Lucene search
+L

Laravel .env disclosure

🗓️ 30 Jul 2026 05:02:00Reported by KittySploitType 
kittysploit
 kittysploit
🔗 kittysploit.com👁 3 Views

Publicly accessible Laravel .env files may leak database credentials and APP_KEY.

Refs
Code
#!/usr/bin/env python3
# -*- coding: utf-8 -*-
"""Laravel .env sensitive information disclosure."""

import re

from kittysploit import *
from lib.protocols.http.http_client import Http_client


PATHS = (
    "/.env",
    "/.env.bak",
    "/.env.dev",
    "/.env.dev.local",
    "/.env.development.local",
    "/.env.prod",
    "/.env.prod.local",
    "/.env.production",
    "/.env.production.local",
    "/.env.local",
    "/.env.example",
    "/.env.stage",
    "/.env.live",
    "/.env.backup",
    "/.env.save",
    "/.env.old",
    "/.env.www",
    "/.env_1",
    "/.env_sample",
    "/api/.env",
)

_APP_RE = re.compile(r"(?mi)^APP_(NAME|ENV|KEY|DEBUG|URL|PASSWORD)=")
_DB_RE = re.compile(r"(?mi)^DB_(HOST|PASSWORD|DATABASE)=")


class Module(Scanner, Http_client):
    __info__ = {
        "name": "Laravel .env disclosure",
        "description": (
            "Detects publicly accessible Laravel .env files that may leak database "
            "credentials, APP_KEY, and other secrets."
        ),
        "author": ["KittySploit Team"],
        "severity": "high",
        "references": [
            "https://laravel.com/docs/master/configuration#environment-configuration",
        ],
        "tags": ["web", "scanner", "exposure", "laravel", "config", "env", "vuln"],
        "agent": {
            "risk": "active",
            "effects": ["network_probe"],
            "expected_requests": 20,
            "reversible": True,
            "approval_required": False,
            "produces": ["tech_hints", "risk_signals", "endpoints"],
            "cost": 1.0,
            "noise": 0.4,
            "value": 1.4,
            "requires": {
                "min_endpoints": 0,
                "min_params": 0,
                "tech_hints_any": [],
                "tech_hints_all": [],
                "specializations_any": [],
                "risk_signals_any": [],
                "auth_session": False,
                "capabilities_any": [],
                "capabilities_all": [],
                "confidence_min": {},
                "confidence_min_any": {},
                "endpoint_pattern_any": [],
                "param_any": [],
                "api_surface_ready": False,
            },
            "chain": {
                "produces_capabilities": [{"capability": "file_read", "from_detail": ""}],
                "consumes_capabilities": [],
                "option_bindings": {},
                "suggested_followups": [],
            },
        },
    }

    def _url(self, path: str) -> str:
        host = str(getattr(getattr(self, "target", None), "value", getattr(self, "target", "")) or "").strip()
        port = getattr(getattr(self, "port", None), "value", getattr(self, "port", 80))
        ssl = getattr(getattr(self, "ssl", None), "value", getattr(self, "ssl", False))
        scheme = "https" if ssl in (True, "true", "True", 1, "1") else "http"
        try:
            port_i = int(port)
        except (TypeError, ValueError):
            port_i = 443 if scheme == "https" else 80
        if host.startswith(("http://", "https://")):
            return host.rstrip("/") + path
        return f"{scheme}://{host}:{port_i}{path}"

    def run(self):
        host = str(getattr(getattr(self, "target", None), "value", getattr(self, "target", "")) or "").strip()
        extra = []
        if host:
            # Also probe host/domain-derived .env names
            host_clean = host.split("://")[-1].split("/")[0].split(":")[0]
            parts = [p for p in host_clean.split(".") if p]
            if parts:
                extra.append(f"/.env.{parts[0]}")
            if len(parts) >= 2:
                extra.append(f"/.env.{'.'.join(parts[-2:])}")

        for path in list(PATHS) + extra:
            response = self.http_request(method="GET", path=path, allow_redirects=False)
            if not response or response.status_code != 200:
                continue
            ctype = str(response.headers.get("Content-Type") or "").lower()
            if "text/html" in ctype:
                continue
            body = response.text or ""
            if not (_APP_RE.search(body) or _DB_RE.search(body)):
                continue

            keys = []
            if _APP_RE.search(body):
                keys.append("APP_*")
            if _DB_RE.search(body):
                keys.append("DB_*")
            self.report_finding(
                "Laravel .env file exposed",
                severity="high",
                evidence={
                    "url": self._url(path),
                    "status_code": 200,
                    "path": path,
                    "keys_found": keys,
                },
                impact={
                    "summary": "Environment secrets (APP_KEY, DB credentials, tokens) may be disclosed.",
                    "business_risk": "Credential and source-config disclosure",
                },
                remediation={
                    "summary": "Block public access to .env files.",
                    "actions": [
                        "Deny .env* in the web server / reverse proxy",
                        "Move secrets to a secure vault or host env",
                        "Rotate exposed APP_KEY and database credentials",
                    ],
                },
            )
            return True
        return False

Data

Build on a solid foundation with Vulners data

We provide the essential building blocks for cybersecurity solutions with comprehensive, structured, and constantly updated vulnerability and exploits data

Api

Power your application with Vulners API

The Vulners REST API offers reliable, high-performance access to vulnerability intelligence, with 99.9% SLA uptime and CDN-backed data delivery for seamless global access

App

Assess and manage vulnerabilities with Vulners tools

Built on top of Vulners' database and SDK, end-user solutions give security professionals and developers lightweight and powerful tools for vulnerability remediation

30 Jul 2026 05:02Current
5.5Medium risk
Vulners AI Score5.5
3