 988->1023 (+35, Command Code wire-model normalization for bare ids + reasoning field fallback for opencode-routed gateways). Cohesive bug fixes at the existing capability-resolution / executor chokepoints; not extractable mid-fix. Covered by tests/unit/model-capabilities-command-code-codex-textonly-10703.test.ts, tests/unit/command-code-vision.test.ts, tests/unit/opencode-mimo-reasoning-details-nonstream.test.ts. Pushed directly to release (own-session miss: the original rebaseline was made in a throwaway validation worktree and never landed on the PR branch or the release before merge).",
    "_rebaseline_2026_08_21_10907_sticky_pin_clear": "#10907 own growth: open-sse/executors/commandCode.ts 1023->1038 (+15, effort-suffix sanitization threading for the sticky-pin-clear fix). Cohesive change at the existing executor chokepoint. Covered by tests/unit/command-code-executor.test.ts.",
-    "_rebaseline_2026_08_21_10986_reasoning_only_content": "#10986 own growth: open-sse/executors/commandCode.ts 1038->1059 (+21, reasoning-only content fallback — when upstream emits only reasoning-delta events and never a text-delta, surface the reasoning text as message.content in createJsonResponse and emit a synthetic content delta in createStreamResponse). Cohesive bug fix at the existing executor chokepoint (mirrors precedent style of #10907/#10859). Covered by tests/unit/command-code-executor.test.ts (2 new cases: non-stream + streaming)."
+    "_rebaseline_2026_08_21_10986_reasoning_only_content": "#10986 own growth: open-sse/executors/commandCode.ts 1038->1059 (+21, reasoning-only content fallback — when upstream emits only reasoning-delta events and never a text-delta, surface the reasoning text as message.content in createJsonResponse and emit a synthetic content delta in createStreamResponse). Cohesive bug fix at the existing executor chokepoint (mirrors precedent style of #10907/#10859). Covered by tests/unit/command-code-executor.test.ts (2 new cases: non-stream + streaming).",
+    "_rebaseline_2026_08_21_11069_m365_har_import": "#11069 own growth: AddApiKeyModal.tsx 1073->1080 (+7 = Import .har file button for the copilot-m365-web credential modal — M365 is the only provider whose credential (access_token+chathubPath) must be extracted from a DevTools HAR WebSocket URL, added as a new modal affordance). Cohesive UI at the existing modal chokepoint; not extractable. Covered by tests/unit/m365-har-import*.test.ts."
  },
  "_rebaseline_base_2026_08_10_proxyfetch": "Base-red fix (green-prs sweep, issue #9985): open-sse/utils/proxyFetch.ts 1207 > cap 1000 — new proxied-TLS fetch helper introduced by the Fal reference-image work. Owner-authorized quick rebaseline to green; structural slim tracked for v3.9.0.",
  "_rebaseline_2026_07_27_v3849_train2": "Merge-train 2 (7 PRs) — owner-approved 2026-07-27. Single entry: chatCore.ts 4955->5006 (#8595, Responses multi-turn image compaction before the context hard-reject). Genuine irreducible growth at the existing compaction chokepoint in handleChatCore — the PR adds a last-resort retry against the concrete budget plus the estimateFinalInputTokens helper, both wired at the pre-existing call site rather than a new branch. Covered by tests/unit/8560-responses-image-compaction.test.ts (4 tests).",
--- a/open-sse/config/providers/registry/command-code/index.ts
+++ b/open-sse/config/providers/registry/command-code/index.ts
@@ -8,7 +8,11 @@ export const command_codeProvider: RegistryEntry = {
  format: "openai",
  executor: "command-code",
  baseUrl: "https://api.commandcode.ai",
-  chatPath: "/alpha/generate",
+  // Chat uses the documented /provider/v1/chat/completions (OpenAI-format)
+  // endpoint — NOT the CLI-only /alpha/generate endpoint, which Command Code
+  // version-gates and proxy-blocks for external callers (#10265). Discovery
+  // already targets the sibling /provider/v1/models endpoint.
+  chatPath: "/provider/v1/chat/completions",
  modelsUrl: "https://api.commandcode.ai/provider/v1/models",
  // The discovery response is a partial routing catalog; static registry
  // entries omitted from it can still be accepted by the gateway.
--- a/open-sse/executors/commandCode.ts
+++ b/open-sse/executors/commandCode.ts
--- a/open-sse/executors/copilot-m365-frames.ts
+++ b/open-sse/executors/copilot-m365-frames.ts
@@ -30,17 +30,45 @@ export const HANDSHAKE_REQUEST = { protocol: "json", version: 1 } as const;
 export const KEEPALIVE_PING = { type: 6 } as const;

 /**
- * Allowed message types observed in the 2026-08 recapture of the working
- * `m365.cloud.microsoft/chat` client (#10718). The old 11-entry list is no longer
- * seen on the wire — the stale shape gets closed immediately after the type:4.
+ * Allowed message types observed in a 2026-08-21 live capture of a working
+ * `m365.cloud.microsoft/chat` session (issue: "Stream ended before producing a
+ * non-ping SSE event" on every individual/consumer M365 Copilot call). The
+ * #10718 6-entry shape above no longer produces a `type:1 target:"update"`
+ * frame at all — the socket only replies with SignalR keepalive pings and then
+ * closes, which is exactly what surfaces client-side as that generic stream
+ * error. 30 entries, up from 6.
 */
 export const ALLOWED_MESSAGE_TYPES = [
  "Chat",
  "Suggestion",
+  "InternalSearchQuery",
  "Disengaged",
-  "Progress",
-  "EndOfRequest",
  "InternalLoaderMessage",
+  "Progress",
+  "GeneratedCode",
+  "RenderCardRequest",
+  "AdsQuery",
+  "SemanticSerp",
+  "GenerateContentQuery",
+  "GenerateGraphicArt",
+  "SearchQuery",
+  "ConfirmationCard",
+  "AuthError",
+  "DeveloperLogs",
+  "TriggerPlugin",
+  "HintInvocation",
+  "MemoryUpdate",
+  "EndOfRequest",
+  "TriggerConfirmation",
+  "ResumeInvokeAction",
+  "ResumeUserInputRequest",
+  "TriggerUserInputRequest",
+  "EscapeHatch",
+  "TriggerPluginAuth",
+  "ResumePluginAuth",
+  "SideBySide",
+  "ReferencesListComplete",
+  "SwitchRespondingEndpoint",
 ] as const;

 /**
@@ -78,19 +106,26 @@ export const M365_ENTERPRISE_EXTRA_MESSAGE_TYPES = [
 ] as const;

 /**
- * Individual / EDU option sets from the 2026-08 recapture (#10718) — 14 entries.
- * The previous 25-entry consumer/MSA set (enable_msa_user, pdnascan, cwc_code_*,
- * …) is no longer observed on the wire and belongs to the shape the substrate
- * now drops silently.
+ * Individual / EDU option sets from a 2026-08-21 live capture — 34 entries, up
+ * from the #10718 14-entry shape (which itself superseded an earlier 25-entry
+ * shape). Each recapture so far has been additive/reshuffled rather than a
+ * wholesale replacement — treat this as the protocol continuing to drift, not
+ * a one-time fix; a future capture may again need to update this list.
 */
 export const M365_DEFAULT_OPTION_SETS = [
  "search_result_progress_messages_with_search_queries",
  "update_textdoc_response_after_streaming",
  "deepleo_networking_timeout_10minutes_canmore",
  "cwc_flux_image",
+  "cwc_code_interpreter",
+  "cwc_code_interpreter_amsfix",
  "cwcfluxgptv",
  "flux_v3_gptv_enable_upload_multi_image_in_turn_wo_ch",
  "gptvnorm2048",
+  "cwc_code_interpreter_citation_fix",
+  "code_interpreter_interactive_charts",
+  "cwc_code_interpreter_interactive_charts_inline_image",
+  "code_interpreter_matplotlib_patching",
  "cwc_fileupload_odb",
  "update_memory_plugin",
  "add_custom_instructions",
@@ -98,6 +133,20 @@ export const M365_DEFAULT_OPTION_SETS = [
  "flux_v3_progress_messages",
  "enable_batch_token_processing",
  "enable_gg_gpt",
+  "async_client_interaction",
+  "flux_v3_references",
+  "flux_v3_references_entities",
+  "flux_v3_references_ci",
+  "add_filestore_filetype",
+  "cwc_code_interpreter_citation_sourceannotations",
+  "cdxcwc_code_interpreter_hallucinated_url_filter",
+  "flux_v3_image_gen_enable_dimensions",
+  "flux_v3_image_gen_enable_non_watermarked_storage",
+  "flux_v3_image_gen_enable_icon_dimensions",
+  "flux_v3_image_gen_enable_system_text_with_params",
+  "flux_v3_image_gen_enable_designer_dimensions_meta_prompting_in_system_prompts",
+  "flux_v3_image_gen_enable_story",
+  "rich_responses",
 ] as const;

 /** Append the record separator to a JSON-serializable frame. */
@@ -433,12 +482,14 @@ export function resolveChatInvocationOverrides(tier: string | undefined): {
  }
  return {
    optionsSets: [...M365_DEFAULT_OPTION_SETS],
-    // #10718 — the 2026-08 recapture sends tone:"magic" (lowercase) on the
-    // individual/EDU surface; the old "" default is part of the dropped shape.
-    tone: "magic",
+    // 2026-08-21 capture — the individual/consumer surface now sends "Magic"
+    // (capitalized), matching the enterprise tone literal. The #10718
+    // lowercase "magic" is part of the shape that gets silently dropped.
+    tone: "Magic",
    allowedMessageTypes: ALLOWED_MESSAGE_TYPES,
-    // Omitted entirely on the individual/EDU wire (see ChatInvocationOptions).
-    disconnectBehavior: undefined,
+    // 2026-08-21 capture — disconnectBehavior:"continue" is now present on the
+    // individual/consumer wire too, not just enterprise (see ChatInvocationOptions).
+    disconnectBehavior: "continue",
  };
 }

@@ -467,16 +518,33 @@ export function resolveToneForModel(model: string | undefined): string | undefin

 /**
 * Build the `type:4` chat invocation frame body (not yet `\x1e`-terminated).
- * Mirrors the argument shape recaptured from a working `m365.cloud.microsoft/chat`
- * client in 2026-08 (#10718). Notable differences from the pre-#10718 shape: a
- * populated `clientInfo` + `productThreadType:"Office"`, a `conversationId`
- * matching the WS URL query, a rich `message` object, and no
- * `spokenTextMode` / `extraExtensionParameters` / `isSbsSupported` /
- * `renderReferencesBehindEOS` / `disconnectBehavior` — none of those are still
- * observed on the wire, and the stale shape gets closed immediately after the
- * invocation.
+ * Base shape from the #10718 recapture (populated `clientInfo` +
+ * `productThreadType:"Office"`, a `conversationId` matching the WS URL query, a
+ * rich `message` object), extended per a 2026-08-21 live capture that found the
+ * #10718 shape alone no longer produces a `type:1 target:"update"` frame — the
+ * socket only replies with keepalive pings and closes. The additions below
+ * (richer `clientInfo`, non-empty `plugins`, `extraExtensionParameters`,
+ * `isSbsSupported`, `renderReferencesBehindEOS`,
+ * `message.connectedFederatedConnections`, and `disconnectBehavior` on every
+ * tier) are exactly the fields the 2026-08-21 capture had that this shape was
+ * missing; the #10718 fields (`conversationId`, `productThreadType`,
+ * `toolChoice`, `message.attachments`) are kept as-is since removing them was
+ * not verified against a live socket.
 */
 export function buildChatInvocation(opts: ChatInvocationOptions): Record<string, unknown> {
+  const clientInfo = {
+    clientAppName: "Office",
+    clientPlatform: "mcmcopilot-web",
+    clientEntrypoint: "mcmcopilot-officeweb",
+    clientSessionId: opts.sessionId,
+    ProductCategory: "Chat",
+    clientAppType: "Web",
+    productEntryPoint: "ChatPanel",
+    deviceOS: "Windows",
+    deviceType: "Desktop",
+    clientPlatformVersion: "10",
+  };
+
  return {
    type: 4,
    target: "chat",
@@ -487,17 +555,17 @@ export function buildChatInvocation(opts: ChatInvocationOptions): Record<string,
          ? [...opts.allowedMessageTypes]
          : [...ALLOWED_MESSAGE_TYPES],
        clientCorrelationId: opts.clientCorrelationId ?? opts.traceId,
-        clientInfo: {
-          clientAppName: "Office",
-          clientPlatform: "mcmcopilot-web",
-        },
+        clientInfo,
        conversationId: opts.conversationId,
+        extraExtensionParameters: {},
        isStartOfSession: opts.isStartOfSession ?? true,
        message: {
          adaptiveCards: [],
          attachments: null,
          author: "user",
+          clientInfo,
          clientPreferences: {},
+          connectedFederatedConnections: ["dummyId"],
          entityAnnotationTypes: ["People", "File", "Event", "Email", "TeamsMessage"],
          experienceType: "Default",
          inputMethod: "Keyboard",
@@ -510,22 +578,27 @@ export function buildChatInvocation(opts: ChatInvocationOptions): Record<string,
          requestId: opts.requestId,
          text: opts.text,
        },
+        isSbsSupported: true,
        options: {},
        optionsSets: opts.optionsSets ?? [...M365_DEFAULT_OPTION_SETS],
-        plugins: opts.plugins ?? [],
+        // 2026-08-21 capture (#11069): BingWebSearch is now the universal
+        // BuiltIn plugin on individual/consumer tier; keep an opt-out override.
+        plugins: opts.plugins ?? [{ Id: "BingWebSearch", Source: "BuiltIn" }],
        ...(opts.customInstructions ? { customInstructions: opts.customInstructions } : {}),
        productThreadType: "Office",
+        renderReferencesBehindEOS: true,
        sessionId: opts.sessionId,
        sliceIds: [],
        source: "officeweb",
        streamingMode: "ConciseWithPadding",
        threadLevelGptId: {},
-        tone: opts.tone ?? "magic",
+        // 2026-08-21 capture (#11069): tone is now capitalized "Magic" on both tiers.
+        tone: opts.tone ?? "Magic",
        toolChoice: opts.toolChoice ?? null,
        traceId: opts.traceId,
-        // #8971 keeps "continue" for the enterprise tier; the individual/EDU wire
-        // omits the key, so only include it when actually set (#10718).
-        ...(opts.disconnectBehavior ? { disconnectBehavior: opts.disconnectBehavior } : {}),
+        // 2026-08-21 capture — disconnectBehavior:"continue" is sent on every
+        // tier now, not gated to enterprise as the #8971 comment described.
+        disconnectBehavior: opts.disconnectBehavior ?? "continue",
      },
    ],
  };
--- a/package-lock.json
+++ b/package-lock.json
@@ -18,7 +18,6 @@
        "@dnd-kit/core": "^6.3.1",
        "@dnd-kit/sortable": "^10.0.0",
        "@dnd-kit/utilities": "^3.2.2",
-        "@huggingface/transformers": "^4.2.0",
        "@lobehub/icons": "^5.16.0",
        "@modelcontextprotocol/sdk": "^1.29.0",
        "@monaco-editor/react": "^4.7.0",
@@ -61,7 +60,6 @@
        "next-themes": "^0.4.6",
        "node-machine-id": "^1.1.12",
        "omniglyph": "^1.4.0",
-        "onnxruntime-node": "1.24.3",
        "open": "^11.0.1",
        "ora": "^9.4.1",
        "parse5": "^8.0.1",
@@ -156,9 +154,11 @@
      },
      "optionalDependencies": {
        "@atjsh/llmlingua-2": "3.0.0",
+        "@huggingface/transformers": "^4.2.0",
        "better-sqlite3": "^13.0.2",
        "js-tiktoken": "^1.0.20",
        "keytar": "^7.9.0",
+        "onnxruntime-node": "1.24.3",
        "sqlite-vec": "^0.1.9",
        "tls-client-node": "^0.2.0",
        "wreq-js": "^3.0.0"
@@ -4510,6 +4510,7 @@
      "resolved": "https://registry.npmjs.org/@huggingface/jinja/-/jinja-0.5.9.tgz",
      "integrity": "sha512-uWTG+l3VJRsl7EXxYizuL3P+cCPoc3cRqbWWRcQN0FhejRfbdq0RNhCmbY/YDtnTcz9icdLYuLDjsnz4d8JMuw==",
      "license": "MIT",
+      "optional": true,
      "engines": {
        "node": ">=18"
      }
@@ -4518,13 +4519,15 @@
      "version": "0.1.3",
      "resolved": "https://registry.npmjs.org/@huggingface/tokenizers/-/tokenizers-0.1.3.tgz",
      "integrity": "sha512-8rF/RRT10u+kn7YuUbUg0OF30K8rjTc78aHpxT+qJ1uWSqxT1MHi8+9ltwYfkFYJzT/oS+qw3JVfHtNMGAdqyA==",
-      "license": "Apache-2.0"
+      "license": "Apache-2.0",
+      "optional": true
    },
    "node_modules/@huggingface/transformers": {
      "version": "4.2.0",
      "resolved": "https://registry.npmjs.org/@huggingface/transformers/-/transformers-4.2.0.tgz",
      "integrity": "sha512-8BRCoBMH0XsWaEIamuR0LrJGAfftgHAfb2Vrffy0VKlSAE/MnUJ5/h/zTfEP3fDIft+nk7TqB8xXEyABGitBjQ==",
      "license": "Apache-2.0",
+      "optional": true,
      "dependencies": {
        "@huggingface/jinja": "^0.5.6",
        "@huggingface/tokenizers": "^0.1.3",
@@ -9483,30 +9486,35 @@
      "version": "1.1.2",
      "resolved": "https://registry.npmjs.org/@protobufjs/aspromise/-/aspromise-1.1.2.tgz",
      "integrity": "sha512-j+gKExEuLmKwvz3OgROXtrJ2UG2x8Ch2YZUxahh+s1F2HZ+wAceUNLkvy6zKCPVRkU++ZWQrdxsUeQXmcg4uoQ==",
+      "devOptional": true,
      "license": "BSD-3-Clause"
    },
    "node_modules/@protobufjs/base64": {
      "version": "1.1.2",
      "resolved": "https://registry.npmjs.org/@protobufjs/base64/-/base64-1.1.2.tgz",
      "integrity": "sha512-AZkcAA5vnN/v4PDqKyMR5lx7hZttPDgClv83E//FMNhR2TMcLUhfRUBHCmSl0oi9zMgDDqRUJkSxO3wm85+XLg==",
+      "devOptional": true,
      "license": "BSD-3-Clause"
    },
    "node_modules/@protobufjs/codegen": {
      "version": "2.0.5",
      "resolved": "https://registry.npmjs.org/@protobufjs/codegen/-/codegen-2.0.5.tgz",
      "integrity": "sha512-zgXFLzW3Ap33e6d0Wlj4MGIm6Ce8O89n/apUaGNB/jx+hw+ruWEp7EwGUshdLKVRCxZW12fp9r40E1mQrf/34g==",
+      "devOptional": true,
      "license": "BSD-3-Clause"
    },
    "node_modules/@protobufjs/eventemitter": {
      "version": "1.1.1",
      "resolved": "https://registry.npmjs.org/@protobufjs/eventemitter/-/eventemitter-1.1.1.tgz",
      "integrity": "sha512-vW1GmwMZNnL+gMRaovlh9yZX74kc+TTU3FObkkurpMaRtBfLP3ldjS9KQWlwZgraRE0+dheEEoAxdzcJQ8eXZg==",
+      "devOptional": true,
      "license": "BSD-3-Clause"
    },
    "node_modules/@protobufjs/fetch": {
      "version": "1.1.1",
      "resolved": "https://registry.npmjs.org/@protobufjs/fetch/-/fetch-1.1.1.tgz",
      "integrity": "sha512-GpptLrs57adMSuHi3VNj0mAF8dwh36LMaYF6XyJ6JMWlVsc+t42tm1HSEDmOs3A8fC9yyeisgLhsTVQokOZ0zw==",
+      "devOptional": true,
      "license": "BSD-3-Clause",
      "dependencies": {
        "@protobufjs/aspromise": "^1.1.1"
@@ -9516,24 +9524,28 @@
      "version": "1.0.2",
      "resolved": "https://registry.npmjs.org/@protobufjs/float/-/float-1.0.2.tgz",
      "integrity": "sha512-Ddb+kVXlXst9d+R9PfTIxh1EdNkgoRe5tOX6t01f1lYWOvJnSPDBlG241QLzcyPdoNTsblLUdujGSE4RzrTZGQ==",
+      "devOptional": true,
      "license": "BSD-3-Clause"
    },
    "node_modules/@protobufjs/path": {
      "version": "1.1.2",
      "resolved": "https://registry.npmjs.org/@protobufjs/path/-/path-1.1.2.tgz",
      "integrity": "sha512-6JOcJ5Tm08dOHAbdR3GrvP+yUUfkjG5ePsHYczMFLq3ZmMkAD98cDgcT2iA1lJ9NVwFd4tH/iSSoe44YWkltEA==",
+      "devOptional": true,
      "license": "BSD-3-Clause"
    },
    "node_modules/@protobufjs/pool": {
      "version": "1.1.0",
      "resolved": "https://registry.npmjs.org/@protobufjs/pool/-/pool-1.1.0.tgz",
      "integrity": "sha512-0kELaGSIDBKvcgS4zkjz1PeddatrjYcmMWOlAuAPwAeccUrPHdUqo/J6LiymHHEiJT5NrF1UVwxY14f+fy4WQw==",
+      "devOptional": true,
      "license": "BSD-3-Clause"
    },
    "node_modules/@protobufjs/utf8": {
      "version": "1.1.1",
      "resolved": "https://registry.npmjs.org/@protobufjs/utf8/-/utf8-1.1.1.tgz",
      "integrity": "sha512-oOAWABowe8EAbMyWKM0tYDKi8Yaox52D+HWZhAIJqQXbqe0xI/GV7FhLWqlEKreMkfDjshR5FKgi3mnle0h6Eg==",
+      "devOptional": true,
      "license": "BSD-3-Clause"
    },
    "node_modules/@radix-ui/number": {
@@ -12737,6 +12749,7 @@
      "version": "26.2.0",
      "resolved": "https://registry.npmjs.org/@types/node/-/node-26.2.0.tgz",
      "integrity": "sha512-5IviulTZeRNp2vAJ514cc/HUlY5nZ9fCbq9DMyC52BrhFZACo3nI0R7qBxhQmo/d27NFe96ur/b7Wwxklda+kg==",
+      "devOptional": true,
      "license": "MIT",
      "dependencies": {
        "undici-types": "~8.3.0"
@@ -13998,6 +14011,7 @@
      "resolved": "https://registry.npmjs.org/adm-zip/-/adm-zip-0.6.0.tgz",
      "integrity": "sha512-XleryMhbuksdKtofnWZ9Sk+4CUTbms4Mb/EU32SZwToAyZ5RgVos/ki8n+yr0LWHOGKuakbXTuuYNHLQjhddgg==",
      "license": "MIT",
+      "optional": true,
      "engines": {
        "node": ">=14.0"
      }
@@ -14971,7 +14985,8 @@
      "resolved": "https://registry.npmjs.org/boolean/-/boolean-3.2.0.tgz",
      "integrity": "sha512-d0II/GO9uf9lfUHH2BQsjxzRJZBdsjgsBiW4BvhWk/3qoKwQFjIDVN19PfX8F2D/r9PCMTtLWjYVCFrpeYUzsw==",
      "deprecated": "Package no longer supported. Contact Support at https://www.npmjs.com/support for more info.",
-      "license": "MIT"
+      "license": "MIT",
+      "optional": true
    },
    "node_modules/bottleneck": {
      "version": "2.19.5",
@@ -17935,6 +17950,7 @@
      "version": "1.1.4",
      "resolved": "https://registry.npmjs.org/define-data-property/-/define-data-property-1.1.4.tgz",
      "integrity": "sha512-rBMvIzlpA8v6E+SJZoo++HAYqsLrkg7MSfIinMPFhmkorw7X+dOXVJQs+QT69zGkzMyfDnIMN2Wid1+NbL3T+A==",
+      "devOptional": true,
      "license": "MIT",
      "dependencies": {
        "es-define-property": "^1.0.0",
@@ -17964,6 +17980,7 @@
      "version": "1.2.1",
      "resolved": "https://registry.npmjs.org/define-properties/-/define-properties-1.2.1.tgz",
      "integrity": "sha512-8QmQKqEASLd5nx0U1B1okLElbUuuttJ/AnYmRXbbbGDWh6uS208EjD4Xqq/I9wK7u0v6O08XhTWnt5XtEbR6Dg==",
+      "devOptional": true,
      "license": "MIT",
      "dependencies": {
        "define-data-property": "^1.0.1",
@@ -18064,7 +18081,8 @@
      "version": "2.1.0",
      "resolved": "https://registry.npmjs.org/detect-node/-/detect-node-2.1.0.tgz",
      "integrity": "sha512-T0NIuQpnTvFDATNuHN5roPwSBG83rFsuO+MXXH9/3N1eFbn4wcPjttvjMLEPWJ0RGUYgQE7cGgS3tNxbqCGM7g==",
-      "license": "MIT"
+      "license": "MIT",
+      "optional": true
    },
    "node_modules/detect-node-es": {
      "version": "1.1.0",
@@ -18908,7 +18926,8 @@
      "version": "4.1.1",
      "resolved": "https://registry.npmjs.org/es6-error/-/es6-error-4.1.1.tgz",
      "integrity": "sha512-Um/+FxMr9CISWh0bi5Zv0iOD+4cFh5qLeks1qhAopKVAJw3drgKbKySikp7wGhDL0HPeaja0P5ULZrxLkniUVg==",
-      "license": "MIT"
+      "license": "MIT",
+      "optional": true
    },
    "node_modules/es6-promisify": {
      "version": "7.0.0",
@@ -20400,7 +20419,8 @@
      "version": "25.9.23",
      "resolved": "https://registry.npmjs.org/flatbuffers/-/flatbuffers-25.9.23.tgz",
      "integrity": "sha512-MI1qs7Lo4Syw0EOzUl0xjs2lsoeqFku44KpngfIduHBYvzm8h2+7K8YMQh1JtVVVrUvhLpNwqVi4DERegUJhPQ==",
-      "license": "Apache-2.0"
+      "license": "Apache-2.0",
+      "optional": true
    },
    "node_modules/flatted": {
      "version": "3.4.2",
@@ -21226,6 +21246,7 @@
      "resolved": "https://registry.npmjs.org/global-agent/-/global-agent-3.0.0.tgz",
      "integrity": "sha512-PT6XReJ+D07JvGoxQMkT6qji/jVNfX/h364XHZOWeRzy64sSFr+xJ5OX7LI3b4MPQzdL4H8Y8M0xzPpsVMwA8Q==",
      "license": "BSD-3-Clause",
+      "optional": true,
      "dependencies": {
        "boolean": "^3.0.1",
        "es6-error": "^4.1.1",
@@ -21243,6 +21264,7 @@
      "resolved": "https://registry.npmjs.org/semver/-/semver-7.8.5.tgz",
      "integrity": "sha512-Y7/KDsb8LjooZpwaqGyulO6DQlksgCncchHGk+sZIY4SBvUocMBEFH5Ur1fI4dV+Jvl0w6cjvucaIi40puRioA==",
      "license": "ISC",
+      "optional": true,
      "bin": {
        "semver": "bin/semver.js"
      },
@@ -21291,6 +21313,7 @@
      "version": "1.0.4",
      "resolved": "https://registry.npmjs.org/globalthis/-/globalthis-1.0.4.tgz",
      "integrity": "sha512-DpLKbNU4WylpxJykQujfCcwYWiV/Jhm50Goo0wrVILAv5jOr9d+H+UR3PhSCD2rCCEIg0uc+G+muBTwD54JhDQ==",
+      "devOptional": true,
      "license": "MIT",
      "dependencies": {
        "define-properties": "^1.2.1",
@@ -21645,7 +21668,8 @@
      "version": "1.0.9",
      "resolved": "https://registry.npmjs.org/guid-typescript/-/guid-typescript-1.0.9.tgz",
      "integrity": "sha512-Y8T4vYhEfwJOTbouREvG+3XDsjr8E3kIr7uf+JZ0BYloFsttiHU0WfvANVsR7TxNUJa/WpCnw/Ino/p+DeBhBQ==",
-      "license": "ISC"
+      "license": "ISC",
+      "optional": true
    },
    "node_modules/hachure-fill": {
      "version": "0.5.2",
@@ -21679,6 +21703,7 @@
      "version": "1.0.2",
      "resolved": "https://registry.npmjs.org/has-property-descriptors/-/has-property-descriptors-1.0.2.tgz",
      "integrity": "sha512-55JNKuIW+vq4Ke1BjOTjM2YctQIvCT7GFzHwmfZPGo5wnrgkid0YQtnAleFSqumZm4az3n2BS+erby5ipJdgrg==",
+      "devOptional": true,
      "license": "MIT",
      "dependencies": {
        "es-define-property": "^1.0.0"
@@ -24790,7 +24815,8 @@
      "version": "5.0.1",
      "resolved": "https://registry.npmjs.org/json-stringify-safe/-/json-stringify-safe-5.0.1.tgz",
      "integrity": "sha512-ZClg6AaYvamvYEE82d3Iyd3vSSIjQ+odgjaTzRuO3s7toCdFKczob2i0zCh7JE8kWn17yvAWhUVxvqGwUalsRA==",
-      "license": "ISC"
+      "license": "ISC",
+      "optional": true
    },
    "node_modules/json5": {
      "version": "2.2.3",
@@ -26654,6 +26680,7 @@
      "resolved": "https://registry.npmjs.org/matcher/-/matcher-3.0.0.tgz",
      "integrity": "sha512-OkeDaAZ/bQCxeFAozM55PKcKU0yJMPGifLwV4Qgjitu+5MoAfSQN4lsLJeXZ1b8w0x+/Emda6MZgXS1jvsapng==",
      "license": "MIT",
+      "optional": true,
      "dependencies": {
        "escape-string-regexp": "^4.0.0"
      },
@@ -29425,6 +29452,7 @@
      "version": "1.1.1",
      "resolved": "https://registry.npmjs.org/object-keys/-/object-keys-1.1.1.tgz",
      "integrity": "sha512-NuAESUOUMrlIXOfHKzD6bpPu3tYt3xvjNdRIQ+FeT0lNb4K8WR70CaDxhuNguS2XG+GjkyMwOzsN5ZktImfhLA==",
+      "devOptional": true,
      "license": "MIT",
      "engines": {
        "node": ">= 0.4"
@@ -29632,7 +29660,8 @@
      "version": "1.24.3",
      "resolved": "https://registry.npmjs.org/onnxruntime-common/-/onnxruntime-common-1.24.3.tgz",
      "integrity": "sha512-GeuPZO6U/LBJXvwdaqHbuUmoXiEdeCjWi/EG7Y1HNnDwJYuk6WUbNXpF6luSUY8yASul3cmUlLGrCCL1ZgVXqA==",
-      "license": "MIT"
+      "license": "MIT",
+      "optional": true
    },
    "node_modules/onnxruntime-node": {
      "version": "1.24.3",
@@ -29640,6 +29669,7 @@
      "integrity": "sha512-JH7+czbc8ALA819vlTgcV+Q214/+VjGeBHDjX81+ZCD0PCVCIFGFNtT0V4sXG/1JXypKPgScQcB3ij/hk3YnTg==",
      "hasInstallScript": true,
      "license": "MIT",
+      "optional": true,
      "os": [
        "win32",
        "darwin",
@@ -29656,6 +29686,7 @@
      "resolved": "https://registry.npmjs.org/onnxruntime-web/-/onnxruntime-web-1.26.0-dev.20260416-b7804b056c.tgz",
      "integrity": "sha512-MD6Ss4GSpQBo6zqoJzyT9LRbKYs7x/JVN23FT24EcEvlqF4VuzPOeH6X38orZPKHQDbprn7K+SBpu0/mj2CQiw==",
      "license": "MIT",
+      "optional": true,
      "dependencies": {
        "flatbuffers": "^25.1.24",
        "guid-typescript": "^1.0.9",
@@ -29669,13 +29700,15 @@
      "version": "5.3.2",
      "resolved": "https://registry.npmjs.org/long/-/long-5.3.2.tgz",
      "integrity": "sha512-mNAgZ1GmyNhD7AuqnTG3/VQ26o760+ZYBPKjPvugO8+nLbYfX6TVpJPseBvopbdY+qpZ/lKUnmEc1LeZYS3QAA==",
-      "license": "Apache-2.0"
+      "license": "Apache-2.0",
+      "optional": true
    },
    "node_modules/onnxruntime-web/node_modules/onnxruntime-common": {
      "version": "1.24.0-dev.20251116-b39e144322",
      "resolved": "https://registry.npmjs.org/onnxruntime-common/-/onnxruntime-common-1.24.0-dev.20251116-b39e144322.tgz",
      "integrity": "sha512-BOoomdHYmNRL5r4iQ4bMvsl2t0/hzVQ3OM3PHD0gxeXu1PmggqBv3puZicEUVOA3AtHHYmqZtjMj9FOfGrATTw==",
-      "license": "MIT"
+      "license": "MIT",
+      "optional": true
    },
    "node_modules/open": {
      "version": "11.0.1",
@@ -30906,7 +30939,8 @@
      "version": "1.3.6",
      "resolved": "https://registry.npmjs.org/platform/-/platform-1.3.6.tgz",
      "integrity": "sha512-fnWVljUchTro6RiCFvCXBbNhJc2NijN7oIQxbwsyL0buWJPG85v81ehlHI9fXrJsMNgTofEoWIQeClKpgxFLrg==",
-      "license": "MIT"
+      "license": "MIT",
+      "optional": true
    },
    "node_modules/playwright": {
      "version": "1.62.1",
@@ -31861,6 +31895,7 @@
      "version": "7.6.5",
      "resolved": "https://registry.npmjs.org/protobufjs/-/protobufjs-7.6.5.tgz",
      "integrity": "sha512-/FPD0nUc9jH6rfFjji9IBqOz4pcSE3CsT1m7Ep6Mdb0LxSUMj8hgl6GomOvZzpNpAqqGaXA0P3VSrZLFzIhQrw==",
+      "devOptional": true,
      "hasInstallScript": true,
      "license": "BSD-3-Clause",
      "dependencies": {
@@ -31884,6 +31919,7 @@
      "version": "5.3.2",
      "resolved": "https://registry.npmjs.org/long/-/long-5.3.2.tgz",
      "integrity": "sha512-mNAgZ1GmyNhD7AuqnTG3/VQ26o760+ZYBPKjPvugO8+nLbYfX6TVpJPseBvopbdY+qpZ/lKUnmEc1LeZYS3QAA==",
+      "devOptional": true,
      "license": "Apache-2.0"
    },
    "node_modules/proxy-addr": {
@@ -33280,6 +33316,7 @@
      "resolved": "https://registry.npmjs.org/roarr/-/roarr-2.15.4.tgz",
      "integrity": "sha512-CHhPh+UNHD2GTXNYhPWLnU8ONHdI+5DI+4EYIAOaiD63rHeYlZvyh8P+in5999TTSFgUYuKUAjzRI4mdh/p+2A==",
      "license": "BSD-3-Clause",
+      "optional": true,
      "dependencies": {
        "boolean": "^3.0.1",
        "detect-node": "^2.0.4",
@@ -33655,7 +33692,8 @@
      "version": "1.0.0",
      "resolved": "https://registry.npmjs.org/semver-compare/-/semver-compare-1.0.0.tgz",
      "integrity": "sha512-YM3/ITh2MJ5MtzaM429anh+x2jiLVjqILF4m4oyQB18W7Ggea7BfqdH/wGMK7dDiMghv/6WG7znWMwUDzJiXow==",
-      "license": "MIT"
+      "license": "MIT",
+      "optional": true
    },
    "node_modules/send": {
      "version": "1.2.1",
@@ -33688,6 +33726,7 @@
      "resolved": "https://registry.npmjs.org/serialize-error/-/serialize-error-7.0.1.tgz",
      "integrity": "sha512-8I8TjW5KMOKsZQTvoxjuSIa7foAwPWGOts+6o7sgjz41/qMD9VQHEDxi6PBvK2l0MXUmqZyNpUK+T2tQaaElvw==",
      "license": "MIT",
+      "optional": true,
      "dependencies": {
        "type-fest": "^0.13.1"
      },
@@ -33703,6 +33742,7 @@
      "resolved": "https://registry.npmjs.org/type-fest/-/type-fest-0.13.1.tgz",
      "integrity": "sha512-34R7HTnG0XIJcBSn5XhDd7nNFPRcXYRZrBB2O2jdKqYODldSzBAqzsWoZYYvduky73toYS/ESqxPvkDf/F0XMg==",
      "license": "(MIT OR CC0-1.0)",
+      "optional": true,
      "engines": {
        "node": ">=10"
      },
@@ -34474,7 +34514,8 @@
      "version": "1.1.3",
      "resolved": "https://registry.npmjs.org/sprintf-js/-/sprintf-js-1.1.3.tgz",
      "integrity": "sha512-Oo+0REFV59/rz3gfJNKQiBlwfHaSESl1pcGyABQsnnIfWOFt6JNj5gCog2U6MLZ//IGYD+nA8nI+mTShREReaA==",
-      "license": "BSD-3-Clause"
+      "license": "BSD-3-Clause",
+      "optional": true
    },
    "node_modules/sql.js": {
      "version": "1.14.2",
@@ -36187,6 +36228,7 @@
      "version": "8.3.0",
      "resolved": "https://registry.npmjs.org/undici-types/-/undici-types-8.3.0.tgz",
      "integrity": "sha512-j375ScV60dom+YkPFIfTLcOiPxkN/buHz5GobjLhixFuANaNs3C9l4GmrWqejgXWJ7BbJcFYpTEUkS1Ge8bpZQ==",
+      "devOptional": true,
      "license": "MIT"
    },
    "node_modules/unicode-emoji-modifier-base": {
--- a/package.json
+++ b/package.json
@@ -265,7 +265,6 @@
    "@dnd-kit/core": "^6.3.1",
    "@dnd-kit/sortable": "^10.0.0",
    "@dnd-kit/utilities": "^3.2.2",
-    "@huggingface/transformers": "^4.2.0",
    "@lobehub/icons": "^5.16.0",
    "@modelcontextprotocol/sdk": "^1.29.0",
    "@monaco-editor/react": "^4.7.0",
@@ -308,7 +307,6 @@
    "next-themes": "^0.4.6",
    "node-machine-id": "^1.1.12",
    "omniglyph": "^1.4.0",
-    "onnxruntime-node": "1.24.3",
    "open": "^11.0.1",
    "ora": "^9.4.1",
    "parse5": "^8.0.1",
@@ -343,9 +341,11 @@
  },
  "optionalDependencies": {
    "@atjsh/llmlingua-2": "3.0.0",
+    "@huggingface/transformers": "^4.2.0",
    "better-sqlite3": "^13.0.2",
    "js-tiktoken": "^1.0.20",
    "keytar": "^7.9.0",
+    "onnxruntime-node": "1.24.3",
    "sqlite-vec": "^0.1.9",
    "tls-client-node": "^0.2.0",
    "wreq-js": "^3.0.0"
--- a/skills/omni-webhooks/SKILL.md
+++ b/skills/omni-webhooks/SKILL.md
@@ -1,12 +1,12 @@
 ---
 name: omni-webhooks
-description: Register, list, test, and remove webhook endpoints. Configure event subscriptions (request.completed, provider.error, budget.exceeded, etc.) and manage delivery retries.
+description: Register, list, test, and remove webhook endpoints. Configure event subscriptions (request.completed, request.failed, quota.exceeded, etc.) and manage delivery retries.
 ---
 <!-- generated by src/lib/agentSkills/generator.ts; manual edits will be overwritten -->

 ## Overview

-Register, list, test, and remove webhook endpoints. Configure event subscriptions (request.completed, provider.error, budget.exceeded, etc.) and manage delivery retries.
+Register, list, test, and remove webhook endpoints. Configure event subscriptions (request.completed, request.failed, quota.exceeded, etc.) and manage delivery retries.

 ## Authentication

--- a/src/app/(dashboard)/dashboard/combos/IntelligentComboPanel.tsx
+++ b/src/app/(dashboard)/dashboard/combos/IntelligentComboPanel.tsx
@@ -12,8 +12,13 @@ import {
 } from "@/lib/combos/intelligentRouting";
 import { AI_PROVIDERS } from "@/shared/constants/providers";

-function getI18nOrFallback(t: any, key: string, fallback: string) {
-  if (typeof t?.has === "function" && t.has(key)) return t(key);
+function getI18nOrFallback(
+  t: any,
+  key: string,
+  fallback: string,
+  values?: Record<string, unknown>
+) {
+  if (typeof t?.has === "function" && t.has(key)) return t(key, values);
  return fallback;
 }

@@ -94,10 +99,9 @@ export default function IntelligentComboPanel({
      const updatedCombo = await response.json();
      onComboUpdated?.(updatedCombo);
      notify.success(
-        getI18nOrFallback(t, "modePackUpdated", "Mode pack updated to {pack}.").replace(
-          "{pack}",
-          modePackId
-        )
+        getI18nOrFallback(t, "modePackUpdated", "Mode pack updated to {pack}.", {
+          pack: modePackId,
+        }).replace("{pack}", modePackId)
      );
    } catch (error: any) {
      notify.error(error?.message || "Failed to update mode pack.");
@@ -184,10 +188,9 @@ export default function IntelligentComboPanel({
              </div>
              {savingModePack && (
                <span className="text-[11px] text-text-muted">
-                  {getI18nOrFallback(t, "savingModePack", "Saving {pack}…").replace(
-                    "{pack}",
-                    savingModePack
-                  )}
+                  {getI18nOrFallback(t, "savingModePack", "Saving {pack}…", {
+                    pack: savingModePack,
+                  }).replace("{pack}", savingModePack)}
                </span>
              )}
            </div>
--- a/src/app/(dashboard)/dashboard/combos/page.tsx
+++ b/src/app/(dashboard)/dashboard/combos/page.tsx
@@ -533,9 +533,9 @@ function getStrategyBadgeClass(strategy) {
  return "bg-blue-500/15 text-blue-600 dark:text-blue-400";
 }

-function getI18nOrFallback(t, key, fallback) {
+function getI18nOrFallback(t, key, fallback, values) {
  try {
-    if (typeof t.has === "function" && t.has(key)) return t(key);
+    if (typeof t.has === "function" && t.has(key)) return t(key, values);
  } catch {}
  return fallback;
 }
@@ -1565,7 +1565,8 @@ function StrategyRecommendationsPanel({ strategy, onApply, showNudge }) {
          {getI18nOrFallback(
            t,
            "recommendationsUpdated",
-            "Recommendations updated for {strategy}."
+            "Recommendations updated for {strategy}.",
+            { strategy: strategyLabel }
          ).replace("{strategy}", strategyLabel)}
        </div>
      )}
--- a/src/app/(dashboard)/dashboard/providers/[id]/components/HarImportButton.tsx
+++ b/src/app/(dashboard)/dashboard/providers/[id]/components/HarImportButton.tsx
@@ -0,0 +1,165 @@
+"use client";
+
+import { useRef, useState } from "react";
+import { useTranslations } from "next-intl";
+import {
+  extractM365CredentialFromHar,
+  describeHarImportExpiry,
+  type M365HarImportResult,
+} from "@/shared/utils/m365HarImport";
+import { providerText, type ProviderMessageTranslator } from "../providerPageHelpers";
+
+type HarImporter = (text: string) => M365HarImportResult;
+
+// One entry per web-session provider that can offer HAR import. Add a new
+// key here (and its own extractor in src/shared/utils/) to support another
+// provider — the button renders nothing for any provider not listed.
+const HAR_IMPORTERS: Record<string, HarImporter> = {
+  "copilot-m365-web": extractM365CredentialFromHar,
+};
+
+const ERROR_MESSAGE_KEYS: Record<string, [string, string]> = {
+  notJson: ["harImportErrorNotJson", "That file isn't valid JSON — is it really a .har export?"],
+  noEntries: ["harImportErrorNoEntries", "This HAR has no network entries recorded."],
+  noChathubUrl: [
+    "harImportErrorNoChathubUrl",
+    "No Copilot chat connection found in this HAR. Send at least one chat message in m365.cloud.microsoft before exporting.",
+  ],
+  unparsableUrl: [
+    "harImportErrorUnparsableUrl",
+    "Found the chat connection, but couldn't read its URL.",
+  ],
+  missingFields: [
+    "harImportErrorMissingFields",
+    "Found the chat connection, but the token was missing from it.",
+  ],
+};
+
+export interface HarImportButtonProps {
+  provider: string;
+  onImport: (apiKey: string) => void;
+}
+
+export default function HarImportButton({ provider, onImport }: HarImportButtonProps) {
+  const t = useTranslations("providers") as ProviderMessageTranslator;
+  const importer = HAR_IMPORTERS[provider];
+  const fileInputRef = useRef<HTMLInputElement>(null);
+  const [state, setState] = useState<
+    | { phase: "idle" }
+    | { phase: "reading" }
+    | { phase: "error"; message: string }
+    | { phase: "success"; expiresAt: number | null }
+  >({ phase: "idle" });
+
+  if (!importer) return null;
+
+  async function handleFile(file: File | undefined) {
+    if (!file) return;
+    setState({ phase: "reading" });
+    let text: string;
+    try {
+      text = await file.text();
+    } catch {
+      setState({
+        phase: "error",
+        message: providerText(t, "harImportErrorReadFailed", "Couldn't read that file."),
+      });
+      return;
+    }
+
+    const result = importer(text);
+    if (!result.ok) {
+      const [key, fallback] = ERROR_MESSAGE_KEYS[result.error] ?? [
+        "harImportErrorUnknown",
+        "Couldn't extract a credential from that HAR file.",
+      ];
+      setState({ phase: "error", message: providerText(t, key, fallback) });
+      return;
+    }
+
+    onImport(result.apiKey);
+    setState({ phase: "success", expiresAt: result.expiresAt });
+  }
+
+  const expiry = state.phase === "success" ? describeHarImportExpiry(state.expiresAt) : null;
+  const expiryText =
+    expiry?.tone === "unknown"
+      ? providerText(t, "harImportStatusUnknownExpiry", "Imported. Couldn't read its expiry.")
+      : expiry?.tone === "bad"
+        ? providerText(
+            t,
+            "harImportStatusExpired",
+            "Imported, but this token already expired ({minutes}m ago) — export a fresh HAR.",
+            { minutes: Math.abs(expiry.minutesRemaining ?? 0) }
+          )
+        : expiry?.tone === "warn"
+          ? providerText(
+              t,
+              "harImportStatusExpiringSoon",
+              "Imported — valid for only ~{minutes}m more.",
+              { minutes: expiry.minutesRemaining ?? 0 }
+            )
+          : expiry?.tone === "ok"
+            ? providerText(t, "harImportStatusValid", "Imported — valid for ~{minutes}m.", {
+                minutes: expiry.minutesRemaining ?? 0,
+              })
+            : null;
+
+  return (
+    <div className="flex flex-col gap-1.5">
+      <div className="flex items-center gap-2">
+        <button
+          type="button"
+          onClick={() => fileInputRef.current?.click()}
+          disabled={state.phase === "reading"}
+          data-testid="har-import-button"
+          className="inline-flex items-center gap-1.5 rounded border border-border px-2.5 py-1.5 text-xs font-medium text-text-main hover:bg-surface-hover disabled:opacity-50"
+        >
+          <span className="material-symbols-outlined text-[16px]" aria-hidden="true">
+            upload_file
+          </span>
+          {state.phase === "reading"
+            ? providerText(t, "harImportButtonBusy", "Importing…")
+            : providerText(t, "harImportButtonLabel", "Import .har file")}
+        </button>
+        <span className="text-xs text-text-muted">
+          {providerText(
+            t,
+            "harImportButtonHint",
+            "Export from DevTools Network tab after sending at least one chat message."
+          )}
+        </span>
+        <input
+          ref={fileInputRef}
+          type="file"
+          accept=".har,application/json"
+          data-testid="har-import-input"
+          className="hidden"
+          onChange={(event) => {
+            void handleFile(event.target.files?.[0]);
+            event.target.value = "";
+          }}
+        />
+      </div>
+      {state.phase === "error" && (
+        <p className="text-xs text-red-600 dark:text-red-400" data-testid="har-import-error">
+          {state.message}
+        </p>
+      )}
+      {state.phase === "success" && expiryText && (
+        <p
+          className={
+            expiry?.tone === "bad"
+              ? "text-xs text-red-600 dark:text-red-400"
+              : expiry?.tone === "warn"
+                ? "text-xs text-amber-700 dark:text-amber-300"
+                : "text-xs text-emerald-700 dark:text-emerald-300"
+          }
+          data-testid="har-import-status"
+        >
+          {expiryText}
+        </p>
+      )}
+    </div>
+  );
+}
--- a/src/app/(dashboard)/dashboard/providers/[id]/components/modals/AddApiKeyModal.tsx
+++ b/src/app/(dashboard)/dashboard/providers/[id]/components/modals/AddApiKeyModal.tsx
@@ -31,6 +31,7 @@ import {
 import { getWebSessionCredentialRequirement } from "../../webSessionCredentials";
 import { useOpenRouterPresetControl } from "../OpenRouterPresetInput";
 import WebSessionCredentialGuide from "../WebSessionCredentialGuide";
+import HarImportButton from "../HarImportButton";
 import CcCompatibleRequestDefaultsFields from "./CcCompatibleRequestDefaultsFields";
 import { buildAddProviderSpecificData } from "./connectionProviderSpecificData";
 import { getCommandCodeAuthPhaseLabel } from "./commandCodeAuthPhase";
@@ -209,13 +210,13 @@ export default function AddApiKeyModal({
        ? "Freebuff uses an authentic CLI auth token obtained via codebuff CLI login or automated harvester."
        : isWebSessionCredential
          ? getWebSessionCredentialHint(t, webSessionCredential, providerDisplayName, false)
-        : isLocalSelfHostedProvider
-          ? t("localProviderApiKeyOptionalHint", {
-              provider: localProviderMetadata?.name || providerName || provider || "",
-            })
-          : apiKeyOptional
-            ? t("apiKeyOptionalHint")
-            : undefined;
+          : isLocalSelfHostedProvider
+            ? t("localProviderApiKeyOptionalHint", {
+                provider: localProviderMetadata?.name || providerName || provider || "",
+              })
+            : apiKeyOptional
+              ? t("apiKeyOptionalHint")
+              : undefined;
  const credentialValidationFailedMessage = isWebSessionCredential
    ? providerText(
        t,
@@ -750,6 +751,12 @@ export default function AddApiKeyModal({
                t={t}
              />
            )}
+            {provider && (
+              <HarImportButton
+                provider={provider}
+                onImport={(apiKey) => setFormData({ ...formData, apiKey })}
+              />
+            )}
            {!isNoAuthWebSessionCredential && (
              <div className="flex gap-2">
                <Input
--- a/src/app/(dashboard)/dashboard/providers/[id]/components/modals/EditConnectionModal.tsx
+++ b/src/app/(dashboard)/dashboard/providers/[id]/components/modals/EditConnectionModal.tsx
@@ -49,6 +49,7 @@ import {
 import { getWebSessionCredentialRequirement } from "../../webSessionCredentials";
 import { useOpenRouterPresetControl } from "../OpenRouterPresetInput";
 import WebSessionCredentialGuide from "../WebSessionCredentialGuide";
+import HarImportButton from "../HarImportButton";
 import CcCompatibleRequestDefaultsFields from "./CcCompatibleRequestDefaultsFields";
 import { CodexConnectionFields } from "./CodexFingerprintFields";
 import { assignEditApiKeyProviderSpecificData } from "./connectionProviderSpecificData";
@@ -909,6 +910,12 @@ export default function EditConnectionModal({
                t={t}
              />
            )}
+            {provider && (
+              <HarImportButton
+                provider={provider}
+                onImport={(apiKey) => setFormData({ ...formData, apiKey })}
+              />
+            )}
            {!isNoAuthWebSessionCredential && (
              <div className="flex gap-2">
                <Input
--- a/src/i18n/messages/en.json
+++ b/src/i18n/messages/en.json
@@ -6269,6 +6269,20 @@
    "webSessionGuideStep3": "Copy the required credential from the provider's own domain. For cookies, copy only the Cookie header value and omit Cookie:.",
    "webSessionGuideStep3Manual": "Manual path: open the browser developer tools (F12 → Network), refresh the page, open an authenticated request, and copy the Cookie header value from Request Headers — omit the Cookie: prefix.",
    "webSessionGuideStep4": "Paste it here and check the connection. If it stops working, sign in again and replace it with a fresh value.",
+    "harImportButtonLabel": "Import .har file",
+    "harImportButtonBusy": "Importing…",
+    "harImportButtonHint": "Export from DevTools Network tab after sending at least one chat message.",
+    "harImportStatusValid": "Imported — valid for ~{minutes}m.",
+    "harImportStatusExpiringSoon": "Imported — valid for only ~{minutes}m more.",
+    "harImportStatusExpired": "Imported, but this token already expired ({minutes}m ago) — export a fresh HAR.",
+    "harImportStatusUnknownExpiry": "Imported. Couldn't read its expiry.",
+    "harImportErrorNotJson": "That file isn't valid JSON — is it really a .har export?",
+    "harImportErrorNoEntries": "This HAR has no network entries recorded.",
+    "harImportErrorNoChathubUrl": "No Copilot chat connection found in this HAR. Send at least one chat message in m365.cloud.microsoft before exporting.",
+    "harImportErrorUnparsableUrl": "Found the chat connection, but couldn't read its URL.",
+    "harImportErrorMissingFields": "Found the chat connection, but the token was missing from it.",
+    "harImportErrorReadFailed": "Couldn't read that file.",
+    "harImportErrorUnknown": "Couldn't extract a credential from that HAR file.",
    "webSessionSecurityHint": "Treat this like a password: it may access your signed-in web account until it expires or is revoked.",
    "webNoAuthGuideTitle": "No credential required",
    "webNoAuthGuideBody": "{provider} does not need an API key or cookie. Save the connection to use its free web endpoint.",
--- a/src/lib/providers/validation/openaiFormat.ts
+++ b/src/lib/providers/validation/openaiFormat.ts
@@ -1,7 +1,6 @@
 // OpenAI/Gemini-format + Bedrock provider key validators (bedrock, openai-like, command-code, gemini-like, openai-compatible).
 // Extracted from validation.ts (god-file decomposition) — top-level functions; behavior is
 // byte-identical to the original inline defs.
-import { randomUUID } from "node:crypto";
 import { getRegistryEntry } from "@omniroute/open-sse/config/providerRegistry.ts";
 import {
  discoverBedrockNativeModels,
@@ -196,13 +195,12 @@ export async function validateOpenAILikeProvider({
 export async function validateCommandCodeProvider({ apiKey, providerSpecificData = {} }: any) {
  const entry = getRegistryEntry("command-code");
  const baseUrl = normalizeBaseUrl(entry?.baseUrl || "https://api.commandcode.ai");
-  const chatPath = entry?.chatPath || "/alpha/generate";
+  const chatPath = entry?.chatPath || "/provider/v1/chat/completions";
  const url = `${baseUrl}${chatPath.startsWith("/") ? chatPath : `/${chatPath}`}`;
  const validationModelId =
    providerSpecificData?.validationModelId ||
    entry?.models?.find((model) => model.id === "deepseek/deepseek-v4-flash")?.id ||
    "deepseek/deepseek-v4-flash";
-  const { COMMAND_CODE_VERSION } = await import("@omniroute/open-sse/executors/commandCode.ts");

  return validateDirectChatProvider({
    url,
@@ -210,37 +208,13 @@ export async function validateCommandCodeProvider({ apiKey, providerSpecificData
    headers: {
      "Content-Type": "application/json",
      Authorization: `Bearer ${apiKey}`,
-      "x-command-code-version": COMMAND_CODE_VERSION,
-      "x-cli-environment": "external",
-      "x-project-slug": "pi-cc",
-      "x-taste-learning": "false",
-      "x-co-flag": "false",
-      "x-session-id": randomUUID(),
+      Accept: "text/event-stream",
    },
    body: {
-      config: {
-        workingDir: "/workspace",
-        date: new Date().toISOString().slice(0, 10),
-        environment: "external",
-        structure: [],
-        isGitRepo: false,
-        currentBranch: "",
-        mainBranch: "",
-        gitStatus: "",
-        recentCommits: [],
-      },
-      memory: "",
-      taste: "",
-      skills: "",
-      permissionMode: "standard",
-      params: {
-        model: validationModelId,
-        messages: [{ role: "user", content: "test" }],
-        tools: [],
-        system: "",
-        max_tokens: 1,
-        stream: true,
-      },
+      model: validationModelId,
+      messages: [{ role: "user", content: "test" }],
+      stream: true,
+      max_tokens: 1,
    },
  });
 }
--- a/src/shared/constants/providers/apikey/gateways.ts
+++ b/src/shared/constants/providers/apikey/gateways.ts
@@ -83,7 +83,7 @@ export const APIKEY_PROVIDERS_GATEWAYS = {
    textIcon: "CC",
    website: "https://commandcode.ai/",
    authHint:
-      "Use a Command Code API key. Requests are sent to Command Code's /alpha/generate endpoint.",
+      "Use a Command Code API key. Requests are sent to Command Code's /provider/v1/chat/completions endpoint.",
    apiHint: "Create or copy an API key from Command Code, then paste it here as a Bearer token.",
  },
  openrouter: {
--- a/src/shared/utils/m365HarImport.ts
+++ b/src/shared/utils/m365HarImport.ts
@@ -0,0 +1,116 @@
+/**
+ * Pure, transport-free helpers that extract a `copilot-m365-web` credential
+ * (`access_token=...; chathubPath=...`) directly from a DevTools HAR export,
+ * so the "Add connection" / "Edit connection" dialogs can offer a one-click
+ * "Import .har file" button instead of the user hand-extracting the token
+ * from the WebSocket URL (see `HarImportButton.tsx`).
+ *
+ * The token rides in the query string of the BizChat ChatHub WebSocket URL
+ * (`wss://substrate.office.com/m365Copilot/Chathub/<oid>@<tenant>?...
+ * access_token=...`) — the same URL the executor itself connects to (see
+ * `open-sse/executors/copilot-m365-connection.ts`). No network calls here;
+ * everything operates on the HAR text already in the browser.
+ */
+
+/** Prefix identifying the BizChat ChatHub WebSocket request in a HAR entry. */
+export const M365_CHATHUB_WS_PREFIX = "wss://substrate.office.com/m365Copilot/Chathub/";
+
+export type M365HarImportResult =
+  | { ok: true; apiKey: string; chathubPath: string; expiresAt: number | null }
+  | { ok: false; error: string };
+
+interface HarEntryLike {
+  request?: { url?: unknown };
+}
+
+interface HarLike {
+  log?: { entries?: unknown };
+}
+
+/** Decode a JWT payload WITHOUT verification — exp is a display hint only. */
+function decodeJwtExpiry(token: string): number | null {
+  const parts = token.split(".");
+  if (parts.length !== 3) return null;
+  try {
+    const b64 = parts[1].replace(/-/g, "+").replace(/_/g, "/");
+    const padded = b64 + "=".repeat((4 - (b64.length % 4)) % 4);
+    const json = JSON.parse(atob(padded)) as { exp?: unknown };
+    return typeof json.exp === "number" ? json.exp * 1000 : null;
+  } catch {
+    return null;
+  }
+}
+
+/**
+ * Extract the `copilot-m365-web` credential from raw HAR file text.
+ *
+ * Scans every request in `log.entries` for the ChatHub WebSocket URL and
+ * uses the LAST match (a HAR may contain several turns of the same session;
+ * the most recent one carries the freshest token). Returns a structured
+ * error — never throws — so callers can render it directly.
+ */
+export function extractM365CredentialFromHar(text: string): M365HarImportResult {
+  let har: HarLike;
+  try {
+    har = JSON.parse(text) as HarLike;
+  } catch {
+    return { ok: false, error: "notJson" };
+  }
+
+  const entries = har.log?.entries;
+  if (!Array.isArray(entries)) {
+    return { ok: false, error: "noEntries" };
+  }
+
+  let matchedUrl: string | null = null;
+  for (const entry of entries as HarEntryLike[]) {
+    const url = entry?.request?.url;
+    if (typeof url === "string" && url.startsWith(M365_CHATHUB_WS_PREFIX)) {
+      matchedUrl = url;
+    }
+  }
+
+  if (!matchedUrl) {
+    return { ok: false, error: "noChathubUrl" };
+  }
+
+  let parsed: URL;
+  try {
+    parsed = new URL(matchedUrl);
+  } catch {
+    return { ok: false, error: "unparsableUrl" };
+  }
+
+  const token = parsed.searchParams.get("access_token");
+  const chathubPath = decodeURIComponent(parsed.pathname.split("/m365Copilot/Chathub/")[1] || "");
+
+  if (!token || !chathubPath) {
+    return { ok: false, error: "missingFields" };
+  }
+
+  return {
+    ok: true,
+    apiKey: `access_token=${token}; chathubPath=${chathubPath}`,
+    chathubPath,
+    expiresAt: decodeJwtExpiry(token),
+  };
+}
+
+export type HarImportExpiryTone = "ok" | "warn" | "bad" | "unknown";
+
+export interface HarImportExpiryStatus {
+  tone: HarImportExpiryTone;
+  minutesRemaining: number | null;
+}
+
+/** Classify a decoded token expiry for the inline status hint. */
+export function describeHarImportExpiry(
+  expiresAt: number | null,
+  now: number = Date.now()
+): HarImportExpiryStatus {
+  if (expiresAt === null) return { tone: "unknown", minutesRemaining: null };
+  const minutesRemaining = Math.round((expiresAt - now) / 60000);
+  if (minutesRemaining <= 0) return { tone: "bad", minutesRemaining };
+  if (minutesRemaining < 15) return { tone: "warn", minutesRemaining };
+  return { tone: "ok", minutesRemaining };
+}
--- a/tests/unit/build/optional-transformers-dependency.test.ts
+++ b/tests/unit/build/optional-transformers-dependency.test.ts
@@ -9,40 +9,52 @@ function readJson<T = Record<string, unknown>>(relPath: string): T {
  return JSON.parse(readFileSync(join(repoRoot, relPath), "utf8")) as T;
 }

-test("@huggingface/transformers is a regular dependency so npm ci never skips it", () => {
-  // #9962 deliberately moved @huggingface/transformers out of optionalDependencies:
-  // as an optional dep, npm silently skipped the whole subtree on Node 24/26 (old
-  // pin dragged onnxruntime-node@1.21.0 whose NAN build no longer compiles), which
-  // broke `npm ci`/`next build` with "Can't resolve @huggingface/transformers"
-  // (lazy import in src/lib/memory/embedding/transformersLocal.ts). As a regular
-  // dep with onnxruntime-node@~1.24.3 (napi prebuilds, no node-gyp) it stays
-  // installable and the memory embedding path requires() cleanly.
+test("ONNX chain (@huggingface/transformers + onnxruntime-node) stays optional so Termux/Android installs succeed", () => {
+  // #11095: onnxruntime-node declares os ["win32","darwin","linux"], so while
+  // these lived in `dependencies` every npm install on Android/Termux aborted
+  // with a fatal EBADPLATFORM. As optionalDependencies npm skips only the
+  // unsupported-platform subtree (with a warning) and installs normally
+  // everywhere else. This deliberately reverses the MECHANISM of #9962 while
+  // keeping its goal: #9962's skip happened because the old onnxruntime-node@
+  // 1.21.0 pin built from source (NAN) and failed to compile on Node 24/26;
+  // the current 1.24.3 pin ships napi prebuilds, so on supported platforms the
+  // chain always installs and `npm ci`/`next build` keep resolving it. On
+  // platforms where it IS skipped, both consumers degrade gracefully via lazy/
+  // dynamic imports (asserted below).
  const pkg = readJson<{
    dependencies?: Record<string, string>;
    optionalDependencies?: Record<string, string>;
+    overrides?: Record<string, string>;
  }>("package.json");

  assert.equal(
    pkg.dependencies?.["@huggingface/transformers"],
+    undefined,
+    "transformers must NOT be a hard dependency (fatal EBADPLATFORM on Android)"
+  );
+  assert.equal(
+    pkg.optionalDependencies?.["@huggingface/transformers"],
    "^4.2.0",
-    "transformers must be a regular dependency (never optional) so npm ci cannot skip it"
+    "transformers must be an optionalDependency"
  );
-  assert.equal(pkg.optionalDependencies?.["@huggingface/transformers"], undefined);
-});
-
-test("transformers + onnxruntime-node are regular dependencies (not optional)", () => {
-  const pkg = readJson<{
-    dependencies?: Record<string, string>;
-    optionalDependencies?: Record<string, string>;
-  }>("package.json");
-
  assert.equal(
    pkg.dependencies?.["onnxruntime-node"],
-    "1.24.3",
-    "onnxruntime-node is a regular dep (napi prebuilds, installable on Node 24/26)"
+    undefined,
+    "onnxruntime-node must NOT be a hard dependency (fatal EBADPLATFORM on Android)"
  );
-  assert.equal(pkg.optionalDependencies?.["onnxruntime-node"], undefined);
+  assert.equal(
+    pkg.optionalDependencies?.["onnxruntime-node"],
+    "1.24.3",
+    "onnxruntime-node must be an optionalDependency pinned in lockstep with the overrides pin"
+  );
+  assert.equal(
+    pkg.overrides?.["onnxruntime-node"],
+    "1.24.3",
+    "the overrides pin must stay aligned with @huggingface/transformers' own pin (single-copy invariant)"
+  );
+});

+test("lockfile marks the whole ONNX chain optional", () => {
  const lock = readJson<{
    packages: Record<
      string,
@@ -51,26 +63,61 @@ test("transformers + onnxruntime-node are regular dependencies (not optional)",
        dependencies?: Record<string, string>;
        optionalDependencies?: Record<string, string>;
      }
-    >;
+      >;
  }>("package-lock.json");

  assert.equal(
-    lock.packages[""]?.dependencies?.["@huggingface/transformers"],
+    lock.packages[""]?.optionalDependencies?.["@huggingface/transformers"],
    "^4.2.0",
-    "root lock dependencies must hold transformers as a regular (non-optional) dep"
+    "root lock optionalDependencies must hold transformers"
  );
-  // Optional flag is only written `true` for genuinely optional packages;
-  // regular deps leave it absent/null. Assert each is NOT optional.
-  assert.ok(
-    !lock.packages["node_modules/@huggingface/transformers"]?.optional,
-    "transformers must not be marked optional in the lockfile"
+  assert.equal(
+    lock.packages[""]?.optionalDependencies?.["onnxruntime-node"],
+    "1.24.3",
+    "root lock optionalDependencies must hold onnxruntime-node"
  );
  assert.ok(
-    !lock.packages["node_modules/onnxruntime-node"]?.optional,
-    "onnxruntime-node must not be marked optional in the lockfile"
+    lock.packages["node_modules/@huggingface/transformers"]?.optional,
+    "transformers must be marked optional in the lockfile"
  );
  assert.ok(
-    !lock.packages["node_modules/onnxruntime-common"]?.optional,
-    "onnxruntime-common must not be marked optional in the lockfile"
+    lock.packages["node_modules/onnxruntime-node"]?.optional,
+    "onnxruntime-node must be marked optional in the lockfile"
+  );
+  assert.ok(
+    lock.packages["node_modules/onnxruntime-common"]?.optional,
+    "onnxruntime-common must be marked optional in the lockfile"
+  );
+});
+
+test("every @huggingface/transformers consumer loads it lazily so absent installs degrade gracefully", () => {
+  // If any module ever switches to a STATIC import of the optional chain,
+  // startup crashes on platforms where npm skipped it (Android/Termux).
+  // transformersLocal.ts must keep its lazy await import() (D8/D25);
+  // onnxWorker.ts must keep its runtime-variable dynamicImport indirection.
+
+  const embeddingSrc = readFileSync(
+    join(repoRoot, "src/lib/memory/embedding/transformersLocal.ts"),
+    "utf8"
+  );
+  assert.doesNotMatch(
+    embeddingSrc,
+    /^\s*import\s+(?:[^'"]*?\s+from\s+)?["']@huggingface\/transformers["']/m,
+    "transformersLocal.ts must not statically import @huggingface/transformers"
+  );
+  assert.match(
+    embeddingSrc,
+    /await import\(["']@huggingface\/transformers["']\)/,
+    "transformersLocal.ts must load @huggingface/transformers via await import()"
+  );
+
+  const workerSrc = readFileSync(
+    join(repoRoot, "open-sse/services/compression/engines/llmlingua/onnxWorker.ts"),
+    "utf8"
+  );
+  assert.doesNotMatch(
+    workerSrc,
+    /^\s*import\s+(?:[^'"]*?\s+from\s+)?["']@huggingface\/transformers["']/m,
+    "onnxWorker.ts must not statically import @huggingface/transformers"
  );
 });
--- a/tests/unit/command-code-executor.test.ts
+++ b/tests/unit/command-code-executor.test.ts
@@ -8,20 +8,13 @@ const TEST_DATA_DIR = fs.mkdtempSync(path.join(os.tmpdir(), "omniroute-command-c
 process.env.DATA_DIR = TEST_DATA_DIR;

 const { REGISTRY, getRegistryEntry } = await import("../../open-sse/config/providerRegistry.ts");
-const { CommandCodeExecutor, COMMAND_CODE_VERSION } =
-  await import("../../open-sse/executors/commandCode.ts");
+const { CommandCodeExecutor } = await import("../../open-sse/executors/commandCode.ts");
 const { getExecutor, hasSpecializedExecutor } = await import("../../open-sse/executors/index.ts");
-const { createResponsesApiTransformStream } =
-  await import("../../open-sse/transformer/responsesTransformer.ts");
 const core = await import("../../src/lib/db/core.ts");

 const originalFetch = globalThis.fetch;

-type JsonRecord = Record<string, unknown>;
-type ResponsesEvent = {
-  event: string;
-  data: { response: JsonRecord & { usage?: unknown; output?: JsonRecord[] } };
-};
+type FetchCall = { url: string; init: Record<string, unknown>; body?: Record<string, unknown> };

 const PINNED_COMMAND_CODE_MODELS = [
  "claude-opus-4-7",
@@ -44,20 +37,7 @@ const PINNED_COMMAND_CODE_MODELS = [
  "Qwen/Qwen3.6-Plus",
 ];

-function commandCodeStream(lines: unknown[], { sse = false } = {}) {
-  const text = lines
-    .map((line) => {
-      const json = JSON.stringify(line);
-      return sse ? `data: ${json}\n\n` : `${json}\n`;
-    })
-    .join("");
-  return new Response(text, { status: 200, headers: { "Content-Type": "application/x-ndjson" } });
-}
-
-function toPlainHeaders(headers: Headers | Record<string, string>) {
-  if (headers instanceof Headers) return Object.fromEntries(headers.entries());
-  return Object.fromEntries(Object.entries(headers).map(([key, value]) => [key, String(value)]));
-}
+const CHAT_URL = "https://api.commandcode.ai/provider/v1/chat/completions";

 function parseSsePayloads(sse: string) {
  return sse
@@ -68,25 +48,21 @@ function parseSsePayloads(sse: string) {
    .map((line) => JSON.parse(line));
 }

-async function responsesFromChatSse(sse: string): Promise<ResponsesEvent[]> {
-  const input = new ReadableStream<Uint8Array>({
-    start(controller) {
-      controller.enqueue(new TextEncoder().encode(sse));
-      controller.close();
-    },
-  });
-  const transformed = await new Response(
-    input.pipeThrough(createResponsesApiTransformStream(null, 60_000))
-  ).text();
+function openAiSse(obj: unknown): string {
+  return `data: ${JSON.stringify(obj)}\n\n`;
+}

-  return transformed
-    .split("\n\n")
-    .map((part) => {
-      const event = part.match(/^event:\s*(.+)$/m)?.[1];
-      const data = part.match(/^data:\s*(.+)$/m)?.[1];
-      return event && data ? ({ event, data: JSON.parse(data) } as ResponsesEvent) : null;
-    })
-    .filter((entry): entry is ResponsesEvent => entry !== null);
+function captureFetch(body: Record<string, unknown>) {
+  const calls: FetchCall[] = [];
+  globalThis.fetch = async (url, init = {}) => {
+    calls.push({
+      url: String(url),
+      init,
+      body: JSON.parse(String(init.body)),
+    });
+    return new Response(JSON.stringify(body), { status: 200 });
+  };
+  return calls;
 }

 test.afterEach(() => {
@@ -105,7 +81,9 @@ test("Command Code provider catalog has pinned models and alias lookup", () => {
  assert.equal(entry.alias, "cmd");
  assert.equal(entry.executor, "command-code");
  assert.equal(entry.baseUrl, "https://api.commandcode.ai");
-  assert.equal(entry.chatPath, "/alpha/generate");
+  // Chat targets the documented /provider/v1/chat/completions endpoint, NOT the
+  // CLI-only /alpha/generate endpoint (#10265).
+  assert.equal(entry.chatPath, "/provider/v1/chat/completions");
  assert.deepEqual(
    entry.models.map((model) => model.id),
    PINNED_COMMAND_CODE_MODELS
@@ -119,17 +97,10 @@ test("getExecutor returns the specialized Command Code executor", () => {
  assert.ok(getExecutor("cmd") instanceof CommandCodeExecutor);
 });

-type FetchCall = { url: string; init: Record<string, unknown>; body?: unknown };
-
-test("Command Code executor posts wrapped body and required headers to /alpha/generate", async () => {
-  const calls: FetchCall[] = [];
-  globalThis.fetch = async (url, init = {}) => {
-    calls.push({ url: String(url), init });
-    return commandCodeStream([{ type: "text-delta", text: "hello" }, { type: "finish" }]);
-  };
-
+test("Command Code executor posts a flat OpenAI body + standard headers to /provider/v1/chat/completions (#10265)", async () => {
+  const calls = captureFetch({});
  const executor = getExecutor("command-code");
-  const { response, url, headers, transformedBody } = await executor.execute({
+  const { response, url, headers } = await executor.execute({
    model: "gpt-5.4-mini",
    stream: false,
    credentials: { apiKey: "cc_test_key" },
@@ -144,41 +115,34 @@ test("Command Code executor posts wrapped body and required headers to /alpha/ge
    },
  });

-  assert.equal(url, "https://api.commandcode.ai/alpha/generate");
+  assert.equal(url, CHAT_URL);
  assert.equal(calls.length, 1);
-  assert.equal(calls[0].url, "https://api.commandcode.ai/alpha/generate");
+  assert.equal(calls[0].url, CHAT_URL);
  assert.equal(calls[0].init.method, "POST");
  assert.equal(headers.Authorization, "Bearer cc_test_key");
-  assert.equal(headers["x-command-code-version"], COMMAND_CODE_VERSION);
-  assert.equal(headers["x-cli-environment"], "external");
-  assert.equal(headers["x-project-slug"], "pi-cc");
-  assert.equal(headers["x-taste-learning"], "false");
-  assert.equal(headers["x-co-flag"], "false");
-  assert.equal(typeof headers["x-session-id"], "string");
+  // No CLI-impersonation headers.
+  assert.equal(headers["x-command-code-version"], undefined);
+  assert.equal(headers["x-cli-environment"], undefined);
+  assert.equal(headers["x-project-slug"], undefined);

-  const posted = JSON.parse(String(calls[0].init.body));
-  assert.deepEqual(posted, transformedBody);
-  for (const key of ["config", "memory", "taste", "skills", "permissionMode", "params"]) {
-    assert.ok(key in posted, `missing ${key}`);
-  }
-  assert.equal(posted.skills, "");
-  assert.equal(posted.params.model, "gpt-5.4-mini");
-  assert.equal(posted.params.stream, true);
-  assert.equal(posted.params.system, "You are concise.");
-  assert.equal(posted.params.messages[0].role, "user");
-  assert.equal(posted.params.tools[0].name, "lookup");
+  const posted = calls[0].body as Record<string, unknown>;
+  // No CLI envelope.
+  assert.equal(posted.config, undefined, "CLI envelope config must not be sent");
+  assert.equal(posted.params, undefined, "CLI envelope params wrapper must not be sent");
+  assert.equal(posted.model, "gpt-5.4-mini");
+  assert.equal(posted.stream, false);
+  assert.equal((posted.messages as Array<{ role: string }>)[0].role, "system");
+  const tool = (posted.tools as Array<{ function: { name: string } }>)[0];
+  assert.equal(tool.function.name, "lookup", "tools in OpenAI shape (function.name)");
+  assert.equal(posted.max_tokens, 42);

+  // The upstream OpenAI JSON passes through untouched.
  const json = await response.json();
-  assert.equal(json.choices[0].message.content, "hello");
+  assert.deepEqual(json, {});
 });

-test("Command Code executor passes reasoning/thinking fields through to params (#2986 follow-up)", async () => {
-  const calls: FetchCall[] = [];
-  globalThis.fetch = async (url, init = {}) => {
-    calls.push({ url: String(url), init });
-    return commandCodeStream([{ type: "text-delta", text: "ok" }, { type: "finish" }]);
-  };
-
+test("Command Code executor passes reasoning/thinking fields through at the top level of the OpenAI body", async () => {
+  const calls = captureFetch({});
  await getExecutor("command-code").execute({
    model: "deepseek/deepseek-v4-pro",
    stream: false,
@@ -189,30 +153,19 @@ test("Command Code executor passes reasoning/thinking fields through to params (
      reasoning_effort: "high",
      thinking: { type: "enabled" },
      effort: "high",
-      output_config: { effort: "high" },
      extra_body: { enable_thinking: true },
    },
  });

-  const posted = JSON.parse(String(calls[0].init.body));
-  assert.equal(posted.params.reasoning_effort, "high");
-  assert.deepEqual(posted.params.thinking, { type: "enabled" });
-  assert.equal(posted.params.effort, "high");
-  assert.deepEqual(posted.params.output_config, { effort: "high" });
-  assert.deepEqual(posted.params.extra_body, { enable_thinking: true });
+  const posted = calls[0].body as Record<string, unknown>;
+  assert.equal(posted.reasoning_effort, "high");
+  assert.deepEqual(posted.thinking, { type: "enabled" });
+  assert.equal(posted.effort, "high");
+  assert.deepEqual(posted.extra_body, { enable_thinking: true });
 });

 test("Command Code executor honors body.model rewrite from payload rules", async () => {
-  const calls: FetchCall[] = [];
-  globalThis.fetch = async (url, init = {}) => {
-    calls.push({ url: String(url), init });
-    return commandCodeStream([{ type: "text-delta", text: "ok" }, { type: "finish" }]);
-  };
-
-  // Simulate a payload-rule rewrite: combo resolves to "deepseek-v4-pro-max"
-  // (passed as the execute() model arg), but the payload rule overwrites
-  // body.model to "deepseek/deepseek-v4-pro" (the vendor-prefixed form
-  // Command Code's API expects).
+  const calls = captureFetch({});
  await getExecutor("command-code").execute({
    model: "deepseek-v4-pro-max",
    stream: false,
@@ -225,20 +178,13 @@ test("Command Code executor honors body.model rewrite from payload rules", async
    },
  });

-  const posted = JSON.parse(String(calls[0].init.body));
-  assert.equal(posted.params.model, "deepseek/deepseek-v4-pro");
-  assert.equal(posted.params.reasoning_effort, "max");
+  const posted = calls[0].body as Record<string, unknown>;
+  assert.equal(posted.model, "deepseek/deepseek-v4-pro");
+  assert.equal(posted.reasoning_effort, "max");
 });

 test("Command Code executor maps unsupported minimal reasoning_effort to low (upstream 400 regression)", async () => {
-  const calls: FetchCall[] = [];
-  globalThis.fetch = async (url, init = {}) => {
-    calls.push({ url: String(url), init });
-    return commandCodeStream([{ type: "text-delta", text: "ok" }, { type: "finish" }]);
-  };
-
-  // Live upstream rejection: "Validation error: Invalid option: expected one of
-  // \"low\"|\"medium\"|\"high\"|\"xhigh\"|\"max\" at \"params.reasoning_effort\"" —
+  const calls = captureFetch({});
  // `minimal` (a Muse Spark catalog tier) must be downgraded to `low` before
  // the wire body is built, on BOTH the combo and single-model paths.
  await getExecutor("command-code").execute({
@@ -252,20 +198,38 @@ test("Command Code executor maps unsupported minimal reasoning_effort to low (up
    },
  });

-  const posted = JSON.parse(String(calls[0].init.body));
-  assert.equal(posted.params.reasoning_effort, "low", "minimal must map to low");
+  const posted = calls[0].body as Record<string, unknown>;
+  assert.equal(posted.reasoning_effort, "low", "minimal must map to low");
 });

-test("Command Code raw NDJSON stream becomes OpenAI chat SSE chunks", async () => {
-  const calls: FetchCall[] = [];
+test("Command Code executor passes the upstream OpenAI SSE stream through untouched", async () => {
+  const sse =
+    openAiSse({
+      id: "c1",
+      object: "chat.completion.chunk",
+      model: "gpt-5.4",
+      choices: [{ index: 0, delta: { role: "assistant" } }],
+    }) +
+    openAiSse({
+      id: "c1",
+      object: "chat.completion.chunk",
+      model: "gpt-5.4",
+      choices: [{ index: 0, delta: { content: "Hello" } }],
+    }) +
+    openAiSse({
+      id: "c1",
+      object: "chat.completion.chunk",
+      model: "gpt-5.4",
+      choices: [{ index: 0, delta: {}, finish_reason: "stop" }],
+    }) +
+    "data: [DONE]\n\n";
+  let capturedStreamFlag: unknown = null;
  globalThis.fetch = async (url, init = {}) => {
-    calls.push({ url: String(url), init, body: JSON.parse(String(init.body)) });
-    return commandCodeStream([
-      { type: "text-delta", text: "Hello" },
-      { type: "reasoning-delta", text: "thinking" },
-      { type: "tool-call", toolCallId: "call_1", toolName: "search", input: { q: "docs" } },
-      { type: "finish", finishReason: "tool-calls" },
-    ]);
+    capturedStreamFlag = JSON.parse(String(init.body)).stream;
+    return new Response(sse, {
+      status: 200,
+      headers: { "Content-Type": "text/event-stream" },
+    });
  };

  const { response } = await getExecutor("command-code").execute({
@@ -275,39 +239,32 @@ test("Command Code raw NDJSON stream becomes OpenAI chat SSE chunks", async () =
    body: { messages: [{ role: "user", content: "Hi" }] },
  });

-  assert.equal(calls[0].body.params.stream, true);
-  assert.equal(response.headers.get("Content-Type"), "text/event-stream; charset=utf-8");
-  const sse = await response.text();
-  assert.match(sse, /data: \[DONE\]/);
-  const chunks = parseSsePayloads(sse);
-  assert.equal(chunks[0].object, "chat.completion.chunk");
-  assert.deepEqual(chunks[0].choices[0].delta, { role: "assistant" });
+  assert.equal(capturedStreamFlag, true, "stream flag forwarded to upstream");
+  const text = await response.text();
+  assert.equal(text, sse, "OpenAI SSE stream passed through byte-for-byte");
+  assert.ok(text.includes("data: [DONE]"));
+  const chunks = parseSsePayloads(text);
+  assert.equal(chunks[0].choices[0].delta.role, "assistant");
  assert.equal(chunks[1].choices[0].delta.content, "Hello");
-  assert.equal(chunks[2].choices[0].delta.reasoning_content, "thinking");
-  assert.equal(chunks[3].choices[0].delta.tool_calls[0].function.name, "search");
-  assert.equal(chunks.at(-1).choices[0].finish_reason, "tool_calls");
+  assert.equal(chunks[2].choices[0].finish_reason, "stop");
 });

-test("Command Code data: SSE lines aggregate into non-stream ChatCompletion JSON", async () => {
-  globalThis.fetch = async () =>
-    commandCodeStream(
-      [
-        { type: "text-delta", text: "Hel" },
-        { type: "text-delta", text: "lo" },
-        { type: "reasoning-delta", text: "because" },
-        { type: "tool-call", id: "call_2", name: "lookup", arguments: { id: 7 } },
-        {
-          type: "finish",
-          finishReason: "max_tokens",
-          totalUsage: {
-            inputTokens: 3,
-            inputTokenDetails: { cacheReadTokens: 2 },
-            outputTokens: 5,
-          },
-        },
-      ],
-      { sse: true }
-    );
+test("Command Code executor passes the upstream OpenAI JSON through untouched (non-stream)", async () => {
+  const upstreamJson = {
+    id: "chatcmpl-1",
+    object: "chat.completion",
+    model: "gpt-5.4-mini",
+    choices: [{ index: 0, message: { role: "assistant", content: "Hello" }, finish_reason: "stop" }],
+    usage: { prompt_tokens: 3, completion_tokens: 2, total_tokens: 5 },
+  };
+  let capturedStreamFlag: unknown = null;
+  globalThis.fetch = async (url, init = {}) => {
+    capturedStreamFlag = JSON.parse(String(init.body)).stream;
+    return new Response(JSON.stringify(upstreamJson), {
+      status: 200,
+      headers: { "Content-Type": "application/json" },
+    });
+  };

  const { response } = await getExecutor("command-code").execute({
    model: "gpt-5.4-mini",
@@ -316,88 +273,12 @@ test("Command Code data: SSE lines aggregate into non-stream ChatCompletion JSON
    body: { messages: [{ role: "user", content: "Hi" }] },
  });

-  assert.equal(response.headers.get("Content-Type"), "application/json");
-  const json = await response.json();
-  assert.equal(json.object, "chat.completion");
-  assert.equal(json.choices[0].message.content, "Hello");
-  assert.equal(json.choices[0].message.reasoning_content, "because");
-  assert.equal(json.choices[0].message.tool_calls[0].function.arguments, JSON.stringify({ id: 7 }));
-  assert.equal(json.choices[0].finish_reason, "length");
-  assert.deepEqual(json.usage, {
-    prompt_tokens: 3,
-    prompt_tokens_details: { cached_tokens: 2 },
-    completion_tokens: 5,
-    completion_tokens_details: { reasoning_tokens: 0 },
-    total_tokens: 8,
-    cache_read_input_tokens: 2,
-  });
+  assert.equal(capturedStreamFlag, false, "stream flag forwarded as false for non-stream");
+  assert.deepEqual(await response.json(), upstreamJson);
 });

-test("Command Code reasoning-only output falls back to reasoning as content (non-stream)", async () => {
-  globalThis.fetch = async () =>
-    commandCodeStream(
-      [
-        { type: "reasoning-delta", text: "The user wants 79874+93658. " },
-        { type: "reasoning-delta", text: "That equals 173532." },
-        {
-          type: "finish",
-          finishReason: "stop",
-          totalUsage: { inputTokens: 20, outputTokens: 64, outputTokenDetails: { reasoningTokens: 61 } },
-        },
-      ],
-      { sse: true }
-    );
-
-  const { response } = await getExecutor("command-code").execute({
-    model: "meta/muse-spark-1.2-contributor",
-    stream: false,
-    credentials: { apiKey: "cc_test_key" },
-    body: { messages: [{ role: "user", content: "Calculate 79874+93658, and reply with the result only." }] },
-  });
-
-  const json = await response.json();
-  const message = json.choices[0].message;
-  // Regression #10986: when the model emits only reasoning-delta events (never a
-  // text-delta), content must fall back to the reasoning text instead of "" (which
-  // OpenAI-compatible clients treat as null/no answer).
-  assert.equal(message.content, "The user wants 79874+93658. That equals 173532.");
-  // reasoning_content must STAY populated for reasoning-aware clients.
-  assert.equal(message.reasoning_content, "The user wants 79874+93658. That equals 173532.");
-});
-
-test("Command Code reasoning-only output emits a content delta chunk when streaming", async () => {
-  globalThis.fetch = async () =>
-    commandCodeStream(
-      [
-        { type: "reasoning-delta", text: "The result is 173532." },
-        { type: "finish", finishReason: "stop" },
-      ],
-      { sse: true }
-    );
-
-  const { response } = await getExecutor("command-code").execute({
-    model: "meta/muse-spark-1.2-contributor",
-    stream: true,
-    credentials: { apiKey: "cc_test_key" },
-    body: { messages: [{ role: "user", content: "Calcular 79874+93658" }] },
-  });
-
-  const sse = await response.text();
-  assert.match(sse, /data: \[DONE\]/);
-  const chunks = parseSsePayloads(sse);
-  assert.equal(chunks[0].choices[0].delta.role, "assistant");
-  // Regression #10986: the reasoning-only stream must emit a content delta when it
-  // otherwise ends with no content. reasoning_content stays present too.
-  const contentDelta = chunks.find((c) => c.choices[0].delta.content !== undefined);
-  assert.equal(contentDelta.choices[0].delta.content, "The result is 173532.");
-  const reasoningDelta = chunks.find((c) => c.choices[0].delta.reasoning_content !== undefined);
-  assert.equal(reasoningDelta.choices[0].delta.reasoning_content, "The result is 173532.");
-  assert.equal(chunks.at(-1).choices[0].finish_reason, "stop");
-});
-
-test("Command Code executor surfaces upstream and streamed errors", async () => {
-  globalThis.fetch = async () =>
-    new Response("bad key", { status: 401, statusText: "Unauthorized" });
+test("Command Code executor surfaces upstream errors", async () => {
+  globalThis.fetch = async () => new Response("bad key", { status: 401, statusText: "Unauthorized" });
  const upstreamFailure = await getExecutor("command-code").execute({
    model: "gpt-5.4-mini",
    stream: false,
@@ -406,124 +287,68 @@ test("Command Code executor surfaces upstream and streamed errors", async () =>
  });
  assert.equal(upstreamFailure.response.status, 401);
  assert.equal(await upstreamFailure.response.text(), "bad key");
-
-  globalThis.fetch = async () => commandCodeStream([{ type: "error", error: { message: "boom" } }]);
-  await assert.rejects(async () => {
-    await getExecutor("command-code").execute({
-      model: "gpt-5.4-mini",
-      stream: false,
-      credentials: { apiKey: "cc_test_key" },
-      body: { messages: [{ role: "user", content: "Hi" }] },
-    });
-  }, /boom/);
 });

-test("Command Code executor omits max_tokens when the client does not supply one (GLM-5.x)", async () => {
-  const calls: FetchCall[] = [];
-  globalThis.fetch = async (url, init = {}) => {
-    calls.push({ url: String(url), init, body: JSON.parse(String(init.body)) });
-    return commandCodeStream([{ type: "text-delta", text: "ok" }, { type: "finish" }]);
-  };
-
-  // No client max_tokens: we must NOT fabricate one. Omitting the field lets
-  // Command Code's upstream apply the model's own native default.
+test("Command Code executor omits max_tokens when the client does not supply one", async () => {
+  const calls = captureFetch({});
  await getExecutor("command-code").execute({
    model: "zai-org/GLM-5.1",
    stream: false,
    credentials: { apiKey: "cc_test_key" },
    body: { messages: [{ role: "user", content: "Hi" }] },
  });
-  assert.ok(!("max_tokens" in calls[0].body.params));
-});
-
-test("Command Code executor omits max_tokens for DeepSeek v4 when the client does not supply one", async () => {
-  const calls: FetchCall[] = [];
-  globalThis.fetch = async (url, init = {}) => {
-    calls.push({ url: String(url), init, body: JSON.parse(String(init.body)) });
-    return commandCodeStream([{ type: "text-delta", text: "ok" }, { type: "finish" }]);
-  };
-
-  // Regression: previously the executor invented max_tokens from the registry
-  // (384000), which /alpha/generate rejects with a 400
-  // "Too big: expected number to be <=200000". With no client value we now omit
-  // the field entirely, so the request succeeds and upstream picks the default.
-  await getExecutor("command-code").execute({
-    model: "deepseek/deepseek-v4-pro",
-    stream: false,
-    credentials: { apiKey: "cc_test_key" },
-    body: { messages: [{ role: "user", content: "Hi" }] },
-  });
-  assert.ok(!("max_tokens" in calls[0].body.params));
+  const posted = calls[0].body as Record<string, unknown>;
+  assert.ok(!("max_tokens" in posted), "must not fabricate max_tokens");
+  assert.ok(!("max_completion_tokens" in posted), "must not fabricate max_completion_tokens");
 });

 test("Command Code executor clamps an oversized client-supplied max_tokens to the endpoint ceiling", async () => {
-  const calls: FetchCall[] = [];
-  globalThis.fetch = async (url, init = {}) => {
-    calls.push({ url: String(url), init, body: JSON.parse(String(init.body)) });
-    return commandCodeStream([{ type: "text-delta", text: "ok" }, { type: "finish" }]);
-  };
-
-  // A client asking for more than the 200000 endpoint ceiling is clamped down
-  // (not 400'd), mirroring the provider-driven clamp in antigravity.ts.
+  const calls = captureFetch({});
+  // A client asking for more than the 200000 endpoint ceiling is clamped down.
  await getExecutor("command-code").execute({
    model: "deepseek/deepseek-v4-pro",
    stream: false,
    credentials: { apiKey: "cc_test_key" },
    body: { messages: [{ role: "user", content: "Hi" }], max_tokens: 500000 },
  });
-  assert.equal(calls[0].body.params.max_tokens, 200000);
+  assert.equal((calls[0].body as Record<string, unknown>).max_tokens, 200000);
 });

-test("Command Code executor honors a smaller client-provided max_tokens under the per-model cap", async () => {
-  const calls: FetchCall[] = [];
-  globalThis.fetch = async (url, init = {}) => {
-    calls.push({ url: String(url), init, body: JSON.parse(String(init.body)) });
-    return commandCodeStream([{ type: "text-delta", text: "ok" }, { type: "finish" }]);
-  };
-
+test("Command Code executor honors a smaller client-provided max_tokens", async () => {
+  const calls = captureFetch({});
  await getExecutor("command-code").execute({
    model: "zai-org/GLM-5.1",
    stream: false,
    credentials: { apiKey: "cc_test_key" },
    body: { messages: [{ role: "user", content: "Hi" }], max_tokens: 2048 },
  });
-  assert.equal(calls[0].body.params.max_tokens, 2048);
+  assert.equal((calls[0].body as Record<string, unknown>).max_tokens, 2048);
 });

-test("Command Code non-stream aggregation throws when the final error event lacks a trailing newline", async () => {
-  globalThis.fetch = async () =>
-    new Response(
-      `${JSON.stringify({ type: "text-delta", text: "Hello" })}\n${JSON.stringify({
-        type: "error",
-        error: { message: "boom" },
-      })}`,
-      { status: 200, headers: { "Content-Type": "application/x-ndjson" } }
-    );
-
-  await assert.rejects(async () => {
-    await getExecutor("command-code").execute({
+test("Command Code stream preserves the upstream OpenAI usage chunk (passthrough)", async () => {
+  const sse =
+    openAiSse({
+      id: "c1",
+      object: "chat.completion.chunk",
      model: "gpt-5.4-mini",
-      stream: false,
-      credentials: { apiKey: "cc_test_key" },
-      body: { messages: [{ role: "user", content: "Hi" }] },
-    });
-  }, /boom/);
-});
-
-test("Command Code usage chunk surfaces cache_read and no_cache for the stream pipeline", async () => {
-  globalThis.fetch = async () =>
-    commandCodeStream([
-      { type: "text-delta", text: "Hi" },
-      {
-        type: "finish",
-        finishReason: "stop",
-        totalUsage: {
-          inputTokens: 10,
-          inputTokenDetails: { noCacheTokens: 6, cacheReadTokens: 4 },
-          outputTokens: 6,
-        },
+      choices: [{ index: 0, delta: { content: "Hi" } }],
+    }) +
+    openAiSse({
+      id: "c1",
+      object: "chat.completion.chunk",
+      model: "gpt-5.4-mini",
+      choices: [],
+      usage: {
+        prompt_tokens: 10,
+        prompt_tokens_details: { cached_tokens: 4 },
+        completion_tokens: 6,
+        completion_tokens_details: { reasoning_tokens: 1 },
+        total_tokens: 16,
      },
-    ]);
+    }) +
+    "data: [DONE]\n\n";
+  globalThis.fetch = async () =>
+    new Response(sse, { status: 200, headers: { "Content-Type": "text/event-stream" } });

  const { response } = await getExecutor("command-code").execute({
    model: "gpt-5.4-mini",
@@ -532,260 +357,11 @@ test("Command Code usage chunk surfaces cache_read and no_cache for the stream p
    body: { messages: [{ role: "user", content: "Hi" }] },
  });

-  const sse = await response.text();
-  const chunks = parseSsePayloads(sse);
-  const usageChunk = chunks.find(
-    (chunk) => Array.isArray(chunk.choices) && chunk.choices.length === 0
-  );
-  assert.ok(usageChunk, "expected a usage-only chunk (choices: []) in the stream");
-
-  // The usage-only chunk feeds stream.ts's extractUsage, which surfaces
-  // cache_read_input_tokens / no_cache_tokens into the [USAGE] line.
-  const { extractUsage } = await import("../../open-sse/utils/usageTracking.ts");
-  const extracted = extractUsage(usageChunk);
-  assert.ok(extracted, "extractUsage should recognize the usage-only chunk");
-  assert.equal(extracted.prompt_tokens, 10);
-  assert.equal(extracted.completion_tokens, 6);
-  assert.equal(extracted.cache_read_input_tokens, 4);
-  assert.equal(extracted.no_cache_tokens, 6);
-});
-
-test("Command Code stream emits a usage-only chunk with actual tokens before [DONE]", async () => {
-  globalThis.fetch = async () =>
-    commandCodeStream([
-      { type: "text-delta", text: "Hi" },
-      {
-        type: "finish",
-        finishReason: "stop",
-        totalUsage: {
-          inputTokens: 10,
-          inputTokenDetails: { cacheReadTokens: 4, cacheCreationTokens: 2 },
-          outputTokens: 6,
-          reasoningTokenDetails: { reasoningTokens: 1 },
-        },
-      },
-    ]);
-
-  const { response } = await getExecutor("command-code").execute({
-    model: "gpt-5.4-mini",
-    stream: true,
-    credentials: { apiKey: "cc_test_key" },
-    body: { messages: [{ role: "user", content: "Hi" }] },
-  });
-
-  const sse = await response.text();
-  const chunks = parseSsePayloads(sse);
-
-  // Find the usage-only chunk: choices must be [] and usage must carry the
-  // actual upstream numbers. prompt_tokens = inputTokens (10) — cacheRead 4 is
-  // already included in that 10, so it is reported separately, NOT re-added.
-  const usageChunk = chunks.find(
-    (chunk) => Array.isArray(chunk.choices) && chunk.choices.length === 0
-  );
-  assert.ok(usageChunk, "expected a usage-only chunk (choices: []) in the stream");
-  assert.deepEqual(usageChunk.usage, {
-    prompt_tokens: 10,
-    prompt_tokens_details: { cached_tokens: 4 },
-    completion_tokens: 6,
-    completion_tokens_details: { reasoning_tokens: 1 },
-    total_tokens: 16,
-    cache_read_input_tokens: 4,
-    reasoning_tokens: 1,
-  });
-  // The usage chunk must come before the [DONE] marker.
-  assert.match(sse, /"usage":/);
-  const doneIndex = sse.indexOf("data: [DONE]");
-  const usageIndex = sse.indexOf(`"choices":[]`);
-  assert.ok(usageIndex > -1 && usageIndex < doneIndex, "usage chunk must precede [DONE]");
-});
-
-test("Command Code non-stream usage keeps inputTokens as prompt_tokens and reports cache separately", async () => {
-  globalThis.fetch = async () =>
-    commandCodeStream(
-      [
-        { type: "text-delta", text: "ok" },
-        {
-          type: "finish",
-          finishReason: "stop",
-          totalUsage: {
-            inputTokens: 5,
-            inputTokenDetails: { noCacheTokens: 2, cacheReadTokens: 3 },
-            outputTokens: 2,
-          },
-        },
-      ],
-      { sse: true }
-    );
-
-  const { response } = await getExecutor("command-code").execute({
-    model: "gpt-5.4-mini",
-    stream: false,
-    credentials: { apiKey: "cc_test_key" },
-    body: { messages: [{ role: "user", content: "Hi" }] },
-  });
-
-  const json = await response.json();
-  assert.deepEqual(json.usage, {
-    prompt_tokens: 5,
-    prompt_tokens_details: { cached_tokens: 3 },
-    completion_tokens: 2,
-    completion_tokens_details: { reasoning_tokens: 0 },
-    total_tokens: 7,
-    cache_read_input_tokens: 3,
-    no_cache_tokens: 2,
-  });
-});
-
-test("Command Code preserves finish-step usage through a finish without totalUsage", async () => {
-  globalThis.fetch = async () =>
-    commandCodeStream([
-      { type: "text-delta", text: "Hi" },
-      {
-        type: "finish-step",
-        usage: {
-          inputTokens: 7308,
-          inputTokenDetails: { noCacheTokens: 27, cacheReadTokens: 7281 },
-          outputTokens: 177,
-          outputTokenDetails: { textTokens: 12, reasoningTokens: 165 },
-          totalTokens: 7485,
-        },
-      },
-      { type: "finish", finishReason: "stop", totalUsage: null },
-    ]);
-
-  const { response } = await getExecutor("command-code").execute({
-    model: "gpt-5.4-mini",
-    stream: true,
-    credentials: { apiKey: "cc_test_key" },
-    body: { messages: [{ role: "user", content: "Hi" }] },
-  });
-  const sse = await response.text();
-  const chunks = parseSsePayloads(sse);
-  const usageChunk = chunks.find(
-    (chunk) => Array.isArray(chunk.choices) && chunk.choices.length === 0
-  );
-
-  assert.deepEqual(usageChunk?.usage, {
-    prompt_tokens: 7308,
-    prompt_tokens_details: { cached_tokens: 7281 },
-    completion_tokens: 177,
-    completion_tokens_details: { reasoning_tokens: 165 },
-    total_tokens: 7485,
-    cache_read_input_tokens: 7281,
-    no_cache_tokens: 27,
-    reasoning_tokens: 165,
-  });
-
-  const completed = (await responsesFromChatSse(sse)).find(
-    (event) => event.event === "response.completed"
-  );
-  assert.deepEqual(completed?.data.response.usage, {
-    input_tokens: 7308,
-    input_tokens_details: { cached_tokens: 7281 },
-    output_tokens: 177,
-    output_tokens_details: { reasoning_tokens: 165 },
-    total_tokens: 7485,
-  });
-});
-
-test("Command Code accepts OpenAI-style usage aliases with absent optional details", async () => {
-  globalThis.fetch = async () =>
-    commandCodeStream([
-      {
-        type: "finish-step",
-        usage: {
-          prompt_tokens: 11,
-          prompt_tokens_details: { cached_tokens: 4 },
-          completion_tokens: 5,
-          completion_tokens_details: { reasoning_tokens: 2 },
-          total_tokens: 16,
-        },
-      },
-      { type: "finish", finishReason: "stop" },
-    ]);
-
-  const { response } = await getExecutor("command-code").execute({
-    model: "gpt-5.4-mini",
-    stream: true,
-    credentials: { apiKey: "cc_test_key" },
-    body: { messages: [{ role: "user", content: "Hi" }] },
-  });
-  const sse = await response.text();
-  const usageChunk = parseSsePayloads(sse).find(
-    (chunk) => Array.isArray(chunk.choices) && chunk.choices.length === 0
-  );
-  assert.deepEqual(usageChunk?.usage, {
-    prompt_tokens: 11,
-    prompt_tokens_details: { cached_tokens: 4 },
-    completion_tokens: 5,
-    completion_tokens_details: { reasoning_tokens: 2 },
-    total_tokens: 16,
-    cache_read_input_tokens: 4,
-    reasoning_tokens: 2,
-  });
-
-  globalThis.fetch = async () =>
-    commandCodeStream([
-      { type: "finish-step", usage: { inputTokens: 4, outputTokens: 3, totalTokens: 7 } },
-      { type: "finish", finishReason: "stop", totalUsage: null },
-    ]);
-  const fallback = await getExecutor("command-code").execute({
-    model: "gpt-5.4-mini",
-    stream: true,
-    credentials: { apiKey: "cc_test_key" },
-    body: { messages: [{ role: "user", content: "Hi" }] },
-  });
-  const fallbackSse = await fallback.response.text();
-  const completed = (await responsesFromChatSse(fallbackSse)).find(
-    (event) => event.event === "response.completed"
-  );
-  assert.deepEqual(completed?.data.response.usage, {
-    input_tokens: 4,
-    input_tokens_details: { cached_tokens: 0 },
-    output_tokens: 3,
-    output_tokens_details: { reasoning_tokens: 0 },
-    total_tokens: 7,
-  });
-});
-
-test("Command Code preserves tool-call streaming while finalizing finish-step usage", async () => {
-  globalThis.fetch = async () =>
-    commandCodeStream([
-      {
-        type: "tool-call",
-        toolCallId: "call_1",
-        toolName: "lookup",
-        input: { query: "hello" },
-      },
-      { type: "finish-step", usage: { inputTokens: 3, outputTokens: 2, totalTokens: 5 } },
-      { type: "finish", finishReason: "tool-calls" },
-    ]);
-
-  const { response } = await getExecutor("command-code").execute({
-    model: "gpt-5.4-mini",
-    stream: true,
-    credentials: { apiKey: "cc_test_key" },
-    body: { messages: [{ role: "user", content: "Hi" }] },
-  });
-  const sse = await response.text();
-  assert.ok(
-    parseSsePayloads(sse).some(
-      (chunk) => chunk.choices?.[0]?.delta?.tool_calls?.[0]?.id === "call_1"
-    )
-  );
-
-  const completed = (await responsesFromChatSse(sse)).find(
-    (event) => event.event === "response.completed"
-  );
-  assert.equal(
-    completed?.data.response.output?.some((item) => item.type === "function_call"),
-    true
-  );
-  assert.deepEqual(completed?.data.response.usage, {
-    input_tokens: 3,
-    input_tokens_details: { cached_tokens: 0 },
-    output_tokens: 2,
-    output_tokens_details: { reasoning_tokens: 0 },
-    total_tokens: 5,
-  });
-});
+  const text = await response.text();
+  // The upstream OpenAI usage chunk passes through unchanged, including the
+  // standard OpenAI usage shape the stream pipeline already understands.
+  assert.ok(text.includes('"prompt_tokens":10'));
+  assert.ok(text.includes('"cached_tokens":4'));
+  assert.ok(text.includes('"reasoning_tokens":1'));
+  assert.ok(text.includes("data: [DONE]"));
+});
--- a/tests/unit/command-code-maxtokens-negative-5166.test.ts
+++ b/tests/unit/command-code-maxtokens-negative-5166.test.ts
@@ -34,7 +34,7 @@ test.after(() => {
  core.resetDbInstance();
 });

-async function captureParams(body: Record<string, unknown>): Promise<FetchCall> {
+async function captureBody(body: Record<string, unknown>): Promise<FetchCall> {
  const calls: FetchCall[] = [];
  globalThis.fetch = async (url: any, init: any = {}) => {
    calls.push({ url: String(url), init, body: JSON.parse(String(init.body)) });
@@ -50,27 +50,27 @@ async function captureParams(body: Record<string, unknown>): Promise<FetchCall>
 }

 test("Command Code omits max_tokens when the client sends max_tokens: -1 (#5166)", async () => {
-  const call = await captureParams({ max_tokens: -1 });
+  const call = await captureBody({ max_tokens: -1 });
  assert.ok(
-    !("max_tokens" in call.body.params),
-    `max_tokens:-1 must be omitted, got params.max_tokens=${call.body.params.max_tokens}`
+    !("max_tokens" in call.body),
+    `max_tokens:-1 must be omitted, got max_tokens=${call.body.max_tokens}`
  );
 });

 test("Command Code omits max_tokens when the client sends max_completion_tokens: -1 (#5166)", async () => {
-  const call = await captureParams({ max_completion_tokens: -1 });
+  const call = await captureBody({ max_completion_tokens: -1 });
  assert.ok(
-    !("max_tokens" in call.body.params),
-    `max_completion_tokens:-1 must be omitted, got params.max_tokens=${call.body.params.max_tokens}`
+    !("max_tokens" in call.body),
+    `max_completion_tokens:-1 must be omitted, got max_tokens=${call.body.max_tokens}`
  );
 });

 test("Command Code omits max_tokens when the client sends 0 (#5166)", async () => {
-  const call = await captureParams({ max_tokens: 0 });
-  assert.ok(!("max_tokens" in call.body.params), "max_tokens:0 must be omitted");
+  const call = await captureBody({ max_tokens: 0 });
+  assert.ok(!("max_tokens" in call.body), "max_tokens:0 must be omitted");
 });

 test("Command Code still honors a positive client max_tokens after the #5166 fix", async () => {
-  const call = await captureParams({ max_tokens: 2048 });
-  assert.equal(call.body.params.max_tokens, 2048);
-});
+  const call = await captureBody({ max_tokens: 2048 });
+  assert.equal(call.body.max_tokens, 2048);
+});
--- a/tests/unit/command-code-user-array-5166.test.ts
+++ b/tests/unit/command-code-user-array-5166.test.ts
@@ -1,14 +1,13 @@
 /**
- * Regression test for #5166 (user-content-array 400 on Command Code / deepseek-v4-pro).
+ * #5166 (user-content-array 400 on Command Code / deepseek-v4-pro) context.
 *
- * When a client sends a user message whose `content` is an array of content parts
- * (e.g. [{type:"text",text:"Hello"},{type:"text",text:"World"}]), the raw array
- * must NOT reach the Command Code upstream — it requires user content to be a plain
- * string. The executor must normalise the array to a string before posting.
- *
- * NOTE: this file covers ONLY the user-content-array/400 symptom of #5166.
- * The 0-output-token symptom on mimo-v2.5-pro (reasoning-only models) is tracked
- * separately and is NOT addressed here.
+ * The original regression was that a user message whose `content` was an array of
+ * content parts reached the CLI-only /alpha/generate endpoint, which required
+ * user content to be a plain string. Since #10265 the executor posts to the
+ * documented /provider/v1/chat/completions endpoint, which natively speaks the
+ * OpenAI chat.completions format — array content (text + image_url parts) is
+ * valid there and passes through unchanged. These tests pin that OpenAI-shaped
+ * passthrough.
 */
 import test from "node:test";
 import assert from "node:assert/strict";
@@ -26,9 +25,8 @@ const core = await import("../../src/lib/db/core.ts");

 const originalFetch = globalThis.fetch;

-function commandCodeStream(lines: unknown[]) {
-  const text = lines.map((l) => JSON.stringify(l)).join("\n") + "\n";
-  return new Response(text, { status: 200, headers: { "Content-Type": "application/x-ndjson" } });
+function okResponse() {
+  return new Response("{}", { status: 200, headers: { "Content-Type": "application/json" } });
 }

 test.after(() => {
@@ -41,155 +39,100 @@ test.afterEach(() => {
  globalThis.fetch = originalFetch;
 });

-// ── helpers ────────────────────────────────────────────────────────────────────
+// ── helpers ────────────────────────────────────────────────────────────

 type FetchCall = { url: string; init: Record<string, unknown>; body: Record<string, unknown> };

 function captureFetch(response: Response) {
  const calls: FetchCall[] = [];
  globalThis.fetch = async (url, init: RequestInit = {}) => {
-    calls.push({ url: String(url), init: init as Record<string, unknown>, body: JSON.parse(String(init.body)) });
+    calls.push({
+      url: String(url),
+      init: init as Record<string, unknown>,
+      body: JSON.parse(String(init.body)),
+    });
    return response;
  };
  return calls;
 }

-// ── failing tests (before fix, user content is the raw array) ──────────────
+test("#5166 user message with multi-part array content passes through as an OpenAI array", async () => {
+  const calls = captureFetch(okResponse());
+  await getExecutor("command-code").execute({
+    model: "deepseek/deepseek-v4-pro",
+    stream: false,
+    credentials: { apiKey: "cc_test_key" },
+    body: {
+      messages: [
+        {
+          role: "user",
+          content: [
+            { type: "text", text: "Hello" },
+            { type: "text", text: "World" },
+          ],
+        },
+      ],
+    },
+  });

-test(
-  "#5166 user message with multi-part array content is flattened to a string (#5166)",
-  async () => {
-    const calls = captureFetch(
-      commandCodeStream([{ type: "text-delta", text: "ok" }, { type: "finish" }])
-    );
+  const userMsg = (calls[0].body.messages as Record<string, unknown>[])[0];
+  // OpenAI array content is valid on /provider/v1 — forwarded as-is.
+  assert.ok(Array.isArray(userMsg.content), "array content forwarded (no CLI flattening)");
+  const parts = userMsg.content as Record<string, unknown>[];
+  assert.equal(parts.length, 2);
+  assert.equal(parts[0].text, "Hello");
+  assert.equal(parts[1].text, "World");
+});

-    await getExecutor("command-code").execute({
-      model: "deepseek/deepseek-v4-pro",
-      stream: false,
-      credentials: { apiKey: "cc_test_key" },
-      body: {
-        messages: [
-          {
-            role: "user",
-            content: [
-              { type: "text", text: "Hello" },
-              { type: "text", text: "World" },
-            ],
-          },
-        ],
-      },
-    });
+test("#5166 user message with single text-part array passes through", async () => {
+  const calls = captureFetch(okResponse());
+  await getExecutor("command-code").execute({
+    model: "deepseek/deepseek-v4-pro",
+    stream: false,
+    credentials: { apiKey: "cc_test_key" },
+    body: {
+      messages: [{ role: "user", content: [{ type: "text", text: "Hi there" }] }],
+    },
+  });
+  const userMsg = (calls[0].body.messages as Record<string, unknown>[])[0];
+  const parts = userMsg.content as Record<string, unknown>[];
+  assert.equal(parts.length, 1);
+  assert.equal(parts[0].text, "Hi there");
+});

-    const posted = calls[0].body;
-    const userMsg = (posted.params as Record<string, unknown[]>).messages[0] as Record<
-      string,
-      unknown
-    >;
+test("#5166 user message with plain string content passes through unchanged", async () => {
+  const calls = captureFetch(okResponse());
+  await getExecutor("command-code").execute({
+    model: "deepseek/deepseek-v4-pro",
+    stream: false,
+    credentials: { apiKey: "cc_test_key" },
+    body: { messages: [{ role: "user", content: "Plain string message" }] },
+  });
+  const userMsg = (calls[0].body.messages as Record<string, unknown>[])[0];
+  assert.equal(userMsg.content, "Plain string message");
+});

-    // Must be a string — never an array — otherwise Command Code's upstream returns 400.
-    assert.equal(
-      typeof userMsg.content,
-      "string",
-      `user message content must be a string, got ${typeof userMsg.content}`
-    );
-    // Joined text parts with "\n"
-    assert.equal(userMsg.content, "Hello\nWorld");
-  }
-);
-
-test(
-  "#5166 user message with single text-part array is flattened to a plain string",
-  async () => {
-    const calls = captureFetch(
-      commandCodeStream([{ type: "text-delta", text: "ok" }, { type: "finish" }])
-    );
-
-    await getExecutor("command-code").execute({
-      model: "deepseek/deepseek-v4-pro",
-      stream: false,
-      credentials: { apiKey: "cc_test_key" },
-      body: {
-        messages: [
-          {
-            role: "user",
-            content: [{ type: "text", text: "Hi there" }],
-          },
-        ],
-      },
-    });
-
-    const posted = calls[0].body;
-    const userMsg = (posted.params as Record<string, unknown[]>).messages[0] as Record<
-      string,
-      unknown
-    >;
-    assert.equal(typeof userMsg.content, "string");
-    assert.equal(userMsg.content, "Hi there");
-  }
-);
-
-test(
-  "#5166 user message with plain string content passes through unchanged (no regression)",
-  async () => {
-    const calls = captureFetch(
-      commandCodeStream([{ type: "text-delta", text: "ok" }, { type: "finish" }])
-    );
-
-    await getExecutor("command-code").execute({
-      model: "deepseek/deepseek-v4-pro",
-      stream: false,
-      credentials: { apiKey: "cc_test_key" },
-      body: {
-        messages: [
-          {
-            role: "user",
-            content: "Plain string message",
-          },
-        ],
-      },
-    });
-
-    const posted = calls[0].body;
-    const userMsg = (posted.params as Record<string, unknown[]>).messages[0] as Record<
-      string,
-      unknown
-    >;
-    assert.equal(typeof userMsg.content, "string");
-    assert.equal(userMsg.content, "Plain string message");
-  }
-);
-
-test(
-  "#5166 user message with mixed parts (text + image_url) keeps only text parts",
-  async () => {
-    const calls = captureFetch(
-      commandCodeStream([{ type: "text-delta", text: "ok" }, { type: "finish" }])
-    );
-
-    await getExecutor("command-code").execute({
-      model: "deepseek/deepseek-v4-pro",
-      stream: false,
-      credentials: { apiKey: "cc_test_key" },
-      body: {
-        messages: [
-          {
-            role: "user",
-            content: [
-              { type: "text", text: "Describe this:" },
-              { type: "image_url", image_url: { url: "https://example.com/img.png" } },
-            ],
-          },
-        ],
-      },
-    });
-
-    const posted = calls[0].body;
-    const userMsg = (posted.params as Record<string, unknown[]>).messages[0] as Record<
-      string,
-      unknown
-    >;
-    assert.equal(typeof userMsg.content, "string");
-    // Only text parts extracted; image_url part is dropped (not a "text" type)
-    assert.equal(userMsg.content, "Describe this:");
-  }
-);
+test("#5166 user message with mixed parts (text + image_url) keeps all parts", async () => {
+  const calls = captureFetch(okResponse());
+  await getExecutor("command-code").execute({
+    model: "deepseek/deepseek-v4-pro",
+    stream: false,
+    credentials: { apiKey: "cc_test_key" },
+    body: {
+      messages: [
+        {
+          role: "user",
+          content: [
+            { type: "text", text: "Describe this:" },
+            { type: "image_url", image_url: { url: "https://example.com/img.png" } },
+          ],
+        },
+      ],
+    },
+  });
+  const userMsg = (calls[0].body.messages as Record<string, unknown>[])[0];
+  const parts = userMsg.content as Record<string, unknown>[];
+  assert.equal(parts.length, 2, "text + image both preserved");
+  assert.equal(parts[0].text, "Describe this:");
+  assert.equal(parts[1].type, "image_url");
+});
--- a/tests/unit/command-code-vision.test.ts
+++ b/tests/unit/command-code-vision.test.ts
@@ -1,9 +1,13 @@
 /**
 * Vision / multimodal support tests for the Command Code executor.
 *
- * Verifies that vision-capable models (MiniMax M3, MiMo V2.5, Kimi K2, Qwen 3.x, GPT-5, Claude 3/4, Fable 5, Gemini 3.x, Stepfun, Fugu, etc.)
- * receive image parts in Command Code CLI format, while text-only
- * models strip images as before (no regression).
+ * Since #10265 the executor posts to the documented /provider/v1/chat/completions
+ * endpoint, which speaks the standard OpenAI chat.completions format. User image
+ * content (OpenAI `image_url` parts and Anthropic Messages-style source blocks)
+ * passes through unchanged — the endpoint natively understands both shapes, so
+ * there is no CLI-specific conversion (and no CLI-wire image stripping) left to
+ * verify. These tests pin that passthrough plus the #10809 wire-model
+ * normalization, which still applies to /provider/v1.
 */
 import test from "node:test";
 import assert from "node:assert/strict";
@@ -19,9 +23,8 @@ const core = await import("../../src/lib/db/core.ts");

 const originalFetch = globalThis.fetch;

-function commandCodeStream(lines: unknown[]) {
-  const text = lines.map((l) => JSON.stringify(l)).join("\n") + "\n";
-  return new Response(text, { status: 200, headers: { "Content-Type": "application/x-ndjson" } });
+function okResponse() {
+  return new Response("{}", { status: 200, headers: { "Content-Type": "application/json" } });
 }

 test.after(() => {
@@ -52,44 +55,28 @@ function captureFetch(response: Response) {
 }

 function userContent(calls: FetchCall[]): unknown {
-  return (
-    (calls[0].body.params as Record<string, unknown[]>).messages as Record<string, unknown>[]
-  )[0].content;
+  return (calls[0].body.messages as Record<string, unknown>[])[0].content;
+}
+
+function wireModel(calls: FetchCall[]): string {
+  return calls[0].body.model as string;
 }

 // ── wire model normalization (#10809) ────────────────────────────────
-//
-// Command Code's /alpha/generate endpoint serves most models under a
-// vendor-prefixed wire id and defaults an unprefixed id to the `anthropic:`
-// provider (403 "Model/provider not recognized: anthropic:<id>"). A bare id
-// reaches the executor when an operator sets a custom vision model in the
-// Vision Bridge picker (e.g. `command-code/mimo-v2.5`). The executor must
-// normalize to the documented vendor-prefixed wire form.
-
-function wireModel(calls: FetchCall[]): string {
-  return (calls[0].body.params as Record<string, unknown>).model as string;
-}

 test("#10809: command-code/mimo-v2.5 wire model is normalized to xiaomi/mimo-v2.5", async () => {
-  const calls = captureFetch(
-    commandCodeStream([{ type: "text-delta", text: "ok" }, { type: "finish" }])
-  );
+  const calls = captureFetch(okResponse());
  await getExecutor("command-code").execute({
    model: "command-code/mimo-v2.5",
    stream: false,
    credentials: { apiKey: "cc_test_key" },
-    body: {
-      model: "command-code/mimo-v2.5",
-      messages: [{ role: "user", content: "hi" }],
-    },
+    body: { model: "command-code/mimo-v2.5", messages: [{ role: "user", content: "hi" }] },
  });
  assert.equal(wireModel(calls), "xiaomi/mimo-v2.5");
 });

 test("#10809: cmd/mimo-v2.5 (alias prefix) is also normalized", async () => {
-  const calls = captureFetch(
-    commandCodeStream([{ type: "text-delta", text: "ok" }, { type: "finish" }])
-  );
+  const calls = captureFetch(okResponse());
  await getExecutor("command-code").execute({
    model: "cmd/mimo-v2.5",
    stream: false,
@@ -100,9 +87,7 @@ test("#10809: cmd/mimo-v2.5 (alias prefix) is also normalized", async () => {
 });

 test("#10809: already vendor-prefixed wire ids pass through unchanged", async () => {
-  const calls = captureFetch(
-    commandCodeStream([{ type: "text-delta", text: "ok" }, { type: "finish" }])
-  );
+  const calls = captureFetch(okResponse());
  await getExecutor("command-code").execute({
    model: "command-code/deepseek/deepseek-v4-pro",
    stream: false,
@@ -115,13 +100,10 @@ test("#10809: already vendor-prefixed wire ids pass through unchanged", async ()
  assert.equal(wireModel(calls), "deepseek/deepseek-v4-pro");
 });

-// ── vision models: image parts preserved in CC CLI format ─────────────
-
-test("vision model minimax-m3 preserves image_url part as CC CLI {type:image}", async () => {
-  const calls = captureFetch(
-    commandCodeStream([{ type: "text-delta", text: "ok" }, { type: "finish" }])
-  );
+// ── image content passthrough (OpenAI /provider/v1 surface) ──────────

+test("image_url parts pass through unchanged (text + image preserved)", async () => {
+  const calls = captureFetch(okResponse());
  await getExecutor("command-code").execute({
    model: "MiniMaxAI/MiniMax-M3",
    stream: false,
@@ -132,451 +114,25 @@ test("vision model minimax-m3 preserves image_url part as CC CLI {type:image}",
          role: "user",
          content: [
            { type: "text", text: "What's in this?" },
-            {
-              type: "image_url",
-              image_url: { url: "data:image/png;base64,iVBORw0KGgo=" },
-            },
+            { type: "image_url", image_url: { url: "data:image/png;base64,iVBORw0KGgo=" } },
          ],
        },
      ],
    },
  });

-  const content = userContent(calls);
-  assert.ok(Array.isArray(content), "vision model user content must be an array");
-  const parts = content as Record<string, unknown>[];
-  assert.equal(parts.length, 2);
-
-  // Text part preserved
-  assert.equal(parts[0].type, "text");
-  assert.equal(parts[0].text, "What's in this?");
-
-  // Image part converted to CC CLI format
-  assert.equal(parts[1].type, "image");
-  assert.equal(parts[1].image, "data:image/png;base64,iVBORw0KGgo=");
+  const content = userContent(calls) as Record<string, unknown>[];
+  assert.equal(content.length, 2);
+  assert.equal(content[0].type, "text");
+  assert.equal(content[1].type, "image_url", "image_url part preserved as-is");
+  assert.equal(
+    (content[1].image_url as { url: string }).url,
+    "data:image/png;base64,iVBORw0KGgo="
+  );
 });

-test("vision model minimax-m3 preserves image_url with HTTP URL", async () => {
-  const calls = captureFetch(
-    commandCodeStream([{ type: "text-delta", text: "ok" }, { type: "finish" }])
-  );
-
-  await getExecutor("command-code").execute({
-    model: "minimax-m3",
-    stream: false,
-    credentials: { apiKey: "cc_test_key" },
-    body: {
-      messages: [
-        {
-          role: "user",
-          content: [
-            { type: "text", text: "Describe" },
-            {
-              type: "image_url",
-              image_url: { url: "https://example.com/photo.jpg" },
-            },
-          ],
-        },
-      ],
-    },
-  });
-
-  const content = userContent(calls);
-  assert.ok(Array.isArray(content));
-  const parts = content as Record<string, unknown>[];
-  assert.equal(parts.length, 2);
-  assert.equal(parts[1].type, "image");
-  assert.equal(parts[1].image, "https://example.com/photo.jpg");
-});
-
-test("vision model mimo-v2.5 preserves image parts", async () => {
-  const calls = captureFetch(
-    commandCodeStream([{ type: "text-delta", text: "ok" }, { type: "finish" }])
-  );
-
-  await getExecutor("command-code").execute({
-    model: "mimo-v2.5",
-    stream: false,
-    credentials: { apiKey: "cc_test_key" },
-    body: {
-      messages: [
-        {
-          role: "user",
-          content: [
-            { type: "text", text: "Analyze" },
-            {
-              type: "image_url",
-              image_url: { url: "https://example.com/img.png" },
-            },
-          ],
-        },
-      ],
-    },
-  });
-
-  const content = userContent(calls);
-  assert.ok(Array.isArray(content));
-  const parts = content as Record<string, unknown>[];
-  assert.equal(parts.length, 2);
-  assert.equal(parts[1].type, "image");
-});
-
-test("vision model mimo-v2.5-pro is text-only (no image parts)", async () => {
-  const calls = captureFetch(
-    commandCodeStream([{ type: "text-delta", text: "ok" }, { type: "finish" }])
-  );
-
-  await getExecutor("command-code").execute({
-    model: "mimo-v2.5-pro",
-    stream: false,
-    credentials: { apiKey: "cc_test_key" },
-    body: {
-      messages: [
-        {
-          role: "user",
-          content: [
-            { type: "text", text: "Hi" },
-            {
-              type: "image_url",
-              image_url: { url: "https://example.com/img.png" },
-            },
-          ],
-        },
-      ],
-    },
-  });
-
-  const content = userContent(calls);
-  // mimo-v2.5-pro is text-only — content must be flattened to a plain string
-  assert.equal(typeof content, "string");
-  assert.equal(content, "Hi");
-});
-
-test("vision model mimo-v2-omni preserves image parts", async () => {
-  const calls = captureFetch(
-    commandCodeStream([{ type: "text-delta", text: "ok" }, { type: "finish" }])
-  );
-
-  await getExecutor("command-code").execute({
-    model: "mimo-v2-omni",
-    stream: false,
-    credentials: { apiKey: "cc_test_key" },
-    body: {
-      messages: [
-        {
-          role: "user",
-          content: [
-            { type: "text", text: "Check" },
-            {
-              type: "image_url",
-              image_url: { url: "data:image/jpeg;base64,/9j/4AAQ=" },
-            },
-          ],
-        },
-      ],
-    },
-  });
-
-  const content = userContent(calls);
-  assert.ok(Array.isArray(content));
-  const parts = content as Record<string, unknown>[];
-  assert.equal(parts.length, 2);
-  assert.equal(parts[1].type, "image");
-  assert.equal(parts[1].image, "data:image/jpeg;base64,/9j/4AAQ=");
-});
-
-// ── non-vision models: images still stripped (no regression) ──────────
-
-test("text-only model deepseek-v4-pro strips image_url parts", async () => {
-  const calls = captureFetch(
-    commandCodeStream([{ type: "text-delta", text: "ok" }, { type: "finish" }])
-  );
-
-  await getExecutor("command-code").execute({
-    model: "deepseek/deepseek-v4-pro",
-    stream: false,
-    credentials: { apiKey: "cc_test_key" },
-    body: {
-      messages: [
-        {
-          role: "user",
-          content: [
-            { type: "text", text: "Hello" },
-            {
-              type: "image_url",
-              image_url: { url: "https://example.com/img.png" },
-            },
-          ],
-        },
-      ],
-    },
-  });
-
-  const content = userContent(calls);
-  // Non-vision model: content is a plain string, images stripped
-  assert.equal(typeof content, "string");
-  assert.equal(content, "Hello");
-});
-
-test("text-only model deepseek-v4-flash strips image_url parts (no regression)", async () => {
-  const calls = captureFetch(
-    commandCodeStream([{ type: "text-delta", text: "ok" }, { type: "finish" }])
-  );
-
-  await getExecutor("command-code").execute({
-    model: "deepseek/deepseek-v4-flash",
-    stream: false,
-    credentials: { apiKey: "cc_test_key" },
-    body: {
-      messages: [
-        {
-          role: "user",
-          content: [
-            { type: "text", text: "Text only" },
-            {
-              type: "image_url",
-              image_url: { url: "data:image/png;base64,AAA=" },
-            },
-          ],
-        },
-      ],
-    },
-  });
-
-  const content = userContent(calls);
-  assert.equal(typeof content, "string");
-  assert.equal(content, "Text only");
-});
-
-// ── edge cases ────────────────────────────────────────────────────────
-
-test("vision model with only image content emits empty text fallback", async () => {
-  const calls = captureFetch(
-    commandCodeStream([{ type: "text-delta", text: "ok" }, { type: "finish" }])
-  );
-
-  await getExecutor("command-code").execute({
-    model: "minimax-m3",
-    stream: false,
-    credentials: { apiKey: "cc_test_key" },
-    body: {
-      messages: [
-        {
-          role: "user",
-          content: [
-            {
-              type: "image_url",
-              image_url: { url: "data:image/png;base64,iVBOR=" },
-            },
-          ],
-        },
-      ],
-    },
-  });
-
-  const content = userContent(calls);
-  assert.ok(Array.isArray(content));
-  const parts = content as Record<string, unknown>[];
-  // Single image part preserved — no empty text injected because
-  // the image itself keeps content non-empty.
-  assert.equal(parts.length, 1);
-  assert.equal(parts[0].type, "image");
-  assert.equal(parts[0].image, "data:image/png;base64,iVBOR=");
-});
-
-test("vision model passes plain string content through unchanged", async () => {
-  const calls = captureFetch(
-    commandCodeStream([{ type: "text-delta", text: "ok" }, { type: "finish" }])
-  );
-
-  await getExecutor("command-code").execute({
-    model: "minimax-m3",
-    stream: false,
-    credentials: { apiKey: "cc_test_key" },
-    body: {
-      messages: [{ role: "user", content: "Plain string message" }],
-    },
-  });
-
-  const content = userContent(calls);
-  assert.equal(typeof content, "string");
-  assert.equal(content, "Plain string message");
-});
-
-test("vision model honors body.model rewrite for vision detection", async () => {
-  // #5166 scenario: body.model overwrites the execute model arg.
-  // Vision detection must use the rewritten model id.
-  const calls = captureFetch(
-    commandCodeStream([{ type: "text-delta", text: "ok" }, { type: "finish" }])
-  );
-
-  // execute() gets a non-vision combo model, body.model rewrites to a vision model
-  await getExecutor("command-code").execute({
-    model: "gpt-5.4-mini",
-    stream: false,
-    credentials: { apiKey: "cc_test_key" },
-    body: {
-      model: "MiniMaxAI/MiniMax-M3",
-      messages: [
-        {
-          role: "user",
-          content: [
-            { type: "text", text: "Describe" },
-            {
-              type: "image_url",
-              image_url: { url: "https://example.com/img.png" },
-            },
-          ],
-        },
-      ],
-    },
-  });
-
-  const content = userContent(calls);
-  // body.model = MiniMax-M3 (vision) → images preserved
-  assert.ok(Array.isArray(content));
-  const parts = content as Record<string, unknown>[];
-  assert.equal(parts.length, 2);
-  assert.equal(parts[1].type, "image");
-});
-
-test("vision model with multiple image parts preserves all of them", async () => {
-  const calls = captureFetch(
-    commandCodeStream([{ type: "text-delta", text: "ok" }, { type: "finish" }])
-  );
-
-  await getExecutor("command-code").execute({
-    model: "minimax-m3",
-    stream: false,
-    credentials: { apiKey: "cc_test_key" },
-    body: {
-      messages: [
-        {
-          role: "user",
-          content: [
-            { type: "text", text: "Compare" },
-            {
-              type: "image_url",
-              image_url: { url: "https://example.com/a.jpg" },
-            },
-            {
-              type: "image_url",
-              image_url: { url: "https://example.com/b.jpg" },
-            },
-          ],
-        },
-      ],
-    },
-  });
-
-  const content = userContent(calls);
-  assert.ok(Array.isArray(content));
-  const parts = content as Record<string, unknown>[];
-  assert.equal(parts.length, 3);
-  assert.equal(parts[0].type, "text");
-  assert.equal(parts[1].type, "image");
-  assert.equal(parts[1].image, "https://example.com/a.jpg");
-  assert.equal(parts[2].type, "image");
-  assert.equal(parts[2].image, "https://example.com/b.jpg");
-});
-
-test("vision model with image_url as plain string (no object wrapper) still works", async () => {
-  const calls = captureFetch(
-    commandCodeStream([{ type: "text-delta", text: "ok" }, { type: "finish" }])
-  );
-
-  await getExecutor("command-code").execute({
-    model: "minimax-m3",
-    stream: false,
-    credentials: { apiKey: "cc_test_key" },
-    body: {
-      messages: [
-        {
-          role: "user",
-          content: [
-            { type: "text", text: "Look" },
-            {
-              type: "image_url",
-              image_url: "https://example.com/img.png",
-            },
-          ],
-        },
-      ],
-    },
-  });
-
-  const content = userContent(calls);
-  assert.ok(Array.isArray(content));
-  const parts = content as Record<string, unknown>[];
-  assert.equal(parts.length, 2);
-  assert.equal(parts[1].type, "image");
-  assert.equal(parts[1].image, "https://example.com/img.png");
-});
-
-// ── CC vision models (Command Code docs registry) ──────────────────
-
-const VISION_CASES = [
-  ["Kimi K2.6", "moonshotai/Kimi-K2.6"],
-  ["Kimi K2.7 Code", "moonshotai/Kimi-K2.7-Code"],
-  ["Kimi K2.5", "moonshotai/Kimi-K2.5"],
-  ["Qwen 3.6 Plus", "Qwen/Qwen3.6-Plus"],
-  ["Qwen 3.7 Plus", "Qwen/Qwen3.7-Plus"],
-  ["Step 3.7 Flash", "stepfun/Step-3.7-Flash"],
-  ["GPT-5.5", "gpt-5.5"],
-  ["GPT-5.4", "gpt-5.4"],
-  ["GPT-5.3 Codex", "gpt-5.3-codex"],
-  ["GPT-5.4 Mini", "gpt-5.4-mini"],
-  ["Claude Fable 5", "claude-fable-5"],
-  ["Sakana Fugu Ultra", "sakana/fugu-ultra"],
-  ["Claude Opus 4.7 (isVisionModelId)", "claude-opus-4-7"],
-  ["Claude Sonnet 4.6 (isVisionModelId)", "claude-sonnet-4-6"],
-  ["Gemini 3.5 Flash (isVisionModelId)", "google/gemini-3.5-flash"],
-];
-
-for (const [name, model] of VISION_CASES) {
-  test(`vision model ${name} preserves image parts`, async () => {
-    const calls = captureFetch(
-      commandCodeStream([{ type: "text-delta", text: "ok" }, { type: "finish" }])
-    );
-    await getExecutor("command-code").execute({
-      model,
-      stream: false,
-      credentials: { apiKey: "cc_test_key" },
-      body: {
-        messages: [
-          {
-            role: "user",
-            content: [
-              { type: "text", text: "Check" },
-              {
-                type: "image_url",
-                image_url: { url: "https://example.com/img.png" },
-              },
-            ],
-          },
-        ],
-      },
-    });
-    const content = userContent(calls);
-    assert.ok(Array.isArray(content), `${name} user content must be an array`);
-    const parts = content;
-    assert.equal(parts.length, 2);
-    assert.equal(parts[1].type, "image");
-  });
-}
-
-// ── Anthropic-shaped image blocks (Zoo Code / Claude-Code-compatible clients) ──
-
-test("vision model mimo-v2.5 preserves Anthropic source.base64 image block", async () => {
-  // Zoo Code sends Messages-API-shaped content blocks to the OpenAI
-  // /v1/chat/completions surface: { type:"image", source:{ base64 } }.
-  // The vision-bridge guardrail skips vision-capable models (cmd/xiaomi/mimo-v2.5
-  // resolves supportsVision=true via the mimo-v2.5 leaf spec), so the raw block
-  // must survive to the executor and be converted to CC CLI { type:"image" }.
-  const calls = captureFetch(
-    commandCodeStream([{ type: "text-delta", text: "ok" }, { type: "finish" }])
-  );
-
+test("Anthropic Messages-style source image blocks pass through unchanged", async () => {
+  const calls = captureFetch(okResponse());
  await getExecutor("command-code").execute({
    model: "xiaomi/mimo-v2.5",
    stream: false,
@@ -601,24 +157,15 @@ test("vision model mimo-v2.5 preserves Anthropic source.base64 image block", asy
    },
  });

-  const content = userContent(calls);
-  assert.ok(Array.isArray(content), "user content must be an array");
-  const parts = content as Record<string, unknown>[];
-  assert.equal(parts.length, 2, "text + image parts preserved");
-  assert.equal(parts[0].type, "text");
-  assert.equal(parts[1].type, "image");
-  assert.equal(
-    parts[1].image,
-    "data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAAEAAAABCAYAAAAfFcSJAAAADUlEQVR42mP8z8BQDwAEhQGAhKmMIQAAAABJRU5ErkJggg==",
-    "base64 payload is rebuilt into a CC CLI data URL"
-  );
+  const content = userContent(calls) as Record<string, unknown>[];
+  assert.equal(content.length, 2, "text + image parts preserved");
+  assert.equal(content[1].type, "image");
+  assert.equal((content[1].source as { type: string }).type, "base64");
+  assert.equal((content[1].source as { data: string }).data, "iVBORw0KGgoAAAANSUhEUgAAAAEAAAABCAYAAAAfFcSJAAAADUlEQVR42mP8z8BQDwAEhQGAhKmMIQAAAABJRU5ErkJggg==");
 });

-test("vision model preserves Anthropic source.url image block", async () => {
-  const calls = captureFetch(
-    commandCodeStream([{ type: "text-delta", text: "ok" }, { type: "finish" }])
-  );
-
+test("Anthropic source.url image block passes through unchanged", async () => {
+  const calls = captureFetch(okResponse());
  await getExecutor("command-code").execute({
    model: "xiaomi/mimo-v2.5",
    stream: false,
@@ -629,31 +176,23 @@ test("vision model preserves Anthropic source.url image block", async () => {
          role: "user",
          content: [
            { type: "text", text: "Look" },
-            {
-              type: "image",
-              source: { type: "url", url: "https://example.com/img.png" },
-            },
+            { type: "image", source: { type: "url", url: "https://example.com/img.png" } },
          ],
        },
      ],
    },
  });

-  const content = userContent(calls);
-  assert.ok(Array.isArray(content));
-  const parts = content as Record<string, unknown>[];
-  assert.equal(parts.length, 2);
-  assert.equal(parts[1].type, "image");
-  assert.equal(parts[1].image, "https://example.com/img.png");
+  const content = userContent(calls) as Record<string, unknown>[];
+  assert.equal(content.length, 2);
+  assert.equal(content[1].type, "image");
+  assert.deepEqual(content[1].source, { type: "url", url: "https://example.com/img.png" });
 });

-test("text-only model deepseek-v4-flash strips Anthropic source.base64 image block", async () => {
-  const calls = captureFetch(
-    commandCodeStream([{ type: "text-delta", text: "ok" }, { type: "finish" }])
-  );
-
+test("multiple image parts are all preserved", async () => {
+  const calls = captureFetch(okResponse());
  await getExecutor("command-code").execute({
-    model: "deepseek/deepseek-v4-flash",
+    model: "minimax-m3",
    stream: false,
    credentials: { apiKey: "cc_test_key" },
    body: {
@@ -661,44 +200,41 @@ test("text-only model deepseek-v4-flash strips Anthropic source.base64 image blo
        {
          role: "user",
          content: [
-            { type: "text", text: "Text only" },
-            {
-              type: "image",
-              source: {
-                type: "base64",
-                media_type: "image/png",
-                data: "iVBORw0KGgoAAAANSUhEUgAAAAEAAAABCAYAAAAfFcSJAAAADUlEQVR42mP8z8BQDwAEhQGAhKmMIQAAAABJRU5ErkJggg==",
-              },
-            },
+            { type: "text", text: "Compare" },
+            { type: "image_url", image_url: { url: "https://example.com/a.jpg" } },
+            { type: "image_url", image_url: { url: "https://example.com/b.jpg" } },
          ],
        },
      ],
    },
  });

+  const content = userContent(calls) as Record<string, unknown>[];
+  assert.equal(content.length, 3);
+  assert.equal(content[1].type, "image_url");
+  assert.equal(content[2].type, "image_url");
+});
+
+test("plain string content passes through unchanged", async () => {
+  const calls = captureFetch(okResponse());
+  await getExecutor("command-code").execute({
+    model: "minimax-m3",
+    stream: false,
+    credentials: { apiKey: "cc_test_key" },
+    body: { messages: [{ role: "user", content: "Plain string message" }] },
+  });
+
  const content = userContent(calls);
-  // Text-only model: content flattened to plain string, image stripped.
  assert.equal(typeof content, "string");
-  assert.equal(content, "Text only");
+  assert.equal(content, "Plain string message");
 });

-// ── conservative vision family lock (gpt-5.4-mini / gpt-5.3-codex) ─────
-//
-// These two ids stay INSIDE the `/gpt-5/` vision family: both accept image
-// input on the OpenAI API, and there is no verified Command Code backend data
-// marking them text-only. These tests pin that conservative executor behavior
-// so a future "narrow the regex" change cannot silently strip images from models
-// that can see them (the #4071 regression class). The #10703 Vision Bridge
-// candidate-list fix lives in the shared capability resolution
-// (KNOWN_TEXT_ONLY_DESPITE_SYNC), NOT in the executor's wire transform.
-
-test("gpt-5.4-mini keeps image parts (conservative vision family lock)", async () => {
-  const calls = captureFetch(
-    commandCodeStream([{ type: "text-delta", text: "ok" }, { type: "finish" }])
-  );
-
+test("text-only model still forwards image parts (passthrough, no CLI stripping)", async () => {
+  // The /provider/v1 OpenAI surface accepts image content for any model id; the
+  // executor forwards content untouched, so there is no text-only stripping.
+  const calls = captureFetch(okResponse());
  await getExecutor("command-code").execute({
-    model: "gpt-5.4-mini",
+    model: "deepseek/deepseek-v4-pro",
    stream: false,
    credentials: { apiKey: "cc_test_key" },
    body: {
@@ -706,54 +242,15 @@ test("gpt-5.4-mini keeps image parts (conservative vision family lock)", async (
        {
          role: "user",
          content: [
-            { type: "text", text: "What's in this?" },
-            {
-              type: "image_url",
-              image_url: { url: "https://example.com/img.png" },
-            },
+            { type: "text", text: "Hello" },
+            { type: "image_url", image_url: { url: "https://example.com/img.png" } },
          ],
        },
      ],
    },
  });

-  const content = userContent(calls);
-  assert.ok(Array.isArray(content), "gpt-5.4-mini must be treated as vision-capable");
-  const parts = content as Record<string, unknown>[];
-  assert.equal(parts.length, 2);
-  assert.equal(parts[1].type, "image");
-  assert.equal(parts[1].image, "https://example.com/img.png");
-});
-
-test("gpt-5.3-codex keeps image parts (conservative vision family lock)", async () => {
-  const calls = captureFetch(
-    commandCodeStream([{ type: "text-delta", text: "ok" }, { type: "finish" }])
-  );
-
-  await getExecutor("command-code").execute({
-    model: "gpt-5.3-codex",
-    stream: false,
-    credentials: { apiKey: "cc_test_key" },
-    body: {
-      messages: [
-        {
-          role: "user",
-          content: [
-            { type: "text", text: "Describe" },
-            {
-              type: "image_url",
-              image_url: { url: "https://example.com/img.png" },
-            },
-          ],
-        },
-      ],
-    },
-  });
-
-  const content = userContent(calls);
-  assert.ok(Array.isArray(content), "gpt-5.3-codex must be treated as vision-capable");
-  const parts = content as Record<string, unknown>[];
-  assert.equal(parts.length, 2);
-  assert.equal(parts[1].type, "image");
-  assert.equal(parts[1].image, "https://example.com/img.png");
-});
+  const content = userContent(calls) as Record<string, unknown>[];
+  assert.equal(content.length, 2, "content array forwarded unchanged");
+  assert.equal(content[1].type, "image_url");
+});
--- a/tests/unit/copilot-m365-enterprise-invocation-7870.test.ts
+++ b/tests/unit/copilot-m365-enterprise-invocation-7870.test.ts
@@ -15,7 +15,10 @@ class MockM365WebSocket {
  closed = false;
  listeners = new Map<string, Listener[]>();

-  constructor(public url: string, public options: unknown) {
+  constructor(
+    public url: string,
+    public options: unknown
+  ) {
    MockM365WebSocket.instances.push(this);
    queueMicrotask(() => this.emit("open"));
  }
@@ -144,30 +147,55 @@ test("#7870: enterprise-tier chat invocation defaults tone to Magic", async () =
  assert.equal(invocationArgs.tone, "Magic");
 });

-test("#10718: individual (no tier) chat invocation carries the recaptured 2026-08 shape", async () => {
+test("2026-08-21: individual (no tier) chat invocation carries the recaptured shape", async () => {
  const invocationArgs = await sendChatInvocation(undefined);
  const optionsSets = invocationArgs.optionsSets as string[];
-  // The 25-entry consumer/MSA set (enable_msa_user, pdnascan, …) is gone from
-  // the wire — the stale set was part of the silently-dropped shape.
+  // The pre-#10718 25-entry consumer/MSA set (enable_msa_user, pdnascan, …) is
+  // still gone from the wire — only the 2026-08-21 34-entry set is current.
  assert.ok(!optionsSets.includes("enable_msa_user"));
  assert.ok(optionsSets.includes("enable_gg_gpt"));
-  // The browser sends tone:"magic" (lowercase) on the individual/EDU surface.
-  assert.equal(invocationArgs.tone, "magic");
+  // The individual/consumer surface now sends tone:"Magic" (capitalized) —
+  // the #10718 lowercase "magic" is stale.
+  assert.equal(invocationArgs.tone, "Magic");
  assert.deepEqual(invocationArgs.allowedMessageTypes, [
    "Chat",
    "Suggestion",
+    "InternalSearchQuery",
    "Disengaged",
-    "Progress",
-    "EndOfRequest",
    "InternalLoaderMessage",
+    "Progress",
+    "GeneratedCode",
+    "RenderCardRequest",
+    "AdsQuery",
+    "SemanticSerp",
+    "GenerateContentQuery",
+    "GenerateGraphicArt",
+    "SearchQuery",
+    "ConfirmationCard",
+    "AuthError",
+    "DeveloperLogs",
+    "TriggerPlugin",
+    "HintInvocation",
+    "MemoryUpdate",
+    "EndOfRequest",
+    "TriggerConfirmation",
+    "ResumeInvokeAction",
+    "ResumeUserInputRequest",
+    "TriggerUserInputRequest",
+    "EscapeHatch",
+    "TriggerPluginAuth",
+    "ResumePluginAuth",
+    "SideBySide",
+    "ReferencesListComplete",
+    "SwitchRespondingEndpoint",
  ]);
 });

-test("#10718: EDU-tier chat invocation carries the same recaptured shape", async () => {
+test("2026-08-21: EDU-tier chat invocation carries the same recaptured shape", async () => {
  const invocationArgs = await sendChatInvocation("edu");
  const optionsSets = invocationArgs.optionsSets as string[];
  assert.ok(!optionsSets.includes("enable_msa_user"));
-  assert.equal(invocationArgs.tone, "magic");
+  assert.equal(invocationArgs.tone, "Magic");
 });

 test("#8971: enterprise-tier chat invocation must send disconnectBehavior=continue", async () => {
@@ -179,11 +207,11 @@ test("#8971: enterprise-tier chat invocation must send disconnectBehavior=contin
  );
 });

-test("#8971/#10718: individual (no tier) chat invocation omits disconnectBehavior (not on the 2026-08 wire)", async () => {
+test("2026-08-21: individual (no tier) chat invocation also sends disconnectBehavior=continue (now on every tier)", async () => {
  const invocationArgs = await sendChatInvocation(undefined);
  assert.equal(
    invocationArgs.disconnectBehavior,
-    undefined,
-    `individual-tier invocation must omit disconnectBehavior; got ${JSON.stringify(invocationArgs.disconnectBehavior)}`
+    "continue",
+    `individual-tier invocation must carry disconnectBehavior="continue"; got ${JSON.stringify(invocationArgs.disconnectBehavior)}`
  );
 });
--- a/tests/unit/copilot-m365-invocation-refresh-10718.test.ts
+++ b/tests/unit/copilot-m365-invocation-refresh-10718.test.ts
@@ -41,7 +41,7 @@ test("#10718: metricsFrame emits the exact bytes observed in the browser capture

 // ── Recaptured invocation shape ────────────────────────────────────────────

-test("#10718: buildChatInvocation matches the recaptured arguments[0] key set", () => {
+test("2026-08-21: buildChatInvocation matches the recaptured arguments[0] key set", () => {
  const arg = buildChatInvocation({
    text: "Say OK in one word.",
    traceId: "11111111-1111-1111-1111-111111111111",
@@ -50,19 +50,25 @@ test("#10718: buildChatInvocation matches the recaptured arguments[0] key set",
    conversationId: "44444444-4444-4444-4444-444444444444",
  }).arguments[0] as Record<string, unknown>;

-  // Exact key set from the capture — additions AND omissions are both pinned,
-  // because the stale keys are exactly what got the shape dropped.
+  // Exact key set from the 2026-08-21 capture (issue: individual/consumer M365
+  // Copilot calls got only SignalR keepalive pings and no type:1 update at all
+  // — "Stream ended before producing a non-ping SSE event" client-side). The
+  // #10718 shape below is missing exactly the keys this capture added.
  assert.deepEqual(Object.keys(arg).sort(), [
    "allowedMessageTypes",
    "clientCorrelationId",
    "clientInfo",
    "conversationId",
+    "disconnectBehavior",
+    "extraExtensionParameters",
+    "isSbsSupported",
    "isStartOfSession",
    "message",
    "options",
    "optionsSets",
    "plugins",
    "productThreadType",
+    "renderReferencesBehindEOS",
    "sessionId",
    "sliceIds",
    "source",
@@ -73,13 +79,32 @@ test("#10718: buildChatInvocation matches the recaptured arguments[0] key set",
    "traceId",
  ]);
  assert.equal(arg.productThreadType, "Office");
-  assert.deepEqual(arg.clientInfo, { clientAppName: "Office", clientPlatform: "mcmcopilot-web" });
+  assert.deepEqual(arg.clientInfo, {
+    clientAppName: "Office",
+    clientPlatform: "mcmcopilot-web",
+    clientEntrypoint: "mcmcopilot-officeweb",
+    clientSessionId: "22222222-2222-2222-2222-222222222222",
+    ProductCategory: "Chat",
+    clientAppType: "Web",
+    productEntryPoint: "ChatPanel",
+    deviceOS: "Windows",
+    deviceType: "Desktop",
+    clientPlatformVersion: "10",
+  });
  assert.equal(arg.conversationId, "44444444-4444-4444-4444-444444444444");
  assert.equal(arg.toolChoice, null);
-  assert.equal(arg.tone, "magic");
+  // The individual/consumer surface now sends "Magic" (capitalized), matching
+  // the enterprise tone literal — the #10718 lowercase "magic" is stale.
+  assert.equal(arg.tone, "Magic");
+  assert.equal(arg.isSbsSupported, true);
+  assert.equal(arg.renderReferencesBehindEOS, true);
+  assert.deepEqual(arg.extraExtensionParameters, {});
+  assert.deepEqual(arg.plugins, [{ Id: "BingWebSearch", Source: "BuiltIn" }]);
+  // Sent on every tier now, not gated to enterprise as #8971 described.
+  assert.equal(arg.disconnectBehavior, "continue");
 });

-test("#10718: the message object carries the recaptured rich shape", () => {
+test("2026-08-21: the message object carries the recaptured rich shape", () => {
  const arg = buildChatInvocation({
    text: "Say OK in one word.",
    traceId: "t",
@@ -93,7 +118,9 @@ test("#10718: the message object carries the recaptured rich shape", () => {
    "adaptiveCards",
    "attachments",
    "author",
+    "clientInfo",
    "clientPreferences",
+    "connectedFederatedConnections",
    "entityAnnotationTypes",
    "experienceType",
    "inputMethod",
@@ -107,26 +134,41 @@ test("#10718: the message object carries the recaptured rich shape", () => {
  assert.equal(message.messageType, "Chat");
  assert.equal(message.requestId, "r");
  assert.equal(message.experienceType, "Default");
-  assert.deepEqual(message.entityAnnotationTypes, ["People", "File", "Event", "Email", "TeamsMessage"]);
+  assert.deepEqual(message.entityAnnotationTypes, [
+    "People",
+    "File",
+    "Event",
+    "Email",
+    "TeamsMessage",
+  ]);
  assert.equal(message.attachments, null);
  assert.deepEqual(message.locationInfo, { timeZone: "UTC", timeZoneOffset: 0 });
+  assert.deepEqual(message.connectedFederatedConnections, ["dummyId"]);
+  // Same clientInfo object echoed inside message, per the capture.
+  assert.deepEqual(message.clientInfo, arg.clientInfo);
 });

-test("#10718: default tier lists are the recaptured 14-entry optionsSets / 6-entry allowedMessageTypes", () => {
+test("2026-08-21: default tier lists are the recaptured 34-entry optionsSets / 30-entry allowedMessageTypes", () => {
  const overrides = resolveChatInvocationOverrides(undefined);
-  assert.equal(overrides.optionsSets.length, 14);
-  assert.equal(overrides.allowedMessageTypes.length, 6);
-  assert.equal(overrides.tone, "magic");
-  // The pre-#10718 consumer/MSA flags are gone from the wire.
+  assert.equal(overrides.optionsSets.length, 34);
+  assert.equal(overrides.allowedMessageTypes.length, 30);
+  assert.equal(overrides.tone, "Magic");
+  assert.equal(overrides.disconnectBehavior, "continue");
  const optionSets = M365_DEFAULT_OPTION_SETS as readonly string[];
  const messageTypes = ALLOWED_MESSAGE_TYPES as readonly string[];
-  for (const stale of ["enable_msa_user", "pdnascan", "cwc_code_interpreter", "rich_responses"]) {
-    assert.ok(!optionSets.includes(stale), `${stale} must not be in the default option sets`);
+  // Entries the 2026-08-21 capture showed that the #10718 lists lacked.
+  for (const present of [
+    "cwc_code_interpreter",
+    "rich_responses",
+    "async_client_interaction",
+    "flux_v3_references",
+  ]) {
+    assert.ok(optionSets.includes(present), `${present} must be in the default option sets`);
  }
-  for (const stale of ["InternalSearchQuery", "GeneratedCode", "RenderCardRequest", "AdsQuery", "SemanticSerp", "GenerateContentQuery"]) {
-    assert.ok(!messageTypes.includes(stale), `${stale} must not be in allowedMessageTypes`);
+  for (const present of ["InternalSearchQuery", "GeneratedCode", "AuthError", "TriggerPlugin"]) {
+    assert.ok(messageTypes.includes(present), `${present} must be in allowedMessageTypes`);
  }
-  // Entries the capture showed and the old lists lacked.
+  // Entries the #10718 capture showed and this capture still confirms.
  assert.ok(optionSets.includes("cwcfluxgptv"));
  assert.ok(messageTypes.includes("EndOfRequest"));
 });
@@ -134,8 +176,7 @@ test("#10718: default tier lists are the recaptured 14-entry optionsSets / 6-ent
 // ── refresh_token helpers ──────────────────────────────────────────────────

 function fakeJwt(claims: Record<string, unknown>): string {
-  const b64 = (value: unknown) =>
-    Buffer.from(JSON.stringify(value)).toString("base64url");
+  const b64 = (value: unknown) => Buffer.from(JSON.stringify(value)).toString("base64url");
  return `${b64({ alg: "none" })}.${b64(claims)}.sig`;
 }

--- a/tests/unit/copilot-m365-tool-calls.test.ts
+++ b/tests/unit/copilot-m365-tool-calls.test.ts
@@ -182,7 +182,9 @@ test("buildChatInvocation defaults stay backward compatible", () => {
    requestId: "r",
    conversationId: "c",
  }).arguments[0] as Record<string, unknown>;
-  assert.deepEqual(arg.plugins, []);
+  // 2026-08-21 capture (#11069): BingWebSearch BuiltIn plugin is now the
+  // default on all tiers (was [] in the #10718 shape); toolChoice stays null.
+  assert.deepEqual(arg.plugins, [{ Id: "BingWebSearch", Source: "BuiltIn" }]);
  assert.equal(arg.toolChoice, null);
  assert.equal(arg.customInstructions, undefined);
 });
--- a/tests/unit/dashboard/m365-har-import.test.ts
+++ b/tests/unit/dashboard/m365-har-import.test.ts
@@ -0,0 +1,78 @@
+import test from "node:test";
+import assert from "node:assert/strict";
+
+import {
+  extractM365CredentialFromHar,
+  describeHarImportExpiry,
+  M365_CHATHUB_WS_PREFIX,
+} from "../../../src/shared/utils/m365HarImport.ts";
+
+function fakeJwt(exp: number): string {
+  const b64 = (value: unknown) => Buffer.from(JSON.stringify(value)).toString("base64url");
+  return `${b64({ alg: "none" })}.${b64({ exp })}.sig`;
+}
+
+function harWithUrl(url: string | null): string {
+  return JSON.stringify({
+    log: { entries: url ? [{ request: { url } }] : [] },
+  });
+}
+
+test("extracts access_token + chathubPath from a matching ChatHub WS entry", () => {
+  const exp = Math.floor(Date.now() / 1000) + 3600;
+  const token = fakeJwt(exp);
+  const url = `${M365_CHATHUB_WS_PREFIX}oid-123%40tenant-456?chatsessionid=abc&access_token=${token}`;
+  const result = extractM365CredentialFromHar(harWithUrl(url));
+
+  assert.equal(result.ok, true);
+  if (!result.ok) return;
+  assert.equal(result.chathubPath, "oid-123@tenant-456");
+  assert.equal(result.apiKey, `access_token=${token}; chathubPath=oid-123@tenant-456`);
+  assert.ok(result.expiresAt !== null && Math.abs(result.expiresAt - exp * 1000) < 1000);
+});
+
+test("picks the LAST matching entry when a HAR has multiple turns", () => {
+  const oldToken = fakeJwt(1000);
+  const freshToken = fakeJwt(9999999999);
+  const har = JSON.stringify({
+    log: {
+      entries: [
+        { request: { url: `${M365_CHATHUB_WS_PREFIX}u%40t?access_token=${oldToken}` } },
+        { request: { url: "https://unrelated.example/" } },
+        { request: { url: `${M365_CHATHUB_WS_PREFIX}u%40t?access_token=${freshToken}` } },
+      ],
+    },
+  });
+  const result = extractM365CredentialFromHar(har);
+  assert.equal(result.ok, true);
+  if (!result.ok) return;
+  assert.ok(result.apiKey.includes(freshToken));
+});
+
+test("returns notJson for malformed input", () => {
+  const result = extractM365CredentialFromHar("not json{{{");
+  assert.deepEqual(result, { ok: false, error: "notJson" });
+});
+
+test("returns noEntries when log.entries is missing/not an array", () => {
+  const result = extractM365CredentialFromHar(JSON.stringify({ log: {} }));
+  assert.deepEqual(result, { ok: false, error: "noEntries" });
+});
+
+test("returns noChathubUrl when no request matches the ChatHub prefix", () => {
+  const result = extractM365CredentialFromHar(harWithUrl("https://example.com/"));
+  assert.deepEqual(result, { ok: false, error: "noChathubUrl" });
+});
+
+test("returns missingFields when the URL matches but lacks access_token or chathubPath", () => {
+  const result = extractM365CredentialFromHar(harWithUrl(`${M365_CHATHUB_WS_PREFIX}`));
+  assert.equal(result.ok, false);
+});
+
+test("describeHarImportExpiry classifies ok/warn/bad/unknown", () => {
+  const now = Date.now();
+  assert.equal(describeHarImportExpiry(now + 30 * 60000, now).tone, "ok");
+  assert.equal(describeHarImportExpiry(now + 5 * 60000, now).tone, "warn");
+  assert.equal(describeHarImportExpiry(now - 60000, now).tone, "bad");
+  assert.equal(describeHarImportExpiry(null, now).tone, "unknown");
+});
--- a/tests/unit/executor-command-code.test.ts
+++ b/tests/unit/executor-command-code.test.ts
@@ -14,11 +14,15 @@ describe("CommandCodeExecutor", () => {
    assert.ok(executor);
  });

-  it("buildUrl returns a string", () => {
+  it("buildUrl targets the documented /provider/v1/chat/completions endpoint (#10265)", () => {
    const executor = new mod.CommandCodeExecutor();
    const url = executor.buildUrl();
    assert.ok(typeof url === "string");
-    assert.ok(url.includes("generate") && url.includes("commandcode"));
+    assert.ok(
+      url.includes("/provider/v1/chat/completions"),
+      `expected the documented provider API endpoint, got: ${url}`
+    );
+    assert.ok(url.includes("commandcode"));
  });

  it("execute throws when no API key", async () => {
@@ -57,7 +61,7 @@ describe("CommandCodeExecutor", () => {
    }
  });

-  it("assistant tool-call conversion always emits a valid required arguments field (#regression input[N] missing required field arguments)", async () => {
+  it("posts a flat OpenAI chat.completions body (no CLI envelope) to /provider/v1/chat/completions (#10265)", async () => {
    const calls: Array<{ url: string; init: RequestInit; body: unknown }> = [];
    const originalFetch = globalThis.fetch;
    globalThis.fetch = (async (url: string | URL | Request, init?: RequestInit) => {
@@ -70,182 +74,8 @@ describe("CommandCodeExecutor", () => {
    }) as typeof fetch;

    const executor = new mod.CommandCodeExecutor();
-    const pairedId = "call_paired";
-    const body = {
-      messages: [
-        { role: "user", content: "hi" },
-        {
-          role: "assistant",
-          content: "",
-          tool_calls: [
-            // Missing arguments entirely -> must still get a valid arguments field
-            { id: "call_missing", type: "function", function: { name: "lookup" } },
-            // Empty string arguments -> "{}"
-            {
-              id: "call_empty",
-              type: "function",
-              function: { name: "lookup", arguments: "" },
-            },
-            // Valid object arguments -> round-trips as JSON string
-            {
-              id: pairedId,
-              type: "function",
-              function: { name: "lookup", arguments: { q: "docs" } },
-            },
-            // Valid string arguments -> preserved as-is
-            {
-              id: "call_string",
-              type: "function",
-              function: { name: "lookup", arguments: '{"q":"string"}' },
-            },
-            // Invalid JSON string arguments -> defaults to "{}"
-            {
-              id: "call_invalid",
-              type: "function",
-              function: { name: "lookup", arguments: "{invalid-json" },
-            },
-            // Tool call without name -> defaults tool-result toolName to "unknown"
-            {
-              id: "call_unnamed",
-              type: "function",
-              function: { arguments: { q: "unnamed" } },
-            },
-          ],
-        },
-        { role: "tool", tool_call_id: "call_missing", content: "r1" },
-        { role: "tool", tool_call_id: "call_empty", content: "r2" },
-        { role: "tool", tool_call_id: pairedId, content: "r3" },
-        { role: "tool", tool_call_id: "call_string", content: "r4" },
-        { role: "tool", tool_call_id: "call_invalid", content: "r5" },
-        { role: "tool", tool_call_id: "call_unnamed", content: "r6" },
-      ],
-    };
-
-    try {
-      await executor.execute({
-        model: "test",
-        body,
-        stream: false,
-        credentials: { apiKey: "fake-key" },
-        signal: null,
-      });
-    } finally {
-      globalThis.fetch = originalFetch;
-    }
-
-    assert.equal(calls.length, 1, "exactly one upstream call");
-    const sentBody = calls[0].body as {
-      params: { messages: Array<{ role: string; content: unknown }> };
-    };
-    const assistant = sentBody.params.messages.find((m) => m.role === "assistant");
-    assert.ok(assistant, "assistant turn present");
-    const parts = assistant.content as Array<Record<string, unknown>>;
-    const toolCalls = parts.filter((p) => p.type === "tool-call");
-    assert.equal(toolCalls.length, 6, "all six paired tool calls converted");
-
-    for (const call of toolCalls) {
-      assert.equal(
-        typeof call.arguments,
-        "string",
-        `tool-call ${String(call.toolCallId)} must carry a string arguments field`
-      );
-      const parsed = JSON.parse(call.arguments as string);
-      assert.equal(typeof parsed, "object");
-      assert.ok(!Array.isArray(parsed), "arguments must parse to a JSON object");
-    }
-
-    const byId = new Map(toolCalls.map((c) => [String(c.toolCallId), c]));
-    assert.equal(byId.get("call_missing").arguments, "{}", "missing arguments -> empty object");
-    assert.equal(byId.get("call_empty").arguments, "{}", "empty string arguments -> empty object");
-    assert.equal(
-      byId.get(pairedId).arguments,
-      '{"q":"docs"}',
-      "object arguments round-trip as JSON string"
-    );
-    assert.equal(
-      byId.get("call_string").arguments,
-      '{"q":"string"}',
-      "valid string arguments preserved as-is"
-    );
-    assert.equal(
-      byId.get("call_invalid").arguments,
-      "{}",
-      "invalid JSON string arguments -> empty object"
-    );
-
-    const toolMsgs = sentBody.params.messages.filter((m) => m.role === "tool");
-    assert.equal(toolMsgs.length, 6, "all 6 tool result messages present");
-    const resultByName = new Map(
-      toolMsgs.map((m) => {
-        const p = (m.content as Array<Record<string, unknown>>)[0];
-        return [String(p.toolCallId), String(p.toolName)];
-      })
-    );
-    assert.equal(resultByName.get("call_missing"), "lookup");
-    assert.equal(
-      resultByName.get("call_unnamed"),
-      "unknown",
-      "unnamed call falls back to 'unknown'"
-    );
-
-    // /alpha/generate also requires `arguments` on tool-result parts; a
-    // missing one is rejected with `input[N] missing required field 'arguments'`
-    // (the index landing on the tool message). Echo the paired call's
-    // normalized arguments.
-    const resultById = new Map(
-      toolMsgs.map((m) => {
-        const p = (m.content as Array<Record<string, unknown>>)[0];
-        return [String(p.toolCallId), p];
-      })
-    );
-    assert.equal(resultById.size, 6, "each tool result maps to its call id");
-    for (const p of resultById.values()) {
-      assert.equal(
-        typeof p.arguments,
-        "string",
-        `tool-result ${String(p.toolCallId)} must carry a string arguments field`
-      );
-      const parsed = JSON.parse(p.arguments as string);
-      assert.equal(typeof parsed, "object");
-      assert.ok(!Array.isArray(parsed), "tool-result arguments must parse to a JSON object");
-    }
-    assert.equal(
-      resultById.get("call_missing").arguments,
-      "{}",
-      "tool-result echoes paired call's missing arguments as empty object"
-    );
-    assert.equal(
-      resultById.get(pairedId).arguments,
-      '{"q":"docs"}',
-      "tool-result echoes paired call's object arguments as JSON string"
-    );
-    assert.equal(
-      resultById.get("call_string").arguments,
-      '{"q":"string"}',
-      "tool-result echoes paired call's valid string arguments as-is"
-    );
-    assert.equal(
-      resultById.get("call_empty").arguments,
-      "{}",
-      "tool-result echoes paired call's empty arguments as empty object"
-    );
-    assert.equal(
-      resultById.get("call_invalid").arguments,
-      "{}",
-      "tool-result echoes paired call's invalid JSON arguments as empty object"
-    );
-  });
-
-  it("COMMAND_CODE_VERSION default constant is 1.15.1", () => {
-    assert.equal(mod.COMMAND_CODE_VERSION, "1.15.1");
-  });
-
-  it("renames tool names colliding with upstream built-ins on the wire and un-renames on the response (#regression input[N] missing required field arguments from a tool_search result)", async () => {
-    // Upstream /alpha/generate normalizes tool-call/result parts against its
-    // OWN built-in registry for matching names; `tool_search` collides and its
-    // result is rejected with `input[N] missing required field 'arguments'`.
-    // Verified live: renaming the pair to a non-colliding name passes.
    const body = {
+      model: "gpt-5.4",
      messages: [
        { role: "user", content: "hi" },
        {
@@ -253,29 +83,19 @@ describe("CommandCodeExecutor", () => {
          content: "",
          tool_calls: [
            {
-              id: "call_00_AAAAAAAAAAAAAAAAA",
+              id: "call_1",
              type: "function",
-              function: { name: "tool_search", arguments: '{"query":"x"}' },
-            },
-            {
-              id: "call_01_BBBBBBBBBBBBBBBBB",
-              type: "function",
-              function: { name: "lookup", arguments: '{"q":"1"}' },
+              function: { name: "lookup", arguments: '{"q":"docs"}' },
            },
+            // Missing arguments entirely stays missing — passthrough, no CLI
+            // envelope injection of a synthetic `arguments` field.
+            { id: "call_2", type: "function", function: { name: "search" } },
          ],
        },
-        { role: "tool", tool_call_id: "call_00_AAAAAAAAAAAAAAAAA", content: "r1" },
-        { role: "tool", tool_call_id: "call_01_BBBBBBBBBBBBBBBBB", content: "r2" },
+        { role: "tool", tool_call_id: "call_1", content: "r1" },
+        { role: "tool", tool_call_id: "call_2", content: "r2" },
      ],
      tools: [
-        {
-          type: "function",
-          function: {
-            name: "tool_search",
-            description: "search tools",
-            parameters: { type: "object", properties: { query: { type: "string" } } },
-          },
-        },
        {
          type: "function",
          function: {
@@ -287,31 +107,9 @@ describe("CommandCodeExecutor", () => {
      ],
    };

-    const calls: Array<{ url: string; init: RequestInit; body: unknown }> = [];
-    const originalFetch = globalThis.fetch;
-    // execute() makes a single upstream fetch; capture the wire request and
-    // return a stream with a tool-call event using the renamed wire name.
-    globalThis.fetch = (async (url: string | URL | Request, init?: RequestInit) => {
-      calls.push({
-        url: String(url),
-        init: init || {},
-        body: JSON.parse(String((init as RequestInit | undefined)?.body)),
-      });
-      const streamBody =
-        'data: {"type":"tool-call","toolCallId":"c1","toolName":"omniroute_tool_search","input":{"query":"x"}}\n\n' +
-        'data: {"type":"finish","finishReason":"tool_use"}\n\n' +
-        "data: [DONE]\n\n";
-      return new Response(streamBody, {
-        status: 200,
-        headers: { "Content-Type": "text/event-stream" },
-      });
-    }) as typeof fetch;
-
-    const executor = new mod.CommandCodeExecutor();
-    let result: { response: Response } | null = null;
    try {
-      result = await executor.execute({
-        model: "test",
+      await executor.execute({
+        model: "gpt-5.4",
        body,
        stream: false,
        credentials: { apiKey: "fake-key" },
@@ -321,46 +119,85 @@ describe("CommandCodeExecutor", () => {
      globalThis.fetch = originalFetch;
    }

-    const sentBody = calls[0].body as {
-      params: {
-        messages: Array<{ role: string; content: unknown }>;
-        tools: Array<{ name: string }>;
-      };
-    };
-    const toolDefNames = sentBody.params.tools.map((t) => t.name);
+    assert.equal(calls.length, 1, "exactly one upstream call");
    assert.ok(
-      toolDefNames.includes("omniroute_tool_search"),
-      "colliding tool def renamed on the wire"
+      calls[0].url.includes("/provider/v1/chat/completions"),
+      `expected documented provider endpoint, got: ${calls[0].url}`
    );
-    assert.ok(toolDefNames.includes("lookup"), "non-colliding tool def untouched");
-
-    const assistant = sentBody.params.messages.find((m) => m.role === "assistant");
-    const toolCallParts = (assistant?.content as Array<Record<string, unknown>>).filter(
-      (p) => p.type === "tool-call"
+    const sent = calls[0].body as Record<string, unknown>;
+    // No CLI envelope.
+    assert.equal(sent.config, undefined, "CLI envelope `config` must not be sent");
+    assert.equal(sent.params, undefined, "CLI envelope `params` wrapper must not be sent");
+    assert.equal(sent.model, "gpt-5.4", "flat OpenAI model at top level");
+    assert.equal((sent.messages as Array<{ role: string }>)[0].role, "user");
+    // Assistant tool_calls pass through unchanged (no CLI tool-call/tool-result parts).
+    const assistant = (sent.messages as Array<Record<string, unknown>>).find(
+      (m) => m.role === "assistant"
    );
-    const toolSearchCall = toolCallParts.find((p) => p.toolName === "omniroute_tool_search");
-    assert.ok(toolSearchCall, "assistant tool-call part renamed on the wire");
-    const lookupCall = toolCallParts.find((p) => p.toolName === "lookup");
-    assert.ok(lookupCall, "non-colliding tool-call part untouched");
-
-    const toolMsgs = sentBody.params.messages.filter((m) => m.role === "tool");
-    const toolSearchResult = toolMsgs.find(
-      (m) => (m.content as Array<Record<string, unknown>>)[0]?.toolName === "omniroute_tool_search"
+    assert.ok(assistant, "assistant turn present");
+    const toolCalls = assistant?.tool_calls as Array<{
+      id: string;
+      function: { name: string; arguments?: string };
+    }>;
+    assert.equal(toolCalls.length, 2, "both tool calls pass through untouched");
+    assert.equal(toolCalls[0].function.name, "lookup");
+    assert.equal(toolCalls[0].function.arguments, '{"q":"docs"}');
+    assert.equal(toolCalls[1].function.arguments, undefined, "missing arguments stays missing (no injection)");
+    // Tool role message (OpenAI flat) preserved.
+    const toolMsg = (sent.messages as Array<Record<string, unknown>>).find(
+      (m) => m.role === "tool"
    );
-    assert.ok(toolSearchResult, "tool-result part renamed on the wire");
-
-    // Response path: upstream emits the renamed wire name; the client must get
-    // its original name back.
-    assert.ok(result, "execute returned a response");
-    const json = (await result.response.json()) as {
-      choices: Array<{ message: { tool_calls?: Array<{ function: { name: string } }> } }>;
-    };
-    const toolCalls = json.choices[0].message.tool_calls ?? [];
-    assert.equal(toolCalls.length, 1, "one tool call translated");
+    assert.equal(toolMsg?.tool_call_id, "call_1");
    assert.equal(
-      toolCalls[0].function.name,
-      "tool_search",
-      "renamed wire name un-renamed for the client"
+      (sent.tools as Array<{ function: { name: string } }>)[0].function.name,
+      "lookup",
+      "tool definitions pass through in OpenAI shape (no rename)"
    );
  });
-});
+
+  it("passes through the upstream OpenAI response and drops CLI-impersonation headers (#10265)", async () => {
+    const calls: Array<{ url: string; init: RequestInit }> = [];
+    const originalFetch = globalThis.fetch;
+    globalThis.fetch = (async (url: string | URL | Request, init?: RequestInit) => {
+      calls.push({ url: String(url), init: init || {} });
+      const chunk =
+        'data: {"id":"c1","object":"chat.completion.chunk","model":"gpt-5.4",' +
+        '"choices":[{"index":0,"delta":{"content":"hi"}}]}\n\n' +
+        'data: {"id":"c1","object":"chat.completion.chunk","model":"gpt-5.4",' +
+        '"choices":[{"index":0,"delta":{},"finish_reason":"stop"}],"usage":' +
+        '{"prompt_tokens":2,"completion_tokens":1,"total_tokens":3}}\n\n' +
+        "data: [DONE]\n\n";
+      return new Response(chunk, {
+        status: 200,
+        headers: { "Content-Type": "text/event-stream" },
+      });
+    }) as typeof fetch;
+
+    const executor = new mod.CommandCodeExecutor();
+    let result: { response: Response; headers: Record<string, string> } | null = null;
+    try {
+      result = await executor.execute({
+        model: "gpt-5.4",
+        body: { messages: [{ role: "user", content: "hi" }] },
+        stream: true,
+        credentials: { apiKey: "fake-key" },
+        signal: null,
+      });
+    } finally {
+      globalThis.fetch = originalFetch;
+    }
+
+    assert.ok(result, "execute returned a result");
+    const headers = result.headers;
+    assert.equal(headers["x-command-code-version"], undefined, "CLI-impersonation header dropped");
+    assert.equal(headers["x-cli-environment"], undefined, "CLI-impersonation header dropped");
+    assert.equal(headers.Authorization, "Bearer fake-key");
+
+    // The upstream OpenAI SSE passes through untouched (no CLI re-parsing).
+    const text = await result.response.text();
+    assert.ok(text.includes("chat.completion.chunk"), "OpenAI-format SSE passed through");
+    assert.ok(text.includes('"content":"hi"'), "delta content preserved");
+    assert.ok(text.includes("[DONE]"), "stream terminator preserved");
+    assert.ok(text.includes('"prompt_tokens":2'), "OpenAI usage block passed through");
+  });
+});
--- a/tests/unit/m365-bizchat-frames-4042.test.ts
+++ b/tests/unit/m365-bizchat-frames-4042.test.ts
@@ -131,7 +131,22 @@ test("buildChatInvocation produces a type:4 chat invocation carrying the user te
  assert.equal(arg.sessionId, "session-id");
  assert.equal(arg.conversationId, "conversation-id");
  assert.equal(arg.productThreadType, "Office");
-  assert.deepEqual(arg.clientInfo, { clientAppName: "Office", clientPlatform: "mcmcopilot-web" });
+  // 2026-08-21 capture (#11069): clientInfo gained 8 keys (clientEntrypoint,
+  // clientSessionId, ProductCategory, clientAppType, productEntryPoint,
+  // deviceOS, deviceType, clientPlatformVersion). The #10718 base shape only
+  // carried clientAppName + clientPlatform.
+  assert.deepEqual(arg.clientInfo, {
+    clientAppName: "Office",
+    clientPlatform: "mcmcopilot-web",
+    clientEntrypoint: "mcmcopilot-officeweb",
+    clientSessionId: "session-id",
+    ProductCategory: "Chat",
+    clientAppType: "Web",
+    productEntryPoint: "ChatPanel",
+    deviceOS: "Windows",
+    deviceType: "Desktop",
+    clientPlatformVersion: "10",
+  });
  assert.equal(arg.isStartOfSession, true);
  assert.ok(Array.isArray(arg.optionsSets));
  assert.ok((arg.optionsSets as string[]).includes("enable_gg_gpt"));
--- a/tests/unit/m365-tone-model-variants.test.ts
+++ b/tests/unit/m365-tone-model-variants.test.ts
@@ -30,10 +30,11 @@ test("model-driven tone overrides the tier default; bare id keeps the tier tone"
  const enterprise = resolveChatInvocationOverrides("enterprise");
  const individual = resolveChatInvocationOverrides(undefined);

-  // enterprise tier default tone is Magic; individual/EDU sends "magic" (#10718
-  // recapture — the old "" default was part of the silently-dropped shape)
+  // enterprise tier default tone is Magic; individual/EDU now also sends
+  // "Magic" (capitalized) — the 2026-08-21 capture (#11069) showed lowercase
+  // "magic" is part of the shape that gets silently dropped (ping-only socket).
  assert.equal(enterprise.tone, "Magic");
-  assert.equal(individual.tone, "magic");
+  assert.equal(individual.tone, "Magic");

  // precedence: resolveToneForModel(model) ?? overrides.tone  (mirrors the executor wiring)
  const toneFor = (model: string | undefined, tierTone: string) =>
@@ -43,9 +44,9 @@ test("model-driven tone overrides the tier default; bare id keeps the tier tone"
  assert.equal(toneFor("copilot-m365-claude-opus", enterprise.tone), "Claude_Opus");
  assert.equal(toneFor("copilot-m365-claude-opus", individual.tone), "Claude_Opus");

-  // the bare id keeps whatever the tier resolved
+  // the bare id keeps whatever the tier resolved (both "Magic" post-#11069)
  assert.equal(toneFor("copilot-m365", enterprise.tone), "Magic");
-  assert.equal(toneFor("copilot-m365", individual.tone), "magic");
+  assert.equal(toneFor("copilot-m365", individual.tone), "Magic");
 });

 test("registry exposes the bare id (first) plus every tone variant", () => {
--- a/tests/unit/provider-models-target-format-scoping.test.ts
+++ b/tests/unit/provider-models-target-format-scoping.test.ts
@@ -8,12 +8,12 @@ import { resolveChatCoreTargetFormat } from "../../open-sse/handlers/chatCore/ta
 // ghe-copilot catalog. getModelTargetFormat falls back to getGlobalModel() when
 // the provider's own catalog lacks the model id, importing the DECLARING
 // provider's endpoint semantics into every other provider serving the same id.
-// command-code's chat-shaped /alpha/generate executor then received a
+// command-code's chat-shaped executor then received a
 // Responses-format body (input, not messages) and shipped `messages: []`
 // upstream — upstream rejected with "Invalid prompt: messages must not be empty"
 // (502). Model-level targetFormat is provider-scoped: it must not leak.
 test("model-level targetFormat does not leak across provider catalogs", () => {
-  // command-code serves gpt-5.6-luna over its chat-shaped /alpha/generate endpoint
+  // command-code serves gpt-5.6-luna over its chat-shaped provider endpoint
  assert.equal(getModelTargetFormat("cmd", "gpt-5.6-luna"), null);
  // raw provider id form behaves identically (alias resolution)
  assert.equal(getModelTargetFormat("command-code", "gpt-5.6-luna"), null);
--- a/tests/unit/provider-validation-specialty.test.ts
+++ b/tests/unit/provider-validation-specialty.test.ts
@@ -16,7 +16,6 @@ const { __setTlsFetchOverrideForTesting: __setPplxTlsFetchOverride } =
 const { __setTlsFetchOverrideForTesting: __setGrokTlsFetchOverride } =
  await import("../../open-sse/services/grokTlsClient.ts");

-const { COMMAND_CODE_VERSION } = await import("../../open-sse/executors/commandCode.ts");

 const originalFetch = globalThis.fetch;

@@ -216,11 +215,11 @@ test("specialty provider validators cover Deepgram, AssemblyAI, ElevenLabs and I

 test("validateCommandCodeProvider ignores caller baseUrl and chatPath overrides", async () => {
  globalThis.fetch = async (url, init = {}) => {
-    assert.equal(String(url), "https://api.commandcode.ai/alpha/generate");
+    assert.equal(String(url), "https://api.commandcode.ai/provider/v1/chat/completions");
    const headers = init.headers as Record<string, string>;
    assert.equal(headers.Authorization, "Bearer cc-key");
    const body = JSON.parse(String(init.body));
-    assert.equal(body.params.model, "command-code-validation-model");
+    assert.equal(body.model, "command-code-validation-model");
    return new Response(JSON.stringify({ ok: true }), { status: 200 });
  };

@@ -239,7 +238,7 @@ test("validateCommandCodeProvider ignores caller baseUrl and chatPath overrides"
 test("validateCommandCodeProvider defaults probe model to DeepSeek flash", async () => {
  globalThis.fetch = async (_url, init = {}) => {
    const body = JSON.parse(String(init.body));
-    assert.equal(body.params.model, "deepseek/deepseek-v4-flash");
+    assert.equal(body.model, "deepseek/deepseek-v4-flash");
    return new Response("", { status: 400 });
  };

@@ -2285,7 +2284,7 @@ test("specialty validator rejects invalid Runway credentials", async () => {
  assert.equal(runway.error, "Invalid API key");
 });

-test("validateCommandCodeProvider sends Command Code probe URL, headers, and wrapper body", async () => {
+test("validateCommandCodeProvider sends Command Code probe URL, headers, and flat OpenAI body", async () => {
  const calls: Array<{
    url: string;
    method?: string;
@@ -2309,22 +2308,21 @@ test("validateCommandCodeProvider sends Command Code probe URL, headers, and wra

  assert.deepEqual(result, { valid: true, error: null });
  assert.equal(calls.length, 1);
-  assert.equal(calls[0].url, "https://api.commandcode.ai/alpha/generate");
+  // Probe targets the documented /provider/v1/chat/completions endpoint, not
+  // the CLI-only /alpha/generate (#10265).
+  assert.equal(calls[0].url, "https://api.commandcode.ai/provider/v1/chat/completions");
  assert.equal(calls[0].method, "POST");
  assert.equal(calls[0].headers.Authorization, "Bearer cc_test_key");
  assert.equal(calls[0].headers["Content-Type"], "application/json");
-  assert.equal(calls[0].headers["x-command-code-version"], COMMAND_CODE_VERSION);
-  assert.equal(calls[0].headers["x-cli-environment"], "external");
-  assert.equal(calls[0].headers["x-project-slug"], "pi-cc");
-  assert.equal(calls[0].headers["x-taste-learning"], "false");
-  assert.equal(calls[0].headers["x-co-flag"], "false");
-  assert.equal(typeof calls[0].headers["x-session-id"], "string");
-  assert.equal(calls[0].body.config.environment, "external");
-  assert.equal(calls[0].body.permissionMode, "standard");
-  assert.equal(calls[0].body.skills, "");
-  assert.equal(calls[0].body.params.model, "gpt-5.4-mini");
-  assert.equal(calls[0].body.params.stream, true);
-  assert.equal(calls[0].body.params.max_tokens, 1);
+  // No CLI-impersonation headers.
+  assert.equal(calls[0].headers["x-command-code-version"], undefined);
+  assert.equal(calls[0].headers["x-cli-environment"], undefined);
+  assert.equal(calls[0].headers["x-project-slug"], undefined);
+  // Flat OpenAI chat.completions body (no CLI wrapper).
+  assert.equal(calls[0].body.params, undefined, "CLI envelope params wrapper must not be sent");
+  assert.equal(calls[0].body.model, "gpt-5.4-mini");
+  assert.equal(calls[0].body.stream, true);
+  assert.equal(calls[0].body.max_tokens, 1);
 });

 for (const status of [400, 422, 429]) {
--- a/tests/unit/responses-handler.test.ts
+++ b/tests/unit/responses-handler.test.ts
@@ -10,7 +10,6 @@ process.env.DATA_DIR = TEST_DATA_DIR;

 const core = await import("../../src/lib/db/core.ts");
 const { handleResponsesCore } = await import("../../open-sse/handlers/responsesHandler.ts");
-const { COMMAND_CODE_VERSION } = await import("../../open-sse/executors/commandCode.ts");

 const originalFetch = globalThis.fetch;

@@ -354,26 +353,31 @@ test("handleResponsesCore transforms Command Code executor SSE through Responses
      input: "hello command code",
    },
    responseFactory() {
+      // /provider/v1/chat/completions returns standard OpenAI SSE (#10265).
+      const chunk = (delta: Record<string, unknown>) =>
+        `data: ${JSON.stringify({
+          id: "c1",
+          object: "chat.completion.chunk",
+          model: "gpt-5.4-mini",
+          choices: [{ index: 0, delta }],
+        })}\n\n`;
      return new Response(
        [
-          `data: ${JSON.stringify({ type: "text-delta", text: "command" })}`,
-          "",
-          `data: ${JSON.stringify({ type: "reasoning-delta", text: "thinking" })}`,
-          "",
-          `data: ${JSON.stringify({ type: "finish", finishReason: "stop" })}`,
-          "",
-        ].join("\n"),
-        { status: 200, headers: { "Content-Type": "application/x-ndjson" } }
+          chunk({ role: "assistant" }),
+          chunk({ content: "command" }),
+          chunk({}),
+        ].join("") + "data: [DONE]\n\n",
+        { status: 200, headers: { "Content-Type": "text/event-stream" } }
      );
    },
  });

  assert.equal(result.success, true);
-  assert.equal(call.url, "https://api.commandcode.ai/alpha/generate");
+  assert.equal(call.url, "https://api.commandcode.ai/provider/v1/chat/completions");
  assert.equal(call.headers.Authorization, "Bearer cc_test_key");
-  assert.equal(call.headers["x-command-code-version"], COMMAND_CODE_VERSION);
-  assert.equal(call.body.params.model, "gpt-5.4-mini");
-  assert.equal(call.body.params.stream, true);
+  assert.equal(call.headers["x-command-code-version"], undefined);
+  assert.equal(call.body.model, "gpt-5.4-mini");
+  assert.equal(call.body.stream, true);

  const sse = await result.response.text();
  assert.match(sse, /event: response\.created/);
Author	SHA1	Message	Date
Xiangzhe	cf1d564e0a	chore(skills): regenerate omni-webhooks after dependency changes (#11095 )	2026-08-22 14:13:57 -03:00
Xiangzhe	7fb921e28e	fix(install): make ONNX chain optional for Android/Termux installs (#11095 )	2026-08-22 13:57:59 -03:00
N123 Project	6cd4d38e21	fix(m365): BizChat invocation shape drift + HAR-import UX + Antigravity alias note (#11069 ) ⭐5 — M365 Copilot (BizChat) individual/consumer path — 3 itens: (1) forma de invocação do #10718 derivou de novo (2026-08-21 capture): optionsSets 14→34, allowedMessageTypes 6→30, tone "magic"→"Magic", plugins []→[{BingWebSearch}], disconnectBehavior em todos os tiers, +8 keys de clientInfo; verificado contra conta real com round-trip WebSocket (ping-then-close → resposta real). (2) Aviso sobre o alias Antigravity gemini-3.1-pro-high ainda não publicado (3.8.49 pré-data). (3) Botão "Import .har file" no modal de credencial M365. Conflito resolvido em copilot-m365-frames.ts (board vs release tip): mantive o forwarding de opts.plugins/toolChoice/customInstructions do HEAD com os NOVOS defaults da captura (BingWebSearch builtin, tone "Magic"). Alinhei 3 testes pré-existentes que afirmavam o contrato antigo (m365-bizchat-frames-4042 clientInfo, m365-tone-model-variants tone, copilot-m365-tool-calls plugins) — propagação de contrato, não mascaramento. Rebaselinei AddApiKeyModal 1073→1080 (crescimento próprio da parte 3, ~Har import button) com anotação. Validação: typecheck limpo, 142/142 testes m365/copilot verdes, changelog-integrit/file-size/eslint OK.	2026-08-21 22:32:19 -03:00
Diego Rodrigues de Sa e Souza	b6412c6fed	fix(command-code): use the documented /provider/v1 chat endpoint (#10265 ) (#11072 ) ⭐5 — Fecha #10265: chat do command-code migra do endpoint CLI-only /alpha/generate (version-gated + proxy-blocked para callers externos) para o documentado /provider/v1/chat/completions (OpenAI format). Removido o envelope CLI reverse-engineered (config/memory/taste/skills + headers CLI-impersonation), substituído por passthrough OpenAI plano com normalização de model id vendor-prefixed (#10809), clamp de max_tokens (#5166), sanitização de reasoning_effort. commandCode.ts 1037→171 linhas. Validado no worktree board sobre tip: typecheck:core limpo; 175/175 testes focados (command-code executor/vision/usage/maxtokens/user-array/validation-specialty/responses-handler/provider-models-scoping); changelog-integrity/file-size/complexity/cognitive todos OK. TDD RED→GREEN documentado.	2026-08-21 22:16:45 -03:00
				`@@ -0,0 +1 @@`
				`- fix(command-code): route chat to the documented /provider/v1/chat/completions endpoint instead of the CLI-only /alpha/generate, which Command Code gates/blocks for external callers (#10265)`
				`@@ -0,0 +1 @@`
				`- fix(install): make the ONNX dependency chain optional so Termux/Android installs succeed again (#11095)`