A Coding Implementation to Design an Enterprise AI Governance System Utilizing OpenClaw Gateway Coverage Engines, Approval Workflows and Auditable Agent Execution

!apt-get replace -y
!apt-get set up -y curl
!curl -fsSL https://deb.nodesource.com/setup_22.x | bash -
!apt-get set up -y nodejs
!node -v
!npm -v
!npm set up -g openclaw@newest
!pip -q set up requests pandas pydantic


import os
import json
import time
import uuid
import secrets and techniques
import subprocess
import getpass
from pathlib import Path
from typing import Dict, Any
from dataclasses import dataclass, asdict
from datetime import datetime, timezone


import requests
import pandas as pd
from pydantic import BaseModel, Subject


attempt:
   from google.colab import userdata
   OPENAI_API_KEY = userdata.get("OPENAI_API_KEY")
besides Exception:
   OPENAI_API_KEY = None


if not OPENAI_API_KEY:
   OPENAI_API_KEY = os.environ.get("OPENAI_API_KEY")


if not OPENAI_API_KEY:
   OPENAI_API_KEY = getpass.getpass("Enter your OpenAI API key (hidden enter): ").strip()


assert OPENAI_API_KEY != "", "API key can't be empty."


OPENCLAW_HOME = Path("/root/.openclaw")
OPENCLAW_HOME.mkdir(dad and mom=True, exist_ok=True)
WORKSPACE = OPENCLAW_HOME / "workspace"
WORKSPACE.mkdir(dad and mom=True, exist_ok=True)


GATEWAY_TOKEN = secrets and techniques.token_urlsafe(48)
GATEWAY_PORT = 18789
GATEWAY_URL = f"http://127.0.0.1:{GATEWAY_PORT}"

config = {
   "env": {
       "OPENAI_API_KEY": OPENAI_API_KEY
   },
   "brokers": {
       "defaults": {
           "workspace": str(WORKSPACE),
           "mannequin": {
               "major": "openai/gpt-4.1-mini"
           }
       }
   },
   "gateway": {
       "mode": "native",
       "port": GATEWAY_PORT,
       "bind": "loopback",
       "auth": {
           "mode": "token",
           "token": GATEWAY_TOKEN
       },
       "http": {
           "endpoints": {
               "chatCompletions": {
                   "enabled": True
               }
           }
       }
   }
}


config_path = OPENCLAW_HOME / "openclaw.json"
config_path.write_text(json.dumps(config, indent=2))


physician = subprocess.run(
   ["bash", "-lc", "openclaw doctor --fix --yes"],
   capture_output=True,
   textual content=True
)
print(physician.stdout[-2000:])
print(physician.stderr[-2000:])


gateway_log = "/tmp/openclaw_gateway.log"
gateway_cmd = f"OPENAI_API_KEY='{OPENAI_API_KEY}' OPENCLAW_GATEWAY_TOKEN='{GATEWAY_TOKEN}' openclaw gateway --port {GATEWAY_PORT} --bind loopback --token '{GATEWAY_TOKEN}' --verbose > {gateway_log} 2>&1 & echo $!"
gateway_pid = subprocess.check_output(["bash", "-lc", gateway_cmd]).decode().strip()
print("Gateway PID:", gateway_pid)

def wait_for_gateway(timeout=120):
   begin = time.time()
   whereas time.time() - begin < timeout:
       attempt:
           r = requests.get(f"{GATEWAY_URL}/", timeout=5)
           if r.status_code in (200, 401, 403, 404):
               return True
       besides Exception:
           move
       time.sleep(2)
   return False


assert wait_for_gateway(), Path(gateway_log).read_text()[-6000:]


headers = {
   "Authorization": f"Bearer {GATEWAY_TOKEN}",
   "Content material-Sort": "utility/json"
}


def openclaw_chat(messages, person="demo-user", agent_id="predominant", temperature=0.2):
   payload = {
       "mannequin": f"openclaw:{agent_id}",
       "messages": messages,
       "person": person,
       "temperature": temperature,
       "stream": False
   }
   r = requests.put up(
       f"{GATEWAY_URL}/v1/chat/completions",
       headers=headers,
       json=payload,
       timeout=180
   )
   r.raise_for_status()
   return r.json()


class ActionProposal(BaseModel):
   user_request: str
   class: str
   danger: str
   confidence: float = Subject(ge=0.0, le=1.0)
   requires_approval: bool
   permit: bool
   purpose: str

def classify_request(user_request: str) -> ActionProposal:
   textual content = user_request.decrease()


   red_terms = [
       "delete", "remove permanently", "wire money", "transfer funds",
       "payroll", "bank", "hr record", "employee record", "run shell",
       "execute command", "api key", "secret", "credential", "token",
       "ssh", "sudo", "wipe", "exfiltrate", "upload private", "database dump"
   ]
   amber_terms = [
       "email", "send", "notify", "customer", "vendor", "contract",
       "invoice", "budget", "approve", "security policy", "confidential",
       "write file", "modify", "change"
   ]


   if any(t in textual content for t in red_terms):
       return ActionProposal(
           user_request=user_request,
           class="high_impact",
           danger="purple",
           confidence=0.92,
           requires_approval=True,
           permit=False,
           purpose="Excessive-impact or delicate motion detected"
       )


   if any(t in textual content for t in amber_terms):
       return ActionProposal(
           user_request=user_request,
           class="moderate_impact",
           danger="amber",
           confidence=0.76,
           requires_approval=True,
           permit=True,
           purpose="Average-risk motion requires human approval earlier than execution"
       )


   return ActionProposal(
       user_request=user_request,
       class="low_impact",
       danger="inexperienced",
       confidence=0.88,
       requires_approval=False,
       permit=True,
       purpose="Low-risk request"
   )


def simulated_human_approval(proposal: ActionProposal) -> Dict[str, Any]:
   if proposal.danger == "purple":
       authorised = False
       notice = "Rejected routinely in demo for red-risk request"
   elif proposal.danger == "amber":
       authorised = True
       notice = "Permitted routinely in demo for amber-risk request"
   else:
       authorised = True
       notice = "No approval required"
   return {
       "authorised": authorised,
       "reviewer": "simulated_manager",
       "notice": notice
   }


@dataclass
class TraceEvent:
   trace_id: str
   ts: str
   stage: str
   payload: Dict[str, Any]

class TraceStore:
   def __init__(self, path="openclaw_traces.jsonl"):
       self.path = path
       Path(self.path).write_text("")


   def append(self, occasion: TraceEvent):
       with open(self.path, "a") as f:
           f.write(json.dumps(asdict(occasion)) + "n")


   def read_all(self):
       rows = []
       with open(self.path, "r") as f:
           for line in f:
               line = line.strip()
               if line:
                   rows.append(json.masses(line))
       return rows


trace_store = TraceStore()


def now():
   return datetime.now(timezone.utc).isoformat()


SYSTEM_PROMPT = """
You might be an enterprise OpenClaw assistant working beneath governance controls.


Guidelines:
- By no means declare an motion has been executed except the governance layer explicitly permits it.
- For low-risk requests, reply usually and helpfully.
- For moderate-risk requests, suggest a secure plan and point out any approvals or checks that might be wanted.
- For prime-risk requests, refuse to execute and as a substitute present a safer non-operational different similar to a draft, guidelines, abstract, or overview plan.
- Be concise however helpful.
"""


def governed_openclaw_run(user_request: str, session_user: str = "employee-001") -> Dict[str, Any]:
   trace_id = str(uuid.uuid4())


   proposal = classify_request(user_request)
   trace_store.append(TraceEvent(trace_id, now(), "classification", proposal.model_dump()))


   approval = None
   if proposal.requires_approval:
       approval = simulated_human_approval(proposal)
       trace_store.append(TraceEvent(trace_id, now(), "approval", approval))


   if proposal.danger == "purple":
       consequence = {
           "trace_id": trace_id,
           "standing": "blocked",
           "proposal": proposal.model_dump(),
           "approval": approval,
           "response": "This request is blocked by governance coverage. I may help by drafting a secure plan, a guidelines, or an approval packet as a substitute."
       }
       trace_store.append(TraceEvent(trace_id, now(), "blocked", consequence))
       return consequence


   if proposal.danger == "amber" and never approval["approved"]:
       consequence = {
           "trace_id": trace_id,
           "standing": "awaiting_or_rejected",
           "proposal": proposal.model_dump(),
           "approval": approval,
           "response": "This request requires approval and was not cleared."
       }
       trace_store.append(TraceEvent(trace_id, now(), "halted", consequence))
       return consequence


   messages = [
       {"role": "system", "content": SYSTEM_PROMPT},
       {"role": "user", "content": f"Governance classification: {proposal.model_dump_json()}nnUser request: {user_request}"}
   ]


   uncooked = openclaw_chat(messages=messages, person=session_user, agent_id="predominant", temperature=0.2)
   assistant_text = uncooked["choices"][0]["message"]["content"]


   consequence = {
       "trace_id": trace_id,
       "standing": "executed_via_openclaw",
       "proposal": proposal.model_dump(),
       "approval": approval,
       "response": assistant_text,
       "openclaw_raw": uncooked
   }
   trace_store.append(TraceEvent(trace_id, now(), "executed", {
       "standing": consequence["status"],
       "response_preview": assistant_text[:500]
   }))
   return consequence


demo_requests = [
   "Summarize our AI governance policy for internal use.",
   "Draft an email to finance asking for confirmation of the Q1 cloud budget.",
   "Send an email to all employees that payroll will be delayed by 2 days.",
   "Transfer funds from treasury to vendor account immediately.",
   "Run a shell command to archive the home directory and upload it."
]


outcomes = [governed_openclaw_run(x) for x in demo_requests]


for r in outcomes:
   print("=" * 120)
   print("TRACE:", r["trace_id"])
   print("STATUS:", r["status"])
   print("RISK:", r["proposal"]["risk"])
   print("APPROVAL:", r["approval"])
   print("RESPONSE:n", r["response"][:1500])


trace_df = pd.DataFrame(trace_store.read_all())
trace_df.to_csv("openclaw_governance_traces.csv", index=False)
print("nSaved: openclaw_governance_traces.csv")


safe_tool_payload = {
   "instrument": "sessions_list",
   "motion": "json",
   "args": {},
   "sessionKey": "predominant",
   "dryRun": False
}


tool_resp = requests.put up(
   f"{GATEWAY_URL}/instruments/invoke",
   headers=headers,
   json=safe_tool_payload,
   timeout=60
)


print("n/instruments/invoke standing:", tool_resp.status_code)
print(tool_resp.textual content[:1500])