Skill delivery

Overview#

Skill delivery makes published skills from one or more Learning containers available to an agent without another CLI download or process restart. A framework adapter adds an alphabetical catalog and two tools.

The model decides when to load and follow a skill.

Developer instructions retain precedence. Skills cannot override the agent's role, safety rules, tool restrictions, or application policy.

Before you start#

The Intelligence SDK adapters load published Skills into your agent automatically. They add a catalog and tools that read skill instructions during a run.

Complete these steps before you run an example:

1. Create a Learning container in your Intelligence project.
2. Publish a Skill and enable Skill delivery in the container's Skills tab.
3. Configure the agent with your project key and Learning container ID.

For one container, you can set the container and revision in code or with environment variables. The examples below show code parameters. To use environment variables instead, omit the corresponding parameters from the code. You can mix both approaches: an explicit code value overrides its environment variable.

For example, this environment configuration supplies the project key and container:

export CPK_INTELLIGENCE_API_KEY="your-project-key"
export CPK_INTELLIGENCE_LEARNING_CONTAINER_ID="support-learning"
# For self-hosted Intelligence, also set INTELLIGENCE_API_URL.

Set your model provider's credentials separately. Replace model placeholders with your configured model and install its provider package when required by your framework.

To follow the latest published Skills, omit the revision parameter in code and leave CPK_INTELLIGENCE_SKILLS_REVISION unset. See container configuration for exact revision pins.

Skill delivery does not send native agent runs to Automatic Learning. To collect new evidence, also assign your Runtime's Threads to a container.

Choose an adapter#

Attach an adapter to the agents that need skills. Adapters do not inspect or rebuild arbitrary graphs or agent hierarchies. Subagent behavior follows the selected framework.

TypeScript uses the canonical client from @copilotkit/runtime/v2. Python uses copilotkit-intelligence-runtime. These dependencies include Runtime features beyond skill delivery. The .NET adapter targets net9.0 and uses CopilotKit.Intelligence.

Native setup#

These adapters require the server and canonical client releases described below.

Each example shows an optional revision pin. Replace "exact-revision-id" with a published revision ID, or remove that parameter to use the environment fallback. If neither code nor environment specifies a revision, the adapter follows the latest published Skills. The SDK does not support "auto" as a special value.

BuiltInAgent#

Configure learnedSkills directly on BuiltInAgent. No wrapper or separate adapter package is required.

Set the container and revision in code below, or omit those parameters and use environment variables. Code values take precedence.

import { BuiltInAgent } from "@copilotkit/runtime/v2";

const agent = new BuiltInAgent({
  model: "openai/gpt-4o",
  prompt: "Follow the application's support policy.",
  learnedSkills: {
    // Set these here, or omit them to use environment variables (linked above).
    containerId: "support-learning",
    revision: "exact-revision-id", // Optional: pin a published revision.
  },
});

Set CPK_INTELLIGENCE_API_KEY for skill delivery and your model provider's key separately. The configuration also accepts an existing CopilotKitIntelligence instance as client, or explicit apiKey and apiUrl. These credentials apply to Intelligence, not the model provider. The environment fallbacks and refresh settings below apply.

Classic mode adds the catalog and both executable skill tools before the model call. When skills are available, the default step limit is 10 so the model can load and use guidance. An explicit maxSteps takes precedence. Without available skills, the existing default step limit remains unchanged.

Factory mode

BuiltInAgent fetches the snapshot before it calls your factory. Connect the catalog and tools to your model call:

Set the container and revision in code below, or omit those parameters and use environment variables. Code values take precedence.

import {
  BuiltInAgent,
  convertMessagesToVercelAISDKMessages,
} from "@copilotkit/runtime/v2";
import { openai } from "@ai-sdk/openai";
import { stepCountIs, streamText } from "ai";

const agent = new BuiltInAgent({
  type: "aisdk",
  learnedSkills: {
    // Set these here, or omit them to use environment variables (linked above).
    containerId: "support-learning",
    revision: "exact-revision-id", // Optional: pin a published revision.
  },
  factory: ({ input, abortSignal, learnedSkills }) =>
    streamText({
      model: openai("gpt-4o"),
      system: [
        "Follow the application's support policy.",
        learnedSkills.catalog,
      ].filter(Boolean).join("\n\n"),
      messages: convertMessagesToVercelAISDKMessages(input.messages),
      tools: { ...learnedSkills.tools },
      stopWhen: stepCountIs(10),
      abortSignal,
    }),
});

Import BuiltInAgentFactoryContext from @copilotkit/runtime/v2 to annotate a factory context. Every factory receives a learnedSkills object. Its catalog is "" and its tools is {} when delivery is unconfigured or the verified snapshot contains no skills. Omitting the configuration disables all skill requests, even when delivery environment variables exist. With configuration, later runs check for newly published skills according to the freshness window.

The tools use AI SDK schemas and executors. TanStack and custom factories receive the same object and must adapt the tools to their model library; they are not TanStack-native tool definitions. Factory code owns the model call and its step limit. Keep the supplied catalog and tools together within that invocation. The tool map is read-only; spread it into a new object to add application tools.

Each run, including a resume, acquires one snapshot before factory or model work. Agent clones share the refresh cache; each invocation retains its own snapshot. Cancellation stops that invocation's delivery wait without cancelling another clone's shared refresh. Snapshots stay out of application state. Existing cached skills can cover transient failures, but a confirmed delivery denial blocks new execution.

Reserve copilotkit_load_skill and copilotkit_read_skill_file for delivery. Classic mode rejects conflicting client, configuration, or MCP tool names. Factory code must avoid overwriting these names when it combines tool maps.

LangGraph Python#

Use Python 3.11 or later with native asynchronous agents from langchain.agents.create_agent.

pip install copilotkit-intelligence-langgraph

Set the container and revision in code below, or omit those parameters and use environment variables. Code values take precedence.

import asyncio

from copilotkit_intelligence_langgraph import create_skill_registry_middleware
from langchain.agents import create_agent


async def main():
    skills = create_skill_registry_middleware(
        # Set these here, or omit them to use environment variables (linked above).
        container_id="support-learning",
        revision="exact-revision-id",  # Optional: pin a published revision.
    )  # Uses the environment above for the project key.
    try:
        await skills.initialize()
        agent = create_agent(
            "your-provider:your-model",
            system_prompt="Follow the application's support policy.",
            middleware=[skills],
        )
        result = await agent.ainvoke(
            {"messages": [{"role": "user", "content": "Help with a refund"}]}
        )
        print(result["messages"][-1].content)
    finally:
        await skills.aclose()


asyncio.run(main())

Use ainvoke or astream; synchronous execution is unsupported. The adapter supports LangChain >=1.2.16,<2 and LangGraph >=1.1.10,<2. Attach middleware explicitly to each agent that needs skills. Arbitrary compiled graphs are outside this integration.

LangGraph TypeScript#

Use Node.js 20.19 or later with native LangChain agents.

npm install @copilotkit/intelligence-langgraph langchain @langchain/core @langchain/langgraph zod
npm install --save-dev tsx

Save this example as agent.mts. Run it with npx tsx agent.mts:

Set the container and revision in code below, or omit those parameters and use environment variables. Code values take precedence.

import { createAgent } from "langchain";
import {
  SkillRegistry,
  createSkillRegistryMiddleware,
} from "@copilotkit/intelligence-langgraph";

const registry = new SkillRegistry({
  // Set these here, or omit them to use environment variables (linked above).
  containerId: "support-learning",
  revision: "exact-revision-id", // Optional: pin a published revision.
}); // Uses the environment above for the project key.
await registry.initialize(); // Catch this error in your application's startup code.
const skills = createSkillRegistryMiddleware({ registry });
const agent = skills.wrapAgent(
  createAgent({
    model: "your-provider:your-model",
    systemPrompt: "Follow the application's support policy.",
    middleware: [skills],
  }),
);
const result = await agent.invoke({
  messages: [{ role: "user", content: "Help with a refund" }],
});
console.log(result.messages.at(-1)?.content);

The wrapper is currently required because affected LangChain/LangGraph versions cannot keep private transient middleware state correctly. Use the wrapped agent for invoke, stream, and streamEvents, including native Command resumes. withConfig retains the wrapper. Apply default cancellation signals with agent.withConfig after wrapping, or pass the signal to each invocation. Stream objects, readers, and native cancellation remain available. Calling the middleware without its wrapper returns INVALID_CONFIG.

After the native framework fix passes the same lifecycle tests, middleware-only setup will become the default. The wrapper will be optional, and existing wrapped agents will continue to work.

Attach this middleware and its wrapper to each selected agent. Several agents can share one registry, but each wrapped invocation captures its own snapshot. Skills do not propagate automatically to arbitrary subagents. The wrapper supports the native agent's graph execution entry points for integrations that access .graph; graph batching and arbitrary compiled graph adaptation are outside this integration.

The supported dependency ranges are LangChain >=1.5.11,<2, LangGraph >=1.4.14,<2, and LangChain core >=1.2.10,<2. The registry owns in-memory snapshots and performs no filesystem writes. Your application retains ownership of an injected canonical client.

Mastra#

Use Node.js 22.13 or later with @mastra/core>=1.0.0,<2.

npm install @copilotkit/intelligence-mastra @mastra/core zod
npm install --save-dev tsx

Save this example as agent.mts. Run it with npx tsx agent.mts. Register the processor and its tools on the native Agent:

Set the container and revision in code below, or omit those parameters and use environment variables. Code values take precedence.

import { Agent } from "@mastra/core/agent";
import {
  SkillRegistry,
  createSkillRegistryProcessor,
} from "@copilotkit/intelligence-mastra";

const registry = new SkillRegistry({
  // Set these here, or omit them to use environment variables (linked above).
  containerId: "support-learning",
  revision: "exact-revision-id", // Optional: pin a published revision.
}); // Uses the environment above for the project key.
await registry.initialize(); // Catch this error in your application's startup code.
const skills = createSkillRegistryProcessor({ registry });
const agent = skills.wrapAgent(
  new Agent({
    id: "support",
    name: "Support",
    model: "openai/gpt-4.1",
    instructions: "Follow the application's support policy.",
    inputProcessors: [skills],
    tools: { ...skills.tools },
  }),
);
const result = await agent.generate("Help with a refund.");

Keep other processors and tools in the same arrays and maps. Reserve the names copilotkit_load_skill and copilotkit_read_skill_file for this adapter.

Call the wrapped agent for generate, stream, resumeGenerate, and resumeStream. The wrapper checks delivery before native execution and preserves Mastra's stream result. It also covers native tool approval and decline methods when your installed Mastra version provides them: approveToolCall, declineToolCall, approveToolCallGenerate, and declineToolCallGenerate. A resume starts a new invocation and captures the current verified snapshot, including when a tool runs before the next model call.

Register the processor and tools, and wrap each selected agent. Several agents can share a registry; each invocation keeps its own snapshot. Subagents follow Mastra's native propagation rules. Networks, legacy methods, background workers, and separate durable-worker dispatch are outside this integration.

Pass abortSignal in the invocation options to cancel both the delivery wait and native execution. A signal set only in the agent's defaultOptions applies to native execution, but cannot cancel the delivery wait that runs before it. Cancelling one invocation does not cancel a registry refresh shared with other callers.

Google ADK#

Use Python 3.11 or later and add SkillToolset to a standard ADK LlmAgent.

pip install copilotkit-intelligence-adk

Set the container and revision in code below, or omit those parameters and use environment variables. Code values take precedence.

import asyncio

from copilotkit_intelligence_adk import SkillRegistry, SkillToolset
from google.adk.agents import LlmAgent
from google.adk.runners import Runner
from google.adk.sessions import InMemorySessionService
from google.genai import types


async def main():
    registry = SkillRegistry(
        # Set these here, or omit them to use environment variables (linked above).
        container_id="support-learning",
        revision="exact-revision-id",  # Optional: pin a published revision.
    )
    try:
        await registry.initialize()
        agent = LlmAgent(
            name="support",
            model="your-model",
            instruction="Follow the application's support policy.",
            tools=[SkillToolset(registry)],
        )
        sessions = InMemorySessionService()
        session = await sessions.create_session(app_name="support", user_id="demo")
        runner = Runner(agent=agent, app_name="support", session_service=sessions)
        message = types.Content(
            role="user", parts=[types.Part(text="Help with a refund.")]
        )
        async for event in runner.run_async(
            user_id="demo", session_id=session.id, new_message=message
        ):
            if event.is_final_response() and event.content:
                for part in event.content.parts or []:
                    if part.text:
                        print(part.text)
    finally:
        await registry.aclose()


asyncio.run(main())

The adapter supports google-adk>=1.17,<2. Several selected agents can share one registry. Closing a toolset does not close that shared registry. Native resumed runs receive a fresh invocation pin; session state contains no snapshot or lock.

Microsoft Agent Framework#

Use the provider's native agent factory with your application's IChatClient:

Set the container and revision in code below, or omit those parameters and use environment variables. Code values take precedence.

using CopilotKit.Intelligence.AgentFramework;
using Microsoft.Agents.AI;

using var skills = new SkillRegistryContextProvider(new SkillRegistryOptions
{
    // Set these here, or omit them to use environment variables (linked above).
    ContainerId = "support-learning",
    Revision = "exact-revision-id", // Optional: pin a published revision.
});
await skills.InitializeAsync();
var agent = skills.CreateAgent(chatClient, new ChatClientAgentOptions
{
    Name = "support",
    ChatOptions = new() { Instructions = "Follow the application's support policy." }
});
var response = await agent.RunAsync("Help with a refund.");

The adapter targets .NET 9 and Agent Framework >=1.0.0,<2.0.0. AddCopilotKitIntelligenceSkills also registers a keyed provider and native agent through dependency injection. Use this extension or CreateAgent for complete invocation checks. Background responses and continuation tokens are unsupported because the framework bypasses context providers for those calls.

Configure several containers#

Use containers to combine Skills from several Learning containers in the same project. Each entry has an id and an optional revision pin. An entry without a revision follows that container's latest published Skills.

BuiltInAgent accepts the list inside learnedSkills:

const agent = new BuiltInAgent({
  model: "openai/gpt-4o",
  learnedSkills: {
    containers: [
      { id: "support", revision: "revision-123" },
      { id: "company-wide" },
    ],
  },
});

Mastra and LangGraph TypeScript accept the same list in SkillRegistry:

const registry = new SkillRegistry({
  containers: [
    { id: "support", revision: "revision-123" },
    { id: "company-wide" },
  ],
});
await registry.initialize();

LangGraph Python accepts the list in its middleware factory:

skills = create_skill_registry_middleware(
    containers=[
        {"id": "support", "revision": "revision-123"},
        {"id": "company-wide"},
    ],
)

Google ADK accepts the same Python list in SkillRegistry(containers=[...]). For Microsoft Agent Framework, use typed container entries:

var options = new SkillRegistryOptions
{
    Containers = new[]
    {
        new SkillContainerSource { Id = "support", Revision = "revision-123" },
        new SkillContainerSource { Id = "company-wide" },
    },
};

The list must contain 1–50 entries, with unique, nonempty IDs. Credentials, timeouts, and the freshness window apply to every entry. All selected containers must belong to the project that the client key identifies.

Do not combine containers with the old containerId or top-level revision fields. The SDK rejects this combination, including the corresponding Python and .NET fields. TypeScript also rejects the combination at compile time.

An explicit containers list ignores CPK_INTELLIGENCE_LEARNING_CONTAINER_ID and CPK_INTELLIGENCE_SKILLS_REVISION. Connection environment variables still apply. The single-container interface and its environment defaults remain supported unchanged.

One request for several containers#

The SDK sends one batch request for all containers that need a refresh. Each entry includes its container ID, optional revision pin, and cached ETag. The response carries a snapshot, an unchanged result, or an error for each container. Fresh cached containers need no request. An explicit containers list uses this endpoint even when it contains one entry.

The new interface requires an Intelligence server that supports POST /api/v1/learning/skills/batch. For self-hosted Intelligence, update the server before you enable containers. The SDK does not fall back to separate requests on older servers. The original single-container interface keeps its existing endpoint.

Skill names and failures#

The new interface prefixes every Skill name with its container ID, even for a list with one entry. For example, support/refund-policy and company-wide/refund-policy remain separate Skills. The SDK URI-encodes the container ID in this prefix. Tool calls use the exact name from the catalog. The old interface keeps its original, unprefixed names.

Each container keeps its own revision, cache, and refresh state. An invocation captures the complete combined catalog before model work starts. It keeps those Skills for all its tool calls, even after a later refresh.

If any container cannot load its first snapshot, the invocation fails. A warm container can use its cached Skills after a transient failure. A confirmed access denial or revoked revision blocks the whole invocation. The SDK never silently supplies only part of the selected catalog.

For the new interface, each entry in status.containers reports its container's revision and refresh status. .NET exposes the entries through MultiSkillRegistryStatus.Containers. The aggregate status has no server revision to pin. Its mode is pinned only when every container has a revision pin.

Configure one container#

The original interface selects one Learning container. Set containerId in TypeScript, container_id in Python, or ContainerId in .NET, as shown above. BuiltInAgent uses learnedSkills.containerId.

The singular fields accept one stable container ID. To combine containers, use the separate containers interface above. Several agents can share a registry.

An injected canonical Intelligence client supplies its existing project key, endpoint, and HTTP lifecycle. Your application retains ownership of that client. The adapter does not construct a second authenticated transport.

Environment-based setup uses these variables. npx copilotkit@latest project select writes the project key. If the CLI has no session, run npx copilotkit@latest login first. If the agent server is a different directory, copy the key into this environment.

CPK_INTELLIGENCE_API_KEY=cpk-...
CPK_INTELLIGENCE_LEARNING_CONTAINER_ID=expense-review
# Self-hosted deployments only:
INTELLIGENCE_API_URL=https://intelligence.example.com
# Optional exact revision; omit to follow latest:
CPK_INTELLIGENCE_SKILLS_REVISION=42

Explicit configuration overrides environment values. An injected client overrides connection environment values. Freshness and request timeout each default to five seconds; debug logging defaults to false. Configure these behavior options in code.

Initialization can run during application startup. A valid empty container initializes successfully. Initialization failures are catchable and retryable; the adapter never terminates the process. Model work cannot start until a verified snapshot exists.

Reuse an Intelligence SDK client#

If your server already creates an Intelligence client, pass that client to the registry. This reuses its project key and endpoint.

Set the container and revision in code below, or omit those parameters and use environment variables. Code values take precedence.

import { CopilotKitIntelligence } from "@copilotkit/runtime/v2";
import { SkillRegistry } from "@copilotkit/intelligence-mastra";

const intelligence = new CopilotKitIntelligence({
  apiKey: process.env.CPK_INTELLIGENCE_API_KEY!,
  apiUrl: process.env.INTELLIGENCE_API_URL,
});
const registry = new SkillRegistry({
  client: intelligence,
  // Set these here, or omit them to use environment variables (linked above).
  containerId: "support-learning",
  revision: "exact-revision-id", // Optional: pin a published revision.
});
await registry.initialize();

For LangGraph TypeScript, import SkillRegistry from @copilotkit/intelligence-langgraph and use the same configuration. Then pass the registry to the framework adapter shown above.

Make sure delivery works#

For framework adapters, this check requires automatic updates. Remove revision (or Revision in .NET) from the example and unset CPK_INTELLIGENCE_SKILLS_REVISION. A pinned adapter continues to use its selected revision after you publish a newer one.

1. Initialize the adapter with a published Skill in its container. Record the loaded revision from the adapter's status.
2. Start a new invocation with a request that relates to that Skill.
3. Inspect the model's tool calls for copilotkit_load_skill and, when needed, copilotkit_read_skill_file.
4. Publish a new revision, wait past the freshness window, and start another invocation.
5. Make sure that the loaded revision in the adapter's status changed to the newly published revision.

For the original single-container interface, read registry.status.revision in TypeScript, skills.status.revision for LangGraph Python, registry.status.revision for ADK, or skills.Status.Revision in .NET. BuiltInAgent does not expose a public registry status; inspect the loaded Skill's content in the tool result for your published change. A valid empty container can initialize successfully, so initialization alone does not prove that a Skill exists.

The model chooses whether to call the skill tools. A run without a skill tool call does not prove that delivery failed. Inspect the catalog and registry status too.

Refresh and invocation behavior#

Before an invocation, the registry evaluates the freshness window for each selected container. When a check is due, that invocation waits for it. With containers, concurrent invocations share one batch request for the containers due for refresh, with no internal retry. An unchanged response resets the freshness window without transferring the ZIP again.

Each invocation keeps one complete snapshot for its model and tool calls. A later refresh affects later invocations. Revision identifiers are opaque: pass the exact string without parsing, incrementing, or comparing it numerically.

Published revisions have no automatic expiry. A revoked revision cannot be used by new invocations once the adapter receives that denial. An invocation already in progress finishes with its captured snapshot.

Read tools#

The framework adapters keep both tools registered even when the snapshot is empty. BuiltInAgent omits both tools for an empty snapshot; its factory receives tools: {}. With available skills, the tools are:

• copilotkit_load_skill(skill_name) returns SKILL.md and the supporting file list.
• copilotkit_read_skill_file(skill_name, path) returns one supporting UTF-8 text file.

Lookups match the invocation's manifest. Unknown names, unknown paths, and unsupported content use the framework's tool-error behavior. The adapter never executes scripts or writes skill files to disk.

Failures and status#

A cold registry returns a typed error when it cannot load a valid snapshot. After a successful load, a network failure, timeout, unavailable server, or unreadable replacement keeps the previous snapshot and marks it stale.

There is no maximum stale age. A network-isolated process can retain its last verified snapshot until connectivity returns. Restarting clears that in-memory snapshot. Confirmed authentication, authorization, entitlement, delivery disablement, or revocation blocks new invocations instead of falling back to stale content.

The read-only status contains initialized, revision, mode, lastCheckedAt, stale, and lastError. Errors expose a stable code, safe message, retryable, and cause. Debug mode uses the normal language logger and excludes credentials, prompts, skill bodies, and file contents.

Migrate from downloaded skills#

Remove the old manual Learning-directory wiring when you enable an adapter for the same container. Keep unrelated static or hand-authored skills as needed. The adapter does not scan, change, or delete downloaded files.

copilotkit skills download remains supported for inspection, offline use, and unsupported frameworks. Pass several container IDs to download them together:

copilotkit skills download support company-wide --output ./skills

The command writes each container under its own directory, such as ./skills/support/refund-policy/SKILL.md. It creates the destination only after every bundle passes validation. The destination must not already exist. A single container keeps the original flat output, such as ./skills/refund-policy/SKILL.md. See the Automatic Learning guide for the manual workflow.

Deployment requirements#

Cloud-hosted and self-hosted Intelligence use the same delivery contract. The server migration and v1 delivery endpoint must deploy before adapters rely on them. Each adapter also requires a published canonical client version with the learned-snapshot operation.

Pausing new Learning runs does not stop delivery. Entitlement, explicit delivery disablement, and revision revocation control access to published snapshots separately.