customizations-in-the-agent-host

bởi microsoft

Kiến trúc và các bài học gỡ lỗi khó nhằn cho việc hỗ trợ tùy chỉnh (plugin, MCP server, agent, skill, hướng dẫn) trong agent host. Sử dụng khi…

npx skills add https://github.com/microsoft/vscode --skill customizations-in-the-agent-host

Customizations in the Agent Host

Customizations are the plugins, MCP servers, agents, skills and instructions that an agent-host session exposes to a model. This skill covers enablement: how a customization's on/off state is decided, stored, published and enforced.

Read this before changing anything in:

  • src/vs/platform/agentHost/node/agentHostCustomizationEnablementService.ts
  • src/vs/platform/agentHost/node/shared/customizationEnablementGate.ts
  • src/vs/platform/agentHost/common/customizationEnablement.ts
  • the per-provider customization paths under node/copilot/, node/claude/, node/codex/

The mental model

Enablement is a list of scoped decisions, not a boolean

A plugin or MCP server carries enablement?: CustomizationEnablement[] — an ordered list of decisions, most specific first:

KindOwnerMeaning
Sessionagent hostthis session only
Workspaceagent hostthis working directory (carries uri)
Globalclient, unless the host discovered iteverywhere

isCustomizationEnabled() picks the first entry; absent entries inherit from the next scope out, and an absent list means enabled. Never re-add a parallel enabled boolean to plugins or MCP servers — a second source of truth is what this design exists to remove. (DirectoryCustomization legitimately keeps a plain enabled; consumers derive its state through isCustomizationEnabled() all the same.)

Who owns Global

The host owns Global for anything it discovered, even when that thing lives inside a plugin the client forwarded. The client owns Global only for what it actually bundled. The discriminator is childEnablementnot clientId. A plugin whose childEnablement record has a key for a child MCP server means "the client bundled this child and owns its global decision"; anything else the host discovered by parsing the plugin directory.

The overlay model

This is the single most important invariant, and getting it wrong caused a self-defeating write that silently erased user decisions:

The client's value is the base. Host decisions layer on top. "Store only what differs" compares against that base — never against a hardcoded true.

Concretely, in agentHostCustomizationEnablementService.ts:

  • _clientGlobalEnablement is the in-memory base, per session. _setClientGlobal writes here and must never touch _persistent.
  • _persistent holds only host-owned decisions that differ from the base.
  • Inheritance runs host → base → DEFAULT_CUSTOMIZATION_ENABLED (_globalEnablement, _workspaceInheritedEnablement).
  • _setPersistentDecision deletes a decision that matches its inherited value, so the UI can restore inheritance without a third "Inherit" action.

Keeping these separate is what stops this loop: replaceEnablement → synchronous republish → client's stale [{global,true}]_setClientGlobal → "store only what differs" deletes the entry just written.

Identity must not depend on tree position

An MCP server's storage key is derived from owningPluginUri, never from where it happened to appear in the published tree. The same server discovered top-level (mcpServers#azure) and nested under a plugin (<pluginURI>#mcp=azure) must produce one key, or a decision written in one session is invisible to the next. targetForMcpServer() and withOwningPluginUri() in the gate are the only supported way to build that target.

Pending means fail closed — but pending must be transient

resolve() returns {kind:'pending'} when the session isn't initialized yet (reason: 'session') or the working directory isn't known (reason: 'workingDirectory'). resolveCustomizationEnablement() records those in pendingCustomizationIds, and isCustomizationSdkEligible() makes SDK-boundary consumers fail closed.

Two rules follow:

  1. Never "fix" a pending customization by defaulting it to enabled.
  2. Never cache the pending verdict. Pending is a temporary state; recompute it, and republish when the enablement service settles. A cached pendingEnablement: true silently drops a plugin forever.

A pending child MCP server also marks its containing plugin pending, because a plugin directory can cause the SDK to discover the child on its own.

The rule that prevents most bugs

Every publication, emit and SDK-handoff path must go through resolution.

Three separate production bugs on this feature came from paths that skipped it (four sites in copilotAgent, one in mcpCustomizationController, plus the post-startup reconcile). When you add a new place that publishes a customization or hands one to a provider, route it through resolveCustomizationEnablement() — do not read a stored snapshot.

The corollary: state computed once must be rebuilt when its inputs change. Merged per-client enablement maps that are only appended to will leak stale decisions onto a newly parsed plugin with the same URI.

Enforcement is per-provider, and they differ

ProviderLaunchMid-session
CopilotdisabledMcpServers on SessionConfig at open/resumerpc.mcp.enable / rpc.mcp.disable
Claudesession options at materializelive reconcile
CodexMCP servers supplied only at thread/start / resumenot possible — documented limitation

Two traps here:

  • The post-startup reconcile runs ~800 ms after launch. Anything gated only there can still start, and request authentication, during the launch window. Gate at launch too.
  • The SDK's not_configured status must not be translated as enabled: true. That bug made a correct "skip this disabled server" decision look like "enable it".

Client-provided MCP servers (ones the client bundled rather than the host discovering) must route their global Disable to the client, not the host — the host correctly refuses to store a client-owned global, so routing it to the host is a silent no-op. isClientBundled is the discriminator; see getBuiltinMcpServerEnablementActions in mcpListWidget.ts.

Debugging playbook

Rank your signals. They disagree, and the model lies:

  1. Primary — the published customization object and the host's stored decision.
  2. Definitive for tool exposuretool_search override … clientMatched=[…] in Agent Host.log.
  3. Secondary — the customizations UI.
  4. Corroborating only — what the model says. A model answered "yes, I have azure tools" while the server was stopped. Never conclude from this alone.

Reproduction recipe that exercises the real failure surface: start a session, send a message to boot everything, disable the thing, then ask a new session whether it has those tools. Cross-session persistence is where identity and overlay bugs surface.

Use the agent-host-logs skill for exported bundles, and the launch skill to drive a real instance.

Known false alarms:

  • The azure plugin contributes 24 azure-* skills in addition to its MCP server. Disabling the server correctly leaves the skills enabled.
  • sleepy is a deliberately broken test server that cannot initialize — useless as a positive control.

Gotchas that cost real time

  • CustomizationType and CustomizationEnablementKind are const enums. A type-only import compiles and lints cleanly, then crashes at runtime. Import them as values.
  • The unit-test runner uses pre-compiled output and does not fail on type errors. Always run npm run typecheck-client separately.
  • Integration-test servers start with --quiet. Wiring that you hoist out of if (!options.quiet) in agentHostServerMain.ts silently changes their behaviour — hoisting setWorktreeIsolation made quiet servers resolve isolation: 'folder' and broke seven unrelated tests.
  • Protocol files under common/state/protocol/ carry a "DO NOT EDIT — auto-generated" banner. Changes there need a matching backport to the agent-host-protocol repo.
  • Electron runner setup aborts (EEXIST/EPERM/ENOENT/ReactiveObjC) are environmental. Recover with rm -rf .build/electron && npm run electron.

Validation

npm run typecheck-client
./scripts/test.sh --grep "customizationEnablement"
./scripts/test-integration.sh --runGlob "**/agentHost/**/{sessionConfig,toolApproval,codexCustomizations}.integrationTest.js"
./scripts/test-integration.sh --run src/vs/platform/agentHost/test/node/claudeAgent.integrationTest.ts

Tests that construct an agent session directly need IAgentHostCustomizationEnablementService in their service collection; use createNoopCustomizationEnablementService() from src/vs/platform/agentHost/test/node/testCustomizationEnablementService.ts.

Thêm skills từ microsoft

oss-growth
microsoft
Cá tính tăng trưởng OSS
agent-framework-azure-ai-py
microsoft
Xây dựng các tác nhân Azure AI Foundry bằng SDK Python của Microsoft Agent Framework (agent-framework-azure-ai). Sử dụng khi tạo các tác nhân bền vững với AzureAIAgentsProvider, sử dụng các công cụ được lưu trữ (trình thông dịch mã, tìm kiếm tệp, tìm kiếm web), tích hợp máy chủ MCP, quản lý chuỗi hội thoại hoặc triển khai phản hồi phát trực tuyến. Bao gồm các công cụ hàm, đầu ra có cấu trúc và các tác nhân đa công cụ.
development
airunway-aks-setup
microsoft
Thiết lập AI Runway trên AKS — từ cụm trống đến mô hình đang chạy. Bao gồm xác minh cụm, cài đặt controller, đánh giá GPU, thiết lập nhà cung cấp và triển khai đầu tiên. KHI NÀO: "thiết lập AI Runway", "onboard cụm AKS", "cài đặt AI Runway", "thiết lập airunway", "triển khai mô hình lên AKS", "suy luận GPU trên AKS", "thiết lập KAITO trên AKS", "chạy LLM trên AKS", "vLLM trên AKS", "thiết lập phục vụ mô hình trên AKS", "AI Runway controller".
devops
appinsights-instrumentation
microsoft
Hướng dẫn để instrument các ứng dụng web với Azure Application Insights. Cung cấp các mẫu telemetry, thiết lập SDK, và tài liệu tham khảo cấu hình. KHI NÀO: cách instrument ứng dụng, App Insights SDK, các mẫu telemetry, App Insights là gì, hướng dẫn Application Insights, ví dụ instrumentation, các phương pháp tốt nhất APM.
devops
applicationinsights-web-ts
microsoft
Instrument các ứng dụng trình duyệt/web bằng SDK JavaScript Application Insights (@microsoft/applicationinsights-web). Dùng cho Real User Monitoring (RUM) — lượt xem trang, nhấp chuột, phụ thuộc AJAX/fetch, ngoại lệ, sự kiện tùy chỉnh và dấu vết tác nhân GenAI phía trình duyệt tương quan với dấu vết OpenTelemetry phía backend. Bao gồm thiết lập SDK Loader Script và npm, tiện ích mở rộng framework (React, React Native, Angular), Click Analytics, trình khởi tạo telemetry và quy ước ngữ nghĩa OTel GenAI cho các span tác nhân/công cụ/mô hình phát ra từ trình duyệt.
devops
azure-ai-anomalydetector-java
microsoft
Xây dựng ứng dụng phát hiện bất thường với Azure AI Anomaly Detector SDK cho Java. Sử dụng khi triển khai phát hiện bất thường đơn biến/đa biến, phân tích chuỗi thời gian hoặc giám sát hỗ trợ AI.
development
azure-ai-language-conversations-py
microsoft
Triển khai Conversational Language Understanding (CLU) bằng SDK Python azure-ai-language-conversations. Sử dụng khi làm việc với ConversationAnalysisClient để phân tích ý định và thực thể trong hội thoại, xây dựng tính năng NLP, hoặc tích hợp hiểu ngôn ngữ vào ứng dụng.
development
azure-ai-ml-py
microsoft
Azure Machine Learning SDK v2 cho Python. Dùng cho không gian làm việc ML, công việc, mô hình, tập dữ liệu, tính toán và quy trình. Kích hoạt: "azure-ai-ml", "MLClient", "không gian làm việc", "đăng ký mô hình", "công việc đào tạo", "tập dữ liệu".
development