agenticdome-python-sdk 1.2.12__tar.gz → 1.2.13__tar.gz

{agenticdome_python_sdk-1.2.12 → agenticdome_python_sdk-1.2.13}/PKG-INFO

@@ -1,12 +1,13 @@
 Metadata-Version: 2.4
 Name: agenticdome-python-sdk
-Version: 1.2.12
+Version: 1.2.13
 Summary: Official Python SDK for AgenticDome security guardrails, DLP, tool authorization, and multi-agent delegation enforcement.
 Author: AgenticDome
 License-Expression: Apache-2.0
 Project-URL: Homepage, https://www.agenticdome.io
 Project-URL: Repository, https://github.com/agenticdome/agenticdome-python-sdk
 Project-URL: Documentation, https://github.com/agenticdome/agenticdome-python-sdk#readme
+Project-URL: Framework Integration Guides, https://github.com/agenticdome/agenticdome-python-sdk/tree/main/docs/frameworks
 Project-URL: MCP Integration Guide, https://github.com/agenticdome/agenticdome-python-sdk/blob/main/docs/mcp-integration.md
 Project-URL: Issues, https://github.com/agenticdome/agenticdome-python-sdk/issues
 Project-URL: Examples, https://github.com/agenticdome/agenticdome-python-sdk/tree/main/examples
@@ -115,7 +116,7 @@ Dynamic: license-file
 [![CI](https://github.com/agenticdome/agenticdome-python-sdk/actions/workflows/ci.yml/badge.svg)](https://github.com/agenticdome/agenticdome-python-sdk/actions/workflows/ci.yml)
 [![License: Apache-2.0](https://img.shields.io/badge/License-Apache--2.0-blue.svg)](https://www.apache.org/licenses/LICENSE-2.0)
 
-[Source and examples](https://github.com/agenticdome/agenticdome-python-sdk) · [MCP integration guide](https://github.com/agenticdome/agenticdome-python-sdk/blob/main/docs/mcp-integration.md) · [Issue tracker](https://github.com/agenticdome/agenticdome-python-sdk/issues) · [Security policy](https://github.com/agenticdome/agenticdome-python-sdk/blob/main/SECURITY.md) · [PyPI package](https://pypi.org/project/agenticdome-python-sdk/)
+[Source and examples](https://github.com/agenticdome/agenticdome-python-sdk) · [Framework integration guides](https://github.com/agenticdome/agenticdome-python-sdk/tree/main/docs/frameworks) · [MCP integration guide](https://github.com/agenticdome/agenticdome-python-sdk/blob/main/docs/mcp-integration.md) · [Issue tracker](https://github.com/agenticdome/agenticdome-python-sdk/issues) · [Security policy](https://github.com/agenticdome/agenticdome-python-sdk/blob/main/SECURITY.md) · [PyPI package](https://pypi.org/project/agenticdome-python-sdk/)
 
 > **Production-grade security guardrails, DLP, tool authorization, and cryptographically verified multi-agent delegation for Python autonomous AI runtimes.**
 
@@ -154,7 +155,7 @@ agenticdome-demo --list-frameworks
 agenticdome-demo --framework all --scenario both
 ```
 
-You can also open the [framework example gallery](examples/README.md) and run the individual example matching your stack. Before editing production code, use the [production integration playbook](examples/PRODUCTION_INTEGRATION.md) to find the exact construction file, public attachment call and local execution boundary for your framework. MCP host and gateway developers can follow the dedicated [MCP Action Firewall guide](https://github.com/agenticdome/agenticdome-python-sdk/blob/main/docs/mcp-integration.md).
+You can also open the [framework example gallery](examples/README.md) and run the individual example matching your stack. Before editing production code, choose the dedicated [framework integration guide](docs/frameworks/README.md), then use the [production integration playbook](examples/PRODUCTION_INTEGRATION.md) for the cross-framework attachment and proof checklist. MCP host and gateway developers can follow the dedicated [MCP Action Firewall guide](https://github.com/agenticdome/agenticdome-python-sdk/blob/main/docs/mcp-integration.md).
 
 ### 5. Connect the same integration to AgenticDome
 
@@ -189,7 +190,7 @@ result = crew.kickoff()          # hostile prompts, unsafe tools, and rogue
 4. [Quickstart](#quickstart)
 5. [Installation](#installation)
 6. [Configuration](#configuration)
-7. [Choosing Your Integration Point](#choosing-your-integration-point)
+7. [Choosing Your Integration Point](#choosing-your-integration-point) — [dedicated framework guides](docs/frameworks/README.md)
 8. [Framework Integrations](#framework-integrations)
    - [CrewAI](#crewai) · [PydanticAI](#pydanticai) · [LangGraph](#langgraph) · [Microsoft Agent Framework](#microsoft-agent-framework) · [Microsoft AutoGen](#microsoft-autogen) · [Microsoft AI Foundry](#microsoft-ai-foundry) · [OpenAI Agents SDK](#openai-agents-sdk) · [Claude Agent SDK](#claude-agent-sdk) · [Hugging Face smolagents](#hugging-face-smolagents) · [Agno](#agno) · [Google ADK](#google-adk) · [LlamaIndex](#llamaindex) · [AWS Bedrock](#aws-bedrock) · [MCP Host / Gateway](#mcp-host--gateway)
 9. [Core SDK Client (Custom Runtimes)](#core-sdk-client-custom-runtimes)

{agenticdome_python_sdk-1.2.12 → agenticdome_python_sdk-1.2.13}/README.md

@@ -5,7 +5,7 @@
 [![CI](https://github.com/agenticdome/agenticdome-python-sdk/actions/workflows/ci.yml/badge.svg)](https://github.com/agenticdome/agenticdome-python-sdk/actions/workflows/ci.yml)
 [![License: Apache-2.0](https://img.shields.io/badge/License-Apache--2.0-blue.svg)](https://www.apache.org/licenses/LICENSE-2.0)
 
-[Source and examples](https://github.com/agenticdome/agenticdome-python-sdk) · [MCP integration guide](https://github.com/agenticdome/agenticdome-python-sdk/blob/main/docs/mcp-integration.md) · [Issue tracker](https://github.com/agenticdome/agenticdome-python-sdk/issues) · [Security policy](https://github.com/agenticdome/agenticdome-python-sdk/blob/main/SECURITY.md) · [PyPI package](https://pypi.org/project/agenticdome-python-sdk/)
+[Source and examples](https://github.com/agenticdome/agenticdome-python-sdk) · [Framework integration guides](https://github.com/agenticdome/agenticdome-python-sdk/tree/main/docs/frameworks) · [MCP integration guide](https://github.com/agenticdome/agenticdome-python-sdk/blob/main/docs/mcp-integration.md) · [Issue tracker](https://github.com/agenticdome/agenticdome-python-sdk/issues) · [Security policy](https://github.com/agenticdome/agenticdome-python-sdk/blob/main/SECURITY.md) · [PyPI package](https://pypi.org/project/agenticdome-python-sdk/)
 
 > **Production-grade security guardrails, DLP, tool authorization, and cryptographically verified multi-agent delegation for Python autonomous AI runtimes.**
 
@@ -44,7 +44,7 @@ agenticdome-demo --list-frameworks
 agenticdome-demo --framework all --scenario both
 ```
 
-You can also open the [framework example gallery](examples/README.md) and run the individual example matching your stack. Before editing production code, use the [production integration playbook](examples/PRODUCTION_INTEGRATION.md) to find the exact construction file, public attachment call and local execution boundary for your framework. MCP host and gateway developers can follow the dedicated [MCP Action Firewall guide](https://github.com/agenticdome/agenticdome-python-sdk/blob/main/docs/mcp-integration.md).
+You can also open the [framework example gallery](examples/README.md) and run the individual example matching your stack. Before editing production code, choose the dedicated [framework integration guide](docs/frameworks/README.md), then use the [production integration playbook](examples/PRODUCTION_INTEGRATION.md) for the cross-framework attachment and proof checklist. MCP host and gateway developers can follow the dedicated [MCP Action Firewall guide](https://github.com/agenticdome/agenticdome-python-sdk/blob/main/docs/mcp-integration.md).
 
 ### 5. Connect the same integration to AgenticDome
 
@@ -79,7 +79,7 @@ result = crew.kickoff()          # hostile prompts, unsafe tools, and rogue
 4. [Quickstart](#quickstart)
 5. [Installation](#installation)
 6. [Configuration](#configuration)
-7. [Choosing Your Integration Point](#choosing-your-integration-point)
+7. [Choosing Your Integration Point](#choosing-your-integration-point) — [dedicated framework guides](docs/frameworks/README.md)
 8. [Framework Integrations](#framework-integrations)
    - [CrewAI](#crewai) · [PydanticAI](#pydanticai) · [LangGraph](#langgraph) · [Microsoft Agent Framework](#microsoft-agent-framework) · [Microsoft AutoGen](#microsoft-autogen) · [Microsoft AI Foundry](#microsoft-ai-foundry) · [OpenAI Agents SDK](#openai-agents-sdk) · [Claude Agent SDK](#claude-agent-sdk) · [Hugging Face smolagents](#hugging-face-smolagents) · [Agno](#agno) · [Google ADK](#google-adk) · [LlamaIndex](#llamaindex) · [AWS Bedrock](#aws-bedrock) · [MCP Host / Gateway](#mcp-host--gateway)
 9. [Core SDK Client (Custom Runtimes)](#core-sdk-client-custom-runtimes)

{agenticdome_python_sdk-1.2.12 → agenticdome_python_sdk-1.2.13}/agenticdome_python_sdk.egg-info/PKG-INFO

@@ -1,12 +1,13 @@
 Metadata-Version: 2.4
 Name: agenticdome-python-sdk
-Version: 1.2.12
+Version: 1.2.13
 Summary: Official Python SDK for AgenticDome security guardrails, DLP, tool authorization, and multi-agent delegation enforcement.
 Author: AgenticDome
 License-Expression: Apache-2.0
 Project-URL: Homepage, https://www.agenticdome.io
 Project-URL: Repository, https://github.com/agenticdome/agenticdome-python-sdk
 Project-URL: Documentation, https://github.com/agenticdome/agenticdome-python-sdk#readme
+Project-URL: Framework Integration Guides, https://github.com/agenticdome/agenticdome-python-sdk/tree/main/docs/frameworks
 Project-URL: MCP Integration Guide, https://github.com/agenticdome/agenticdome-python-sdk/blob/main/docs/mcp-integration.md
 Project-URL: Issues, https://github.com/agenticdome/agenticdome-python-sdk/issues
 Project-URL: Examples, https://github.com/agenticdome/agenticdome-python-sdk/tree/main/examples
@@ -115,7 +116,7 @@ Dynamic: license-file
 [![CI](https://github.com/agenticdome/agenticdome-python-sdk/actions/workflows/ci.yml/badge.svg)](https://github.com/agenticdome/agenticdome-python-sdk/actions/workflows/ci.yml)
 [![License: Apache-2.0](https://img.shields.io/badge/License-Apache--2.0-blue.svg)](https://www.apache.org/licenses/LICENSE-2.0)
 
-[Source and examples](https://github.com/agenticdome/agenticdome-python-sdk) · [MCP integration guide](https://github.com/agenticdome/agenticdome-python-sdk/blob/main/docs/mcp-integration.md) · [Issue tracker](https://github.com/agenticdome/agenticdome-python-sdk/issues) · [Security policy](https://github.com/agenticdome/agenticdome-python-sdk/blob/main/SECURITY.md) · [PyPI package](https://pypi.org/project/agenticdome-python-sdk/)
+[Source and examples](https://github.com/agenticdome/agenticdome-python-sdk) · [Framework integration guides](https://github.com/agenticdome/agenticdome-python-sdk/tree/main/docs/frameworks) · [MCP integration guide](https://github.com/agenticdome/agenticdome-python-sdk/blob/main/docs/mcp-integration.md) · [Issue tracker](https://github.com/agenticdome/agenticdome-python-sdk/issues) · [Security policy](https://github.com/agenticdome/agenticdome-python-sdk/blob/main/SECURITY.md) · [PyPI package](https://pypi.org/project/agenticdome-python-sdk/)
 
 > **Production-grade security guardrails, DLP, tool authorization, and cryptographically verified multi-agent delegation for Python autonomous AI runtimes.**
 
@@ -154,7 +155,7 @@ agenticdome-demo --list-frameworks
 agenticdome-demo --framework all --scenario both
 ```
 
-You can also open the [framework example gallery](examples/README.md) and run the individual example matching your stack. Before editing production code, use the [production integration playbook](examples/PRODUCTION_INTEGRATION.md) to find the exact construction file, public attachment call and local execution boundary for your framework. MCP host and gateway developers can follow the dedicated [MCP Action Firewall guide](https://github.com/agenticdome/agenticdome-python-sdk/blob/main/docs/mcp-integration.md).
+You can also open the [framework example gallery](examples/README.md) and run the individual example matching your stack. Before editing production code, choose the dedicated [framework integration guide](docs/frameworks/README.md), then use the [production integration playbook](examples/PRODUCTION_INTEGRATION.md) for the cross-framework attachment and proof checklist. MCP host and gateway developers can follow the dedicated [MCP Action Firewall guide](https://github.com/agenticdome/agenticdome-python-sdk/blob/main/docs/mcp-integration.md).
 
 ### 5. Connect the same integration to AgenticDome
 
@@ -189,7 +190,7 @@ result = crew.kickoff()          # hostile prompts, unsafe tools, and rogue
 4. [Quickstart](#quickstart)
 5. [Installation](#installation)
 6. [Configuration](#configuration)
-7. [Choosing Your Integration Point](#choosing-your-integration-point)
+7. [Choosing Your Integration Point](#choosing-your-integration-point) — [dedicated framework guides](docs/frameworks/README.md)
 8. [Framework Integrations](#framework-integrations)
    - [CrewAI](#crewai) · [PydanticAI](#pydanticai) · [LangGraph](#langgraph) · [Microsoft Agent Framework](#microsoft-agent-framework) · [Microsoft AutoGen](#microsoft-autogen) · [Microsoft AI Foundry](#microsoft-ai-foundry) · [OpenAI Agents SDK](#openai-agents-sdk) · [Claude Agent SDK](#claude-agent-sdk) · [Hugging Face smolagents](#hugging-face-smolagents) · [Agno](#agno) · [Google ADK](#google-adk) · [LlamaIndex](#llamaindex) · [AWS Bedrock](#aws-bedrock) · [MCP Host / Gateway](#mcp-host--gateway)
 9. [Core SDK Client (Custom Runtimes)](#core-sdk-client-custom-runtimes)

{agenticdome_python_sdk-1.2.12 → agenticdome_python_sdk-1.2.13}/agenticdome_python_sdk.egg-info/SOURCES.txt

@@ -35,6 +35,22 @@ agenticdome_sdk/pydantic.py
 agenticdome_sdk/simulation.py
 agenticdome_sdk/smolagents.py
 docs/mcp-integration.md
+docs/performance-evidence.md
+docs/frameworks/README.md
+docs/frameworks/agno.md
+docs/frameworks/autogen.md
+docs/frameworks/aws-bedrock.md
+docs/frameworks/claude-agent-sdk.md
+docs/frameworks/crewai.md
+docs/frameworks/custom-python.md
+docs/frameworks/google-adk.md
+docs/frameworks/langgraph.md
+docs/frameworks/llamaindex.md
+docs/frameworks/microsoft-agent-framework.md
+docs/frameworks/microsoft-ai-foundry.md
+docs/frameworks/openai-agents.md
+docs/frameworks/pydanticai.md
+docs/frameworks/smolagents.md
 examples/PRODUCTION_INTEGRATION.md
 examples/README.md
 examples/attack_demo.py
@@ -67,6 +83,7 @@ tests/test_claude_integration.py
 tests/test_client.py
 tests/test_crewai_integration.py
 tests/test_execution_broker.py
+tests/test_framework_guides.py
 tests/test_google_adk_integration.py
 tests/test_identity_context.py
 tests/test_langgraph_integration.py

agenticdome_python_sdk-1.2.13/docs/frameworks/README.md

@@ -0,0 +1,43 @@
+# AgenticDome Python framework integration guides
+
+Choose the runtime that owns your agent, graph, tool or gateway boundary. Each
+guide uses the same production contract: configure the SDK, attach it in code,
+authorize immediately before local execution, review returned content, and
+prove both an allowed and blocked path.
+
+| Integration | Guide | Primary protected boundary |
+| --- | --- | --- |
+| CrewAI | [CrewAI](crewai.md) | Crew bootstrap, LLM/tool hooks and local tools |
+| PydanticAI | [PydanticAI](pydanticai.md) | Agent hooks and decorated tools |
+| LangGraph / LangChain | [LangGraph](langgraph.md) | Graph nodes, transitions and tool nodes |
+| Microsoft Agent Framework | [Microsoft Agent Framework](microsoft-agent-framework.md) | Agent run, middleware and function tools |
+| Microsoft AutoGen | [AutoGen](autogen.md) | Team/Core message and tool execution |
+| Microsoft AI Foundry | [Microsoft AI Foundry](microsoft-ai-foundry.md) | Client/run boundary and local tool executor |
+| OpenAI Agents SDK | [OpenAI Agents SDK](openai-agents.md) | Runner guardrails and function tools |
+| Anthropic Claude Agent SDK | [Claude Agent SDK](claude-agent-sdk.md) | Options hooks, query/client and SDK tools |
+| Hugging Face smolagents | [smolagents](smolagents.md) | Agent run, generated code and native tools |
+| Agno | [Agno](agno.md) | Agent/Team hooks, tools and returned content |
+| Google ADK | [Google ADK](google-adk.md) | Agent callbacks and local tools |
+| LlamaIndex | [LlamaIndex](llamaindex.md) | Query, retrieval and FunctionTool boundaries |
+| AWS Bedrock | [AWS Bedrock](aws-bedrock.md) | boto3 calls, tool use and action groups |
+| MCP host / gateway | [MCP](../mcp-integration.md) | JSON-RPC forwarding and returned content |
+| Custom Python | [Custom Python](custom-python.md) | Application-owned prompt/tool/output boundary |
+
+## Shared five-step path
+
+1. Install the matching package extra.
+2. Run `AGENTICDOME_MODE=local_sim` with one allowed and one blocked scenario.
+3. Obtain the assigned runtime sidecar URL, Runtime/SDK key and tenant ID.
+4. Attach the adapter in the application construction path and remove raw
+   execution routes that bypass it.
+5. Prove safe, blocked, redacted and sidecar-unavailable behavior before
+   production, then run SDK Assurance and Performance Smoke.
+
+The offline simulator proves SDK control flow, not tenant enforcement. The
+public SDK does not expose AgenticDome policy algorithms, private threat
+signatures, decision-token formats or control-plane internals.
+
+Use the [production integration playbook](../../examples/PRODUCTION_INTEGRATION.md)
+for the cross-framework handoff checklist and the
+[performance evidence guide](../performance-evidence.md) when publishing
+latency or throughput results.

agenticdome_python_sdk-1.2.13/docs/frameworks/agno.md

@@ -0,0 +1,46 @@
+# Agno integration
+
+Attach AgenticDome centrally after constructing an Agno Agent or Team and
+before its first run. Secure local tools explicitly, especially tools that
+read protected data, mutate state or call an external service.
+
+## Try it without an account
+
+```bash
+pip install "agenticdome-python-sdk[agno]"
+export AGENTICDOME_MODE=local_sim
+agenticdome-demo --framework agno --scenario both
+```
+
+## Attach in production
+
+```python
+from agenticdome_sdk.agno import AgenticDomeAgnoFirewall
+
+firewall = AgenticDomeAgnoFirewall()
+agent = firewall.attach_firewall(agent)
+
+@firewall.secure_tool(
+    tool_name="crm.customer.read",
+    tool_platform="crm",
+)
+def read_customer(agent, customer_id: str):
+    return crm.read_customer(customer_id)
+```
+
+Use `create_hook_bundle()`, `create_middleware()` or `create_plugin()` when the
+selected Agno assembly style expects those objects. Ensure components created
+later by factories receive the same attachment.
+
+See the [Agno API guide](../../README.md#agno) for Agent/Team hooks, plugins,
+secured tools, delegation, retrieval and streaming examples.
+
+## Launch checks
+
+- The central Agent/Team/Workflow factory performs the attachment once.
+- Components dynamically created after startup are also attached.
+- Tool hooks execute sanitized arguments and review returned values.
+- Retrieved content is sanitized before it enters model context.
+- Specialist delegation is verified at the specialist's execution boundary.
+
+An unattached component or raw tool remains outside SDK protection.

agenticdome_python_sdk-1.2.13/docs/frameworks/autogen.md

@@ -0,0 +1,45 @@
+# Microsoft AutoGen integration
+
+AgenticDome supports current AgentChat/Core applications and documented legacy
+`ConversableAgent` deployments. For new Microsoft projects, evaluate Microsoft
+Agent Framework while retaining this integration for existing AutoGen estates.
+
+## Try it without an account
+
+```bash
+pip install "agenticdome-python-sdk[autogen]"
+export AGENTICDOME_MODE=local_sim
+agenticdome-demo --framework autogen --scenario both
+```
+
+## Attach in production
+
+```python
+from agenticdome_sdk.autogen import AgenticDomeAutoGenFirewall
+
+firewall = AgenticDomeAutoGenFirewall()
+secure_team = firewall.wrap_team(
+    team,
+    session_id="stable-session-id",
+    agent_id="operations-team",
+)
+result = await secure_team.run(task=user_prompt)
+```
+
+For AutoGen Core, install `create_intervention_handler(...)` when constructing
+the runtime. For legacy deployments, use `attach_conversable_agent(...)` and
+the dependency-light base SDK. In every variant, wrap local side-effecting
+handlers with `wrap_tool_handler(...)`.
+
+See the [AutoGen API guide](../../README.md#microsoft-autogen) for AgentChat,
+Core intervention, termination/freeze, legacy and tool examples.
+
+## Launch checks
+
+- Application code invokes the secured team proxy, not the original team.
+- Core runtimes include the intervention handler before startup.
+- Session freeze/denial is terminal until an authorized resolution.
+- Local tools cannot be reached through an unwrapped registration path.
+- The selected AutoGen generation is inside its certified dependency range.
+
+Conversation screening alone does not authorize the eventual tool execution.

agenticdome_python_sdk-1.2.13/docs/frameworks/aws-bedrock.md

@@ -0,0 +1,49 @@
+# AWS Bedrock integration
+
+Replace direct application-owned Bedrock Runtime/Agent calls with the matching
+secure method. Separately wrap local tool-use handlers, action-group Lambda
+handlers and knowledge-base retrieval results.
+
+## Try it without an account
+
+```bash
+pip install "agenticdome-python-sdk[bedrock]"
+export AGENTICDOME_MODE=local_sim
+agenticdome-demo --framework bedrock --scenario both
+```
+
+## Attach in production
+
+```python
+import boto3
+from agenticdome_sdk.aws_bedrock import AgenticDomeAWSBedrockFirewall
+
+bedrock = boto3.client("bedrock-runtime", region_name="us-east-1")
+firewall = AgenticDomeAWSBedrockFirewall()
+
+response = await firewall.converse_securely(
+    bedrock_runtime_client=bedrock,
+    model_id=model_id,
+    messages=messages,
+    agent_id="support-agent",
+    session_id="stable-session-id",
+)
+```
+
+Use the corresponding secure methods for streaming, `InvokeModel` and Bedrock
+Agents. Use `wrap_tool_handler(...)`, `secure_tool(...)`, and
+`wrap_action_group_lambda(...)` at local execution boundaries.
+
+See the [AWS Bedrock API guide](../../README.md#aws-bedrock) for Converse,
+InvokeModel, streaming, Agents, action groups, tools, retrieval and handoffs.
+
+## Launch checks
+
+- No production path calls boto3 directly for a protected model operation.
+- Local tool-use and action-group handlers are independently authorized.
+- Streamed events and final responses are reviewed before delivery.
+- AWS principal, account, region and resource identifiers are propagated.
+- Retrieved knowledge-base nodes are sanitized before planner reuse.
+
+The SDK protects calls and content at the application boundary; it cannot
+instrument execution occurring solely inside AWS-managed services.

agenticdome_python_sdk-1.2.13/docs/frameworks/claude-agent-sdk.md

@@ -0,0 +1,47 @@
+# Anthropic Claude Agent SDK integration
+
+Install hooks on the exact `ClaudeAgentOptions` instance passed to the SDK,
+then consume the secured query/client iterator so returned messages cannot
+bypass output review. Secure locally exposed SDK tools separately.
+
+## Try it without an account
+
+```bash
+pip install "agenticdome-python-sdk[claude]"
+export AGENTICDOME_MODE=local_sim
+agenticdome-demo --framework claude --scenario both
+```
+
+## Attach in production
+
+```python
+from agenticdome_sdk.claude import AgenticDomeClaudeFirewall
+
+firewall = AgenticDomeClaudeFirewall()
+options = firewall.install_on_options(options)
+
+async for message in firewall.secure_query(
+    prompt=user_prompt,
+    options=options,
+    session_id="stable-session-id",
+    agent_id="support-agent",
+):
+    consume(message)
+```
+
+Use `run_client_securely(...)` for a constructed SDK client and
+`secure_sdk_tool(...)` for local SDK MCP tools. Every query/client path must use
+the options object containing the installed hooks.
+
+See the [Claude Agent SDK API guide](../../README.md#claude-agent-sdk) for hook
+composition, query/client, tool output, delegation and streaming behavior.
+
+## Launch checks
+
+- The secured options instance is the one supplied to every production client.
+- Pre-tool denials prevent the SDK tool implementation from running.
+- Tool results and final messages pass through the secured iterator.
+- Decision context is bound and consumed at delegated local execution.
+- Alternate query/client construction paths are removed or equivalently wrapped.
+
+Hooks cannot intercept an unrelated client created with different options.

agenticdome_python_sdk-1.2.13/docs/frameworks/crewai.md

@@ -0,0 +1,51 @@
+# CrewAI integration
+
+Use this integration when CrewAI owns the agent/crew lifecycle and local tool
+execution. AgenticDome screens model input, authorizes tool and delegation
+boundaries, and reviews tool/final output routed through the installed hooks.
+
+## Try it without an account
+
+```bash
+pip install "agenticdome-python-sdk[crewai]"
+export AGENTICDOME_MODE=local_sim
+agenticdome-demo --framework crewai --scenario both
+```
+
+This prints one allowed and one blocked decision without importing CrewAI,
+contacting a sidecar or executing either demonstration tool.
+
+## Attach in production
+
+Configure the assigned runtime sidecar URL, Runtime/SDK key and matching tenant
+ID, then import the adapter in application bootstrap **before** constructing
+agents or crews:
+
+```python
+import agenticdome_sdk.crewai  # registers the process-wide hooks
+
+from crewai import Agent, Crew, Task
+
+# Construct agents, tasks and crews only after the security hooks are loaded.
+```
+
+For explicit lifecycle control, instantiate `AgenticDomeCrewAIFirewall` and
+call `attach(crew)`. Protect every high-impact local tool with
+`@firewall.secure_tool(...)`; register the wrapped callable rather than its raw
+implementation. Use the sanitized arguments supplied by the wrapper.
+
+The complete, tested code patterns for scoped attachment, secured tools,
+delegation, streaming output and multi-worker state are in the
+[CrewAI API guide](../../README.md#crewai).
+
+## Launch checks
+
+- Hooks load before any `Crew` executes.
+- A blocked tool never reaches its function body.
+- Allowed and sanitized arguments are the values the real function receives.
+- Manager-to-specialist execution is verified at the specialist boundary.
+- Every worker uses stable session identity; shared delegation uses Redis.
+- A sidecar outage produces the deliberately selected fail posture.
+
+AgenticDome cannot protect a raw tool or alternate execution route that bypasses
+the hooks/wrapper. Offline simulation is not tenant assurance.

agenticdome_python_sdk-1.2.13/docs/frameworks/custom-python.md

@@ -0,0 +1,62 @@
+# Custom Python integration
+
+Use the core client when your application owns the runtime loop: API handlers,
+Celery workers, routers, bespoke agents, tool executors or custom gateways.
+Make the policy check part of the same function that owns execution.
+
+## Try it without an account
+
+```bash
+pip install agenticdome-python-sdk
+export AGENTICDOME_MODE=local_sim
+agenticdome-demo --framework custom-python --scenario both
+```
+
+## Attach in production
+
+```python
+from agenticdome_sdk import AgentGuardClient
+
+client = AgentGuardClient()
+decision = client.guardrail_validate(
+    text="Support agent requests a customer lookup.",
+    agent_id="support-agent",
+    session_id="stable-session-id",
+    direction="outbound",
+    platform="python",
+    tool_name="crm.customer.read",
+    tool_args={"customer_id": customer_id},
+)
+
+if decision.get("verdict") == "BLOCKED":
+    raise PermissionError(decision.get("reason", "Action blocked"))
+
+# Execute only after the decision and use sanitized arguments when returned.
+raw_output = execute_customer_lookup(customer_id)
+safe_output = client.mesh_validate(
+    agent_id="support-agent",
+    session_id="stable-session-id",
+    direction="output",
+    platform="python",
+    text=str(raw_output),
+)
+```
+
+Use the A2A authorization and verification methods when a manager delegates a
+sensitive tool to another agent. For high-impact outbound HTTP execution, use
+the documented broker receipt and enforcement headers at the last responsible
+moment.
+
+See the [core client API guide](../../README.md#core-sdk-client-custom-runtimes)
+for prompt, tool, A2A, DLP, incident and brokered-execution examples.
+
+## Launch checks
+
+- The real function is unreachable before an allowed decision.
+- Sanitized arguments/text replace originals when returned.
+- Output is reviewed before return, storage, logging or planner reuse.
+- Delegation authorization is verified at the target execution boundary.
+- Fail-closed behavior is tested against an unavailable sidecar.
+
+A check performed in an unrelated helper does not protect another raw executor
+path. Keep the decision and execution boundary structurally adjacent.

agenticdome_python_sdk-1.2.13/docs/frameworks/google-adk.md

@@ -0,0 +1,46 @@
+# Google ADK integration
+
+Register AgenticDome callbacks when constructing each `LlmAgent`, attach them to
+an existing agent, or register the plugin helper centrally. Explicitly secure
+local tools that can cause a side effect.
+
+## Try it without an account
+
+```bash
+pip install "agenticdome-python-sdk[google-adk]"
+export AGENTICDOME_MODE=local_sim
+agenticdome-demo --framework google-adk --scenario both
+```
+
+## Attach in production
+
+```python
+from google.adk.agents import LlmAgent
+from agenticdome_sdk.google_adk import AgenticDomeGoogleADKFirewall
+
+firewall = AgenticDomeGoogleADKFirewall()
+agent = LlmAgent(
+    name="support-agent",
+    model="gemini-2.5-flash",
+    instruction="Help support users safely.",
+    **firewall.build_callback_kwargs(),
+)
+```
+
+Use `install_on_agent(...)` for an existing agent, `create_plugin()` for
+central registration and `wrap_tool_handler(...)`/`secure_tool(...)` for local
+tools. Prefer async callbacks in async ADK runtimes.
+
+See the [Google ADK API guide](../../README.md#google-adk) for callbacks,
+plugins, tools, handoffs, schema validation and streaming examples.
+
+## Launch checks
+
+- Every production agent receives callbacks from the central factory.
+- Tool handlers use sanitized arguments before local execution.
+- Stable ADK session/run/trace identifiers are available to every callback.
+- Google Cloud principal and project context is passed when authenticated.
+- Delegation across workers uses shared, one-time verification state.
+
+Configuration alone does not register ADK callbacks, and remote tools remain
+outside the local process boundary.

agenticdome_python_sdk-1.2.13/docs/frameworks/langgraph.md

@@ -0,0 +1,49 @@
+# LangGraph and LangChain integration
+
+LangGraph is graph-native: make security nodes or wrappers part of the graph
+before compilation. For LangChain `create_agent`, install AgenticDome as
+middleware and secure any local side-effecting tools separately.
+
+## Try it without an account
+
+```bash
+pip install "agenticdome-python-sdk[langgraph]"
+export AGENTICDOME_MODE=local_sim
+agenticdome-demo --framework langgraph --scenario both
+```
+
+## Attach in production
+
+```python
+from langgraph.graph import START, StateGraph
+from agenticdome_sdk.langgraph import AgentState, AgenticDomeLangGraphFirewall
+
+firewall = AgenticDomeLangGraphFirewall()
+graph = StateGraph(AgentState)
+graph.add_node("input_security", firewall.input_node(agent_id="support"))
+graph.add_node("agent", agent_node)
+graph.add_node("tool_security", firewall.transition_node(agent_id="support"))
+graph.add_node("tools", tool_node)
+graph.add_node("output_security", firewall.output_node(agent_id="support"))
+graph.add_edge(START, "input_security")
+# Connect input_security -> agent -> tool_security -> tools -> output_security.
+compiled = graph.compile()
+```
+
+Alternatively use `wrap_agent_node()`, `wrap_tool_node()`, or
+`as_langchain_middleware(...)`. Blocked state must route to a non-executing
+security node; it must never fall through to the tool node.
+
+See the [LangGraph API guide](../../README.md#langgraph) for complete graph
+topology, middleware, handoff, retrieval and streaming examples.
+
+## Launch checks
+
+- Security nodes/wrappers are present before `compile()` returns.
+- Every path to a tool node crosses an authorization boundary.
+- Blocked routes terminate without invoking the raw tool.
+- Retrieval is sanitized before it becomes model context.
+- Streaming and final output use the documented output boundary.
+- A delegated specialist verifies the handoff before its tool executes.
+
+An edge or node that calls a raw executor directly is outside SDK coverage.

agenticdome_python_sdk-1.2.13/docs/frameworks/llamaindex.md

@@ -0,0 +1,52 @@
+# LlamaIndex integration
+
+Protect both sides of a RAG application: query/tool execution and the retrieved
+or synthesized content that is inserted into model context or returned to a
+caller.
+
+## Try it without an account
+
+```bash
+pip install "agenticdome-python-sdk[llamaindex]"
+export AGENTICDOME_MODE=local_sim
+agenticdome-demo --framework llamaindex --scenario both
+```
+
+## Attach in production
+
+```python
+from agenticdome_sdk.llamaindex import AgenticDomeLlamaIndexFirewall
+
+firewall = AgenticDomeLlamaIndexFirewall()
+secure_lookup = firewall.to_function_tool(
+    lookup_customer,
+    tool_name="crm.customer.read",
+    tool_platform="crm",
+    agent_id="support-agent",
+    session_id="stable-session-id",
+)
+
+secure_query_engine = firewall.wrap_query_engine(
+    query_engine,
+    agent_id="support-agent",
+    session_id="stable-session-id",
+)
+```
+
+Also wrap retrievers or install the node postprocessor before retrieved text is
+placed into a prompt. Callback handlers add visibility but do not replace hard
+tool/query wrappers.
+
+See the [LlamaIndex API guide](../../README.md#llamaindex) for FunctionTool,
+query, retrieval, node postprocessor, callback and delegation examples.
+
+## Launch checks
+
+- Every side-effecting FunctionTool is created from a secured callable.
+- Query engines invoked directly are wrapped or called through `run_query_securely()`.
+- Retrieved nodes are sanitized before model-context assembly.
+- Final synthesized output is reviewed before return or storage.
+- Callback visibility is not treated as the sole execution control.
+
+Provider-native and remote tools require protection at their local request and
+response boundary.