integrating-with-lauren
Skill0 starsChanged 6 months agoMITView on GitHub
Wires lauren-ai into a lauren web framework application. Use when setting up LLMModule/AgentModule, streaming agent responses as SSE, securing tools with ExecutionContext, routing SignalBus events to WebSocket clients, or injecting multiple runners via AgentRunner[X].
- Reads credentials
In this repository65 other files it gives its agents.
- AGENTS.mdAGENTS.md
- skills/building-agents/agents.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- llms-full.txtllms.txt
- llms.txtllms.txt
- adding-guardrailsSkill
- agent-health-timeoutSkill
- agent-permission-scopingSkill
- agent-session-isolationSkill
- agent-step-loggingSkill
- agent-supervisor-routerSkill
- agent-telemetrySkill
- async-agent-backgroundSkill
- building-agentsSkill
- building-teamsSkill
- building-toolsSkill
- calendar-scheduling-toolSkill
- callback-tracing-handlerSkill
- chain-of-thought-promptingSkill
- chat-model-configurationSkill
- code-execution-sandboxSkill
- common-agent-patternsSkill
- context-window-managementSkill
- conversation-memorySkill
- conversation-state-dbSkill
- database-query-toolSkill
- document-loadersSkill
- document-rerankingSkill
- email-notification-toolSkill
- embedding-model-ingestionSkill
- fallback-model-chainSkill
- few-shot-injectionSkill
- filesystem-toolSkill
- human-in-the-loopSkill
- hybrid-searchSkill
- inspecting-streamsSkill
- json-schema-enforcementSkill
- llm-provider-connectionSkill
- llm-rate-limitingSkill
- long-term-vector-memorySkill
- managing-memorySkill
- migrating-to-lauren-aiSkill
- multi-agent-delegationSkill
- multi-tool-orchestrationSkill
- multi-turn-tool-stateSkill
- output-guardrailsSkill
- plan-and-execute-agentSkill
- prompt-injection-defenseSkill
- prompt-versioningSkill
- rag-pipelineSkill
- react-agent-loopSkill
- response-validation-retrySkill
- rest-api-toolSkill
- securing-agentsSkill
- semantic-cacheSkill
- streaming-response-handlerSkill
- structured-output-parsingSkill
- sub-agent-spawningSkill
- system-prompt-templatingSkill
- testing-agentsSkill
- The first 60 shown.
The file
We split this skill into the rules it gives an agent
1 rule. About 0 are common, 0 are distinctive, and 1 is very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈0
Distinctive · estimate
This author's own choices. Worth a look.
1
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 198
Every AgentModule.for_root() call has its own runner.
≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Use
codemap find "SymbolName"to locate any symbol before reading — it gives exact file + line range and is faster than grep across the whole repo.
Integrating lauren-ai with the lauren Framework
Module wiring
Step 1 — LLMModule
from lauren_ai import LLMConfig, LLMModule
LLMProvider = LLMModule.for_root(
LLMConfig.for_anthropic(model="claude-opus-4-6") # reads ANTHROPIC_API_KEY
# or: LLMConfig.for_openai(model="openai/gpt-4o-mini") # reads OPENAI_API_KEY
)LLMModule.for_root() registers LLMService, EmbedService, LLMConfig, and the
transport layer as singletons. Pass it to the parent @module's imports=.
Step 2 — AgentModule
from lauren_ai import AgentModule, SignalBus
signal_bus = SignalBus() # shared singleton — keep in a dedicated module
AgentProvider = AgentModule.for_root(
agents=[BankingCRMAgent, BankingTransferAgent],
tools=[GetBalanceTool, TransferFundsTool, DelegateToBankingTransfer],
imports=[LLMProvider],
signals=signal_bus,
# No conversation_store here — set per-agent on @agent(conversation_store=...)
)AgentModule.for_root() auto-registers each agent class and tool class as a
singleton provider and generates a unique AgentRunnerBase subclass as the module's
runner token. It also registers AgentRunner[agent_cls] for every agent in
agents=, giving controllers a per-agent DI token without any boilerplate subclass.
Inject with runner: AgentRunner (single-module scope) or with the parameterized
form runner: AgentRunner[MyAgent] (multi-module scope).
Step 3 — parent @module
from lauren import module
@module(
imports=[AgentProvider, WsModule],
controllers=[BankingChatController, MetricsController],
providers=[EventForwarder, TransferAgentRunner], # wiring singletons
exports=[AgentProvider],
)
class AIModule: ...Any class in providers= that subscribes to SignalBus events or wires runner
references in its __init__ must appear here so Lauren's lifecycle scheduler
instantiates it eagerly at startup.
Streaming agent responses as SSE
from lauren import controller, post, EventStream, ServerSentEvent, use_guards
from lauren.types import ExecutionContext
from lauren_ai import AgentRunner
from app.ws.context import current_user_id
@use_guards(SignatureGuard)
@controller("/api/banking")
class BankingChatController:
def __init__(self, runner: AgentRunner, agent: BankingCRMAgent) -> None:
self._runner = runner
self._agent = agent
@post("/chat")
async def stream(self, body: Json[ChatRequest], exec_ctx: ExecutionContext) -> EventStream:
request = exec_ctx.request
user_id = request.state.get("user_id") or body.user_id
async def generate():
# Set ContextVar BEFORE runner.run() so signal handlers emitted
# during the agent loop can route events to the right WebSocket.
# Do NOT call ContextVar.reset() — each SSE keep-alive chunk runs
# in its own asyncio.Task with a copied context; reset() across
# task boundaries raises ValueError.
current_user_id.set(user_id)
try:
response = await self._runner.run(
self._agent,
body.messages[-1].content,
conversation_id=body.conversation_id,
execution_context=exec_ctx, # ← security anchor
)
for i in range(0, len(response.content), 40):
yield ServerSentEvent(event="token", data=response.content[i:i+40])
yield ServerSentEvent(event="done", data="")
except Exception as exc:
yield ServerSentEvent(event="error", data=str(exc))
return EventStream(generate(), keep_alive=15.0)ExecutionContext as the security anchor
ExecutionContext flows from the HTTP handler through AgentRunner.run() into every
ToolContext. Identity should always be read from ctx.execution_context.request.state,
never from LLM-supplied parameters.
SignatureGuard.can_activate()
└─ sets request.state.user_id from HMAC-verified payload body
BankingChatController.stream()
└─ exec_ctx received via ExecutionContext injection (no marker needed)
└─ runner.run(..., execution_context=exec_ctx)
└─ AgentContext.execution_context = exec_ctx
└─ ToolContext.execution_context = exec_ctx
└─ TransferFundsTool._auth_uid(ctx) reads
ctx.execution_context.request.state.get("user_id")def _auth_uid(ctx: ToolContext | None) -> str:
if ctx is None:
return ""
ec = ctx.execution_context
if ec is None:
return ""
state = getattr(getattr(ec, "request", None), "state", None)
if state is None:
return ""
return state.get("user_id") or ""Key rule: never add from_user or authenticated_user to a tool's parameter
list — that exposes them in the LLM-visible JSON schema and creates a
prompt-injection surface.
Routing SignalBus events to WebSocket clients
# app/ai/signals.py — shared singleton to avoid circular imports
from lauren_ai import SignalBus
signal_bus: SignalBus = SignalBus()
# app/ws/context.py — ContextVar propagated by asyncio.gather
from contextvars import ContextVar
current_user_id: ContextVar[str | None] = ContextVar("ws_current_user_id", default=None)from lauren import Scope, injectable
from lauren_ai import ModelCallComplete, ToolCallStarted, AgentRunComplete
from app.ai.signals import signal_bus
from app.ws.context import current_user_id
@injectable(scope=Scope.SINGLETON)
class EventForwarder:
def __init__(self, db: BankDatabase) -> None:
self._connections: dict[str, list[WebSocket]] = {}
self._lock = asyncio.Lock()
# Subscribe once at construction — stable for the app's lifetime
signal_bus.on(ModelCallComplete)(self._on_model_complete)
signal_bus.on(ToolCallStarted)(self._on_tool_started)
signal_bus.on(AgentRunComplete)(self._on_run_complete)
async def _on_model_complete(self, event: ModelCallComplete) -> None:
user_id = current_user_id.get() # asyncio.gather copies ContextVar state
if not user_id:
return
await self.send_to_user(user_id, {"type": "token_usage", ...})SignalBus.emit uses asyncio.gather internally, which copies the calling
task's Context (including all ContextVar values) into every spawned coroutine.
Any ContextVar set before runner.run() is therefore visible inside signal
handlers without any extra wiring.
Multi-runner disambiguation — AgentRunner[X] per agent
Every AgentModule.for_root() call has its own runner. When a controller or
delegation tool can see runners from two or more modules, use the parameterized
form AgentRunner[AgentClass] — a cached real subclass used as an unambiguous
DI token. No named AgentRunnerBase subclass needed.
# Transfer module — AgentRunner[BankingTransferAgent] is auto-registered
TransferMod = AgentModule.for_root(
agents=[BankingTransferAgent],
tools=[TransferFundsTool],
imports=[LLMProvider],
signals=signal_bus,
)
# CRM module — imports Transfer module, owns the delegation tool
CRMMod = AgentModule.for_root(
agents=[BankingCRMAgent],
tools=[DelegateToBankingTransfer], # ← delegation tool lives in the calling module
imports=[LLMProvider, TransferMod], # ← makes AgentRunner[BankingTransferAgent] visible
signals=signal_bus,
)The delegation tool uses AgentRunner[BankingTransferAgent] — DI resolves it
unambiguously from the imported scope:
@tool()
class DelegateToBankingTransfer:
"""Delegate a transfer task to the BankingTransferAgent.
Args:
transfer_request: Transfer instructions.
"""
def __init__(
self,
agent: BankingTransferAgent,
runner: AgentRunner[BankingTransferAgent], # ← parameterized — no boilerplate
) -> None:
self._agent = agent
self._runner = runner
async def run(self, ctx: ToolContext, transfer_request: str) -> dict:
response = await self._runner.run(
self._agent,
transfer_request,
execution_context=ctx.execution_context,
)
return {"content": response.content}The controller injects multiple runners the same way:
from lauren_ai import AgentRunner
class BankingChatController:
def __init__(
self,
crm_runner: AgentRunner[BankingCRMAgent],
transfer_runner: AgentRunner[BankingTransferAgent],
crm_agent: BankingCRMAgent,
transfer_agent: BankingTransferAgent,
) -> None: ...Reference files
| File | Contents |
| docs/guides/lauren-integration.md | Full guide: module wiring, SSE, security, delegation, SignalBus |
| ../securing-agents/SKILL.md | ExecutionContext security patterns and the full trust chain |
| ../building-tools/SKILL.md | @tool(), ToolContext, schema generation |
| ../building-agents/SKILL.md | @agent(), lifecycle hooks, @use_tools |
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

