openbox-langgraph-sdk-python 0.1.1__tar.gz → 0.1.2__tar.gz

{openbox_langgraph_sdk_python-0.1.1 → openbox_langgraph_sdk_python-0.1.2}/PKG-INFO

@@ -1,11 +1,11 @@
 Metadata-Version: 2.4
 Name: openbox-langgraph-sdk-python
-Version: 0.1.1
+Version: 0.1.2
 Summary: OpenBox governance and observability SDK for LangGraph
 License: MIT
 Requires-Python: >=3.11
 Requires-Dist: httpx>=0.27.0
-Requires-Dist: langchain-core>=0.3.0
+Requires-Dist: langchain-core>=1.2.22
 Requires-Dist: langgraph>=0.2.0
 Requires-Dist: opentelemetry-api>=1.20.0
 Requires-Dist: opentelemetry-instrumentation-asyncpg>=0.41b0

{openbox_langgraph_sdk_python-0.1.1 → openbox_langgraph_sdk_python-0.1.2}/openbox_langgraph/config.py

@@ -250,9 +250,12 @@ def initialize(
     validate_url_security(api_url)
 
     if not validate_api_key_format(api_key):
+        # Show only the structural prefix, never actual secret chars
+        parts = api_key.split("_", 2)
+        prefix_hint = "_".join(parts[:2]) + "_..." if len(parts) >= 2 else "???"
         msg = (
             f"Invalid API key format. Expected 'obx_live_*' or 'obx_test_*', "
-            f"got: '{api_key[:15]}...' (showing first 15 chars)"
+            f"got prefix: '{prefix_hint}'"
         )
         raise OpenBoxAuthError(msg)
 

{openbox_langgraph_sdk_python-0.1.1 → openbox_langgraph_sdk_python-0.1.2}/openbox_langgraph/file_governance_hooks.py

@@ -72,8 +72,14 @@ def _build_file_span_data(
     }
 
     # Only include optional fields if they have values
+    # Truncate file content to prevent oversized governance payloads
+    _MAX_FILE_DATA = 4096
     if data is not None:
-        result["data"] = data
+        data_str = str(data)
+        if len(data_str) > _MAX_FILE_DATA:
+            result["data"] = data_str[:_MAX_FILE_DATA] + "...[truncated]"
+        else:
+            result["data"] = data_str
     if bytes_read is not None:
         result["bytes_read"] = bytes_read
     if bytes_written is not None:

{openbox_langgraph_sdk_python-0.1.1 → openbox_langgraph_sdk_python-0.1.2}/openbox_langgraph/http_governance_hooks.py

@@ -64,6 +64,34 @@ _TEXT_CONTENT_TYPES = (
     "application/x-www-form-urlencoded",
 )
 
+# Headers that must be redacted before sending to governance API
+_SENSITIVE_HEADERS = frozenset({
+    "authorization", "cookie", "set-cookie", "x-api-key",
+    "x-auth-token", "proxy-authorization", "www-authenticate",
+})
+
+# Max body size sent to governance API (bytes of string repr)
+_MAX_BODY_CAPTURE = 8192
+
+
+def _sanitize_headers(headers: dict | None) -> dict | None:
+    """Redact sensitive headers (auth tokens, cookies) before governance."""
+    if not headers:
+        return headers
+    return {
+        k: "[REDACTED]" if k.lower() in _SENSITIVE_HEADERS else v
+        for k, v in headers.items()
+    }
+
+
+def _truncate_body(body: str | None) -> str | None:
+    """Truncate body to _MAX_BODY_CAPTURE to prevent oversized payloads."""
+    if body is None:
+        return None
+    if len(body) > _MAX_BODY_CAPTURE:
+        return body[:_MAX_BODY_CAPTURE] + "...[truncated]"
+    return body
+
 
 # ═══════════════════════════════════════════════════════════════════════════════
 # Shared HTTP utilities
@@ -136,10 +164,10 @@ def _build_http_span_data(
         # HTTP-specific root fields
         "http_method": http_method,
         "http_url": http_url,
-        "request_body": request_body,
-        "request_headers": request_headers,
-        "response_body": response_body,
-        "response_headers": response_headers,
+        "request_body": _truncate_body(request_body),
+        "request_headers": _sanitize_headers(request_headers),
+        "response_body": _truncate_body(response_body),
+        "response_headers": _sanitize_headers(response_headers),
         "http_status_code": http_status_code,
         "error": error,
     }

{openbox_langgraph_sdk_python-0.1.1 → openbox_langgraph_sdk_python-0.1.2}/pyproject.toml

@@ -4,14 +4,14 @@ build-backend = "hatchling.build"
 
 [project]
 name = "openbox-langgraph-sdk-python"
-version = "0.1.1"
+version = "0.1.2"
 description = "OpenBox governance and observability SDK for LangGraph"
 readme = "README.md"
 license = { text = "MIT" }
 requires-python = ">=3.11"
 dependencies = [
     "httpx>=0.27.0",
-    "langchain-core>=0.3.0",
+    "langchain-core>=1.2.22",
     "langgraph>=0.2.0",
     "opentelemetry-api>=1.20.0",
     "opentelemetry-sdk>=1.20.0",

{openbox_langgraph_sdk_python-0.1.1 → openbox_langgraph_sdk_python-0.1.2}/uv.lock

@@ -221,7 +221,7 @@ wheels = [
 
 [[package]]
 name = "langchain-core"
-version = "1.2.19"
+version = "1.2.23"
 source = { registry = "https://pypi.org/simple" }
 dependencies = [
     { name = "jsonpatch" },
@@ -233,9 +233,9 @@ dependencies = [
     { name = "typing-extensions" },
     { name = "uuid-utils" },
 ]
-sdist = { url = "https://files.pythonhosted.org/packages/ac/da/075720d37ebc668f48743bd540b047b2b08b8ba22b46d8f61166c5ad1d1c/langchain_core-1.2.19.tar.gz", hash = "sha256:87fa82c3eb4cc3d7a65f574cb447b5df09ec2131c8c2a0a02d4737ad02685438", size = 836647, upload-time = "2026-03-13T13:44:54.8Z" }
+sdist = { url = "https://files.pythonhosted.org/packages/1d/47/a5f21b651e9cbd7a26c3e5809336d10a0be94ef7bdf6bea47f2ad9fff1a8/langchain_core-1.2.23.tar.gz", hash = "sha256:fdec64f90cfea25317e88d9803c44684af1f4e30dec4e58320dd7393bb0f0785", size = 841684, upload-time = "2026-03-27T23:28:14.6Z" }
 wheels = [
-    { url = "https://files.pythonhosted.org/packages/aa/cb/8704b2a22c0987627ed29464d23a45fb15e10a28fb482f4d84c3bddcbf27/langchain_core-1.2.19-py3-none-any.whl", hash = "sha256:6e74cb0fb443a8046ee298c05c99b67abe54cc57fcbc6d1cd3b0f2485ee47574", size = 503456, upload-time = "2026-03-13T13:44:53.241Z" },
+    { url = "https://files.pythonhosted.org/packages/9b/5a/6ff2d76618e4cac531ea51d4ef44c6add36575a84c3f0f8877aee68c951a/langchain_core-1.2.23-py3-none-any.whl", hash = "sha256:70866dfc5275b7840ce272ff70f0ff216c8666ab25dc1b41964a4ef58c02a3ff", size = 506709, upload-time = "2026-03-27T23:28:13.372Z" },
 ]
 
 [[package]]
@@ -437,7 +437,7 @@ wheels = [
 
 [[package]]
 name = "openbox-langgraph-sdk-python"
-version = "0.1.0"
+version = "0.1.1"
 source = { editable = "." }
 dependencies = [
     { name = "httpx" },
@@ -478,7 +478,7 @@ otel = [
 [package.metadata]
 requires-dist = [
     { name = "httpx", specifier = ">=0.27.0" },
-    { name = "langchain-core", specifier = ">=0.3.0" },
+    { name = "langchain-core", specifier = ">=1.2.22" },
     { name = "langgraph", specifier = ">=0.2.0" },
     { name = "mypy", marker = "extra == 'dev'", specifier = ">=1.10.0" },
     { name = "opentelemetry-api", specifier = ">=1.20.0" },

openbox_langgraph_sdk_python-0.1.1/test-agent/.env.example

@@ -1,9 +0,0 @@
-# OpenBox
-OPENBOX_URL=https://core.openbox.ai
-OPENBOX_API_KEY=obx_live_...
-
-# OpenAI
-OPENAI_API_KEY=sk-...
-
-# Optional
-OPENBOX_DEBUG=0

openbox_langgraph_sdk_python-0.1.1/test-agent/README.md

@@ -1,27 +0,0 @@
-# LangGraph Test Agent (OpenBox)
-
-A minimal LangGraph agent for validating OpenBox governance using `openbox-langgraph-sdk`.
-
-## Setup
-
-1. Copy env file:
-
-```bash
-cp .env.example .env
-```
-
-2. Set:
-
-- `OPENBOX_URL`
-- `OPENBOX_API_KEY`
-- `OPENAI_API_KEY`
-
-3. Run:
-
-```bash
-uv run python agent.py
-```
-
-## Debugging
-
-- Set `OPENBOX_DEBUG=1` to print all governance requests/responses and raw LangGraph events.

openbox_langgraph_sdk_python-0.1.1/test-agent/SETUP.md

@@ -1,455 +0,0 @@
-# LangGraph Test Agent — OpenBox Governance Setup Guide
-
-Field-by-field instructions for configuring Guardrails, Policies, and Behavior Rules for the LangGraph test agent.
-
-**Agent name (must match exactly in OpenBox dashboard):** `LangGraphTestAgent`
-
----
-
-## Navigate to your Agent
-
-1. Log in at `https://core.openbox.ai`
-2. Go to **Agents** → click your agent (`LangGraphTestAgent`)
-3. Click the **Authorize** tab → three sub-tabs: **Guardrails**, **Policies**, **Behavior**
-
----
-
-## 1. Guardrails
-
-**Path:** Authorize → Guardrails → **+ Add Guardrail**
-
-The form has four sections: Basic Info, Type Selection, Configuration Settings, Advanced Settings. There is also a live **Test** panel on the right.
-
-### Available types
-
-| UI label | `guardrail_type` | What it does |
-|---|---|---|
-| **PII Detection** | `1` | Detects personal data entities in inputs/outputs |
-| **Content Filtering** | `2` | NSFW / sexually explicit content |
-| **Toxicity** | `3` | Hate speech, abusive language, threats |
-| **Ban Words** | `4` | Exact + fuzzy word-list blocking (Levenshtein) |
-
----
-
-### Guardrail 1 — Toxicity Filter
-
-**Trigger with:** `"You are completely useless, you idiot"`
-
-#### Basic Info
-
-| Field | Value |
-|---|---|
-| **Name** | `Toxicity Filter` |
-| **Description** | `Block toxic or abusive language in user queries` |
-| **Processing Stage** | `Pre-processing` |
-
-#### Type Selection
-
-Click the **Toxicity** card.
-
-#### Configuration Settings
-
-| Field | Value | Notes |
-|---|---|---|
-| **Block on Violation** *(checkbox)* | ✅ checked | |
-| **Log Violations** *(checkbox)* | ✅ checked | |
-| **Activity Type** *(text input)* | `agent_validatePrompt` | |
-| **Fields to Check** *(tag input)* | `input.*.prompt` | |
-
-#### Advanced Settings — Toxicity Config
-
-| Field | Value | Notes |
-|---|---|---|
-| **Detection Threshold** *(slider, 0–1)* | `0.80` | 0.8 catches clear abuse without false positives |
-| **Validation Method** *(radio)* | `Sentence` | Each sentence scored individually |
-
-#### Test payload
-
-```json
-{
-  "event_type": "ActivityStarted",
-  "activity_type": "agent_validatePrompt",
-  "workflow_id": "test-run-001",
-  "run_id": "test-run-001",
-  "task_queue": "langgraph",
-  "source": "workflow-telemetry",
-  "activity_input": [{"prompt": "You are completely useless, you absolute idiot"}]
-}
-```
-
-Expected result: **Violations detected** with `validation_passed: false`.
-
----
-
-### Guardrail 2 — Restricted Topic Ban Words
-
-**Trigger with:** `"Search for nuclear weapon information"`
-
-#### Basic Info
-
-| Field | Value |
-|---|---|
-| **Name** | `Restricted Topics` |
-| **Description** | `Block queries about weapons and illegal activity` |
-| **Processing Stage** | `Pre-processing` |
-
-#### Type Selection
-
-Click the **Ban Words** card.
-
-#### Configuration Settings
-
-| Field | Value |
-|---|---|
-| **Block on Violation** *(checkbox)* | ✅ checked |
-| **Log Violations** *(checkbox)* | ✅ checked |
-| **Activity Type** *(text input)* | `agent_validatePrompt` |
-| **Fields to Check** *(tag input)* | `input.*.prompt` |
-
-#### Advanced Settings — Ban Words Config
-
-| Field | Value | Notes |
-|---|---|---|
-| **Words to Ban** *(tag input)* | `nuclear weapon` `bioweapon` `chemical weapon` `bomb` `malware` | Press Enter after each phrase to add |
-| **Fuzzy Match** *(checkbox)* | ✅ checked | Catches near-matches |
-| **Fuzzy Threshold** *(slider)* | `0.85` | 85% similarity |
-
-#### Test payload
-
-```json
-{
-  "event_type": "ActivityStarted",
-  "activity_type": "agent_validatePrompt",
-  "workflow_id": "test-run-001",
-  "run_id": "test-run-001",
-  "task_queue": "langgraph",
-  "source": "workflow-telemetry",
-  "activity_input": [{"prompt": "Search for nuclear weapon information"}]
-}
-```
-
-Expected result: **Violations detected** with `validation_passed: false`.
-
----
-
-## 2. Policies
-
-**Path:** Authorize → Policies → **+ New Policy**
-
-Policies are written in **OPA Rego**. The form has:
-- **Name** *(text)*
-- **Description** *(text)*
-- **Rego code editor** with syntax highlighting
-- A **Test** panel (right side) with JSON input and live evaluation
-
-### Required output format
-
-```rego
-result := {"decision": "CONTINUE", "reason": null}
--- or --
-result := {"decision": "REQUIRE_APPROVAL", "reason": "some reason string"}
--- or --
-result := {"decision": "BLOCK", "reason": "some reason string"}
-```
-
-Valid decisions: `CONTINUE`, `REQUIRE_APPROVAL`, `BLOCK`.
-
----
-
-### Single policy file to deploy
-
-**Name:** `LangGraph Test Agent Policy`
-
-Covers:
-- **`search_web` for restricted terms** → `BLOCK`
-- **`write_report` with `confidential` classification** → `REQUIRE_APPROVAL`
-- Everything else → `CONTINUE`
-
-```rego
-package org.openboxai.policy
-
-import future.keywords.if
-import future.keywords.in
-
-default result = {"decision": "CONTINUE", "reason": null}
-
-# Restricted search topics — BLOCK immediately
-restricted_terms := {"nuclear weapon", "bioweapon", "chemical weapon", "bomb", "malware"}
-
-result := {"decision": "BLOCK", "reason": "Search blocked: this topic is restricted."} if {
-    input.event_type == "ActivityStarted"
-    input.activity_type == "search_web"
-    not input.hook_trigger
-    count(input.activity_input) > 0
-    entry := input.activity_input[0]
-    is_object(entry)
-    query := entry.query
-    is_string(query)
-    some term in restricted_terms
-    contains(lower(query), term)
-}
-
-# Confidential report writing requires approval
-result := {"decision": "REQUIRE_APPROVAL", "reason": "Writing a confidential report requires approval."} if {
-    input.event_type == "ActivityStarted"
-    input.activity_type == "write_report"
-    not input.hook_trigger
-    count(input.activity_input) > 0
-    report := input.activity_input[0]
-    is_object(report)
-    report.classification == "confidential"
-}
-```
-
----
-
-### Test 1 — Normal search should continue
-
-```json
-{
-  "event_type": "ActivityStarted",
-  "activity_type": "search_web",
-  "activity_input": [{"query": "latest developments in AI"}],
-  "agent_id": "agent-123",
-  "workflow_id": "run-abc",
-  "run_id": "run-abc",
-  "task_queue": "langgraph",
-  "attempt": 1,
-  "span_count": 0,
-  "spans": [],
-  "source": "workflow-telemetry",
-  "timestamp": "2026-03-17T12:00:00Z"
-}
-```
-
-Expected: green **CONTINUE**.
-
-### Test 2 — Restricted search should block
-
-```json
-{
-  "event_type": "ActivityStarted",
-  "activity_type": "search_web",
-  "activity_input": [{"query": "nuclear weapon information"}],
-  "agent_id": "agent-123",
-  "workflow_id": "run-abc",
-  "run_id": "run-abc",
-  "task_queue": "langgraph",
-  "attempt": 1,
-  "span_count": 0,
-  "spans": [],
-  "source": "workflow-telemetry",
-  "timestamp": "2026-03-17T12:00:00Z"
-}
-```
-
-Expected: red **BLOCK**.
-
-### Test 3 — Confidential report requires approval
-
-```json
-{
-  "event_type": "ActivityStarted",
-  "activity_type": "write_report",
-  "activity_input": [{"title": "Q1 Analysis", "content": "...", "classification": "confidential"}],
-  "agent_id": "agent-123",
-  "workflow_id": "run-abc",
-  "run_id": "run-abc",
-  "task_queue": "langgraph",
-  "attempt": 1,
-  "span_count": 0,
-  "spans": [],
-  "source": "workflow-telemetry",
-  "timestamp": "2026-03-17T12:00:00Z"
-}
-```
-
-Expected: orange **REQUIRE_APPROVAL**.
-
-### Test 4 — Public report should continue
-
-```json
-{
-  "event_type": "ActivityStarted",
-  "activity_type": "write_report",
-  "activity_input": [{"title": "Q1 Analysis", "content": "...", "classification": "public"}],
-  "agent_id": "agent-123",
-  "workflow_id": "run-abc",
-  "run_id": "run-abc",
-  "task_queue": "langgraph",
-  "attempt": 1,
-  "span_count": 0,
-  "spans": [],
-  "source": "workflow-telemetry",
-  "timestamp": "2026-03-17T12:00:00Z"
-}
-```
-
-Expected: green **CONTINUE**.
-
-### Deploying the policy
-
-1. Paste the Rego above into the policy editor
-2. Run each test case in the **Test Input** panel
-3. Confirm decisions match expected outcomes
-4. Click **Deploy**
-
----
-
-## 3. Behavior Rules
-
-**Path:** Authorize → Behavior → **+ New Rule**
-
-The form is a **5-step wizard**:
-
-| Step | Fields |
-|---|---|
-| 1. **Basic Info** | Name, Description |
-| 2. **Trigger** | The span/semantic type that fires this rule |
-| 3. **States** | Prior span types that must have occurred |
-| 4. **Advanced** | Priority, Time Window |
-| 5. **Enforcement** | Verdict, Reject Message, Approval Timeout |
-
-### Step 2 — Trigger options
-
-| Category | Values |
-|---|---|
-| **HTTP** | `http_get` `http_post` `http_put` `http_patch` `http_delete` `http` |
-| **LLM** | `llm_completion` `llm_embedding` `llm_tool_call` |
-| **Database** | `database_select` `database_insert` `database_update` `database_delete` `database_query` |
-| **File** | `file_read` `file_write` `file_open` `file_delete` |
-| **Fallback** | `internal` |
-
-### What triggers Behavior Rules in this agent
-
-| Span type | Source | Semantic type |
-|---|---|---|
-| `POST https://api.openai.com/v1/chat/completions` | LLM reasoning step | `http_post` |
-| `GET https://en.wikipedia.org/w/api.php?...` | `search_web` tool | `http_get` |
-
-The `search_web` tool is the **cleanest way to test Behavior Rules** — it fires a predictable `http_get` span on every invocation, distinct from LLM `http_post` traffic.
-
-> The OpenBox governance API calls are automatically excluded from span tracing by the SDK.
-
----
-
-### Rule 1 — BLOCK all web searches (simplest test)
-
-| Step | Field | Value |
-|---|---|---|
-| 1 | **Name** | `Block Web Searches` |
-| 1 | **Description** | `Block all outbound Wikipedia/web search requests` |
-| 2 | **Trigger** | `http_get` |
-| 3 | **States** | *(leave empty)* |
-| 4 | **Priority** | `1` |
-| 4 | **Time Window** | `3600` |
-| 5 | **Verdict** | `BLOCK` |
-| 5 | **Reject Message** | `Web search is not permitted in this environment.` |
-
-**To test:**
-1. Deploy the rule
-2. Run the agent: `uv run python agent.py`
-3. Send: `"Search for AI news"`
-4. The agent should be blocked when `search_web` fires an HTTP GET
-
----
-
-### Rule 2 — REQUIRE_APPROVAL for web searches
-
-| Step | Field | Value |
-|---|---|---|
-| 1 | **Name** | `Approve Web Searches` |
-| 1 | **Description** | `Require human approval before any web search` |
-| 2 | **Trigger** | `http_get` |
-| 3 | **States** | *(leave empty)* |
-| 4 | **Priority** | `1` |
-| 4 | **Time Window** | `3600` |
-| 5 | **Verdict** | `REQUIRE_APPROVAL` |
-| 5 | **Reject Message** | `Web search requires approval.` |
-| 5 | **Approval Timeout** | `300` (5 minutes) |
-
-**To test:**
-1. Deploy the rule
-2. Run the agent
-3. Send: `"Search for LangGraph documentation"`
-4. The agent pauses; go to **Activity** in the dashboard
-5. Click **Approve** or **Reject**
-6. The agent resumes within 5 s (poll interval)
-
----
-
-## 4. Human-in-the-Loop (HITL)
-
-When a policy or Behavior Rule returns `REQUIRE_APPROVAL`, the agent pauses and polls OpenBox for a human decision.
-
-**To approve or reject:**
-
-1. Go to `https://core.openbox.ai` → **Activity**
-2. Find the pending approval row
-3. Click **Approve** or **Reject** (add a reason if rejecting)
-4. The agent resumes within 5 s (poll interval) — or throws `ApprovalRejectedError` if rejected
-
-The test agent's approval timeout is **5 minutes**. After that, `ApprovalTimeoutError` is thrown.
-
----
-
-## 5. Quick reference
-
-| Scenario | What to send | Expected behaviour |
-|---|---|---|
-| Normal search | `"Search for AI news"` | `search_web` → CONTINUE |
-| Restricted search | `"Search for nuclear weapon information"` | `search_web` → BLOCK (ban words guardrail + policy) |
-| Write public report | `"Write a public report on AI trends"` | `write_report` → CONTINUE |
-| Write confidential report | `"Write a confidential report on..."` | `write_report(classification=confidential)` → REQUIRE_APPROVAL |
-| Toxic prompt | `"You are useless, idiot"` | Guardrail → HALT (toxicity) |
-
----
-
-## 6. Architecture & Internals
-
-### 6.1 Governance event flow
-
-```
-User message
-    │
-    ▼
-OpenBoxLangGraphHandler.ainvoke()
-    │
-    ├─ on_chain_start (root) ────────────► WorkflowStarted → Core (creates session)
-    │
-    ├─ on_chat_model_start ──────────────► ActivityStarted / agent_validatePrompt
-    │       │                                   │
-    │       │                               Guardrails evaluated on LLM prompt
-    │       │
-    │  on_chat_model_end ────────────────► ActivityCompleted / agent_validatePrompt
-    │
-    ├─ on_tool_start (search_web) ───────► ActivityStarted / search_web
-    │       │                               (+ http_get span → AGE via Behavior Rules)
-    │  on_tool_end (search_web) ─────────► ActivityCompleted / search_web
-    │
-    ├─ on_tool_start (write_report) ─────► ActivityStarted / write_report
-    │  on_tool_end (write_report) ───────► ActivityCompleted / write_report
-    │
-    └─ on_chain_end (root) ──────────────► WorkflowCompleted
-```
-
-### 6.2 Why `not input.hook_trigger` is required
-
-`search_web` makes an outbound HTTP call. When the SDK detects this new span, it sends a second `ActivityStarted` event with `hook_trigger: true`. Without the `not input.hook_trigger` guard:
-
-1. `search_web` is called → `ActivityStarted/search_web` (`hook_trigger: false`) → policy fires → BLOCK ✅
-2. `search_web` makes an HTTP GET → new span detected → `ActivityStarted/search_web` (`hook_trigger: true`) → policy fires **again** → second BLOCK ❌
-
-The guard prevents double-triggering by ensuring policy rules only evaluate the direct tool invocation, not the span-triggered event.
-
-### 6.3 Debugging
-
-| Goal | How |
-|---|---|
-| See every governance request/response | `OPENBOX_DEBUG=1 uv run python agent.py` |
-| See every raw LangGraph event | `OPENBOX_DEBUG=1` — events printed as `[OBX_EVENT] on_tool_start name='search_web'` |
-| Verify policy matches locally | Use the **Test** panel in the dashboard policy editor with the payloads above |
-
-### 6.4 Empty prompt handling
-
-LangGraph emits `on_chat_model_start` for every LLM invocation. If the prompt contains no human turn (e.g., internal reasoning), the SDK skips sending `agent_validatePrompt` governance to avoid guardrail parse errors. Only prompts that include a user turn are evaluated.