diff --git a/.env.example b/.env.example
new file mode 100644
index 0000000..bc4ca55
--- /dev/null
+++ b/.env.example
@@ -0,0 +1,48 @@
+# ============================================================
+# ERP stack configuration — copy to .env and fill in real values.
+# .env is git-ignored; NEVER commit real credentials.
+#   Local dev:      cp .env.example .env   (defaults work as-is)
+#   Production VPS: /srv/erp/.env with STRONG values for every entry —
+#                   docker-compose.prod.yml refuses to boot without them.
+# ============================================================
+
+# --- Public hosts (production; local defaults live in docker-compose.yml) ---
+APP_HOST=aramesh.bennu-api-bahrestan.ir
+API_HOST=api.bennu-api-bahrestan.ir
+AUTH_HOST=casdoor.bennu-api-bahrestan.ir
+S3_HOST=s3.bennu-api-bahrestan.ir
+STORAGE_HOST=storage.bennu-api-bahrestan.ir
+
+# --- Traefik / Let's Encrypt ---
+TRAEFIK_ACME_EMAIL=you@example.com
+
+# --- Environment: "prod" enables the fail-fast startup checks in the API ---
+ENVIRONMENT=prod
+
+# --- PostgreSQL (app + Casdoor share the instance) ---
+POSTGRES_PASSWORD=CHANGE_ME_strong_db_password
+
+# --- MinIO object storage root credentials ---
+MINIO_ROOT_USER=CHANGE_ME_minio_user
+MINIO_ROOT_PASSWORD=CHANGE_ME_long_random_minio_password
+
+# --- Casdoor OIDC application (must match the Casdoor application record) ---
+CASDOOR_CLIENT_SECRET=CHANGE_ME_casdoor_client_secret
+CASDOOR_ADMIN_PASSWORD=CHANGE_ME_casdoor_console_admin_password
+
+# --- CORS: exact browser origins allowed to call the API (comma-separated) ---
+CORS_ORIGINS=https://aramesh.bennu-api-bahrestan.ir
+
+# --- Developer Mode (X-User-* header identity + login panel) ---
+# MUST stay false in production. Only for local debugging without tokens.
+DEV_LOGIN_ENABLED=false
+
+# --- Deploy webhook (Gitea push → rebuild; see infra/deploy-webhook/) ---
+GITEA_WEBHOOK_SECRET=CHANGE_ME_long_random_webhook_secret
+
+# --- Monitoring stack (see monitoring/docker-compose.monitoring.yml) ---
+OPENPANEL_CLIENT_ID=
+OPENPANEL_CLIENT_SECRET=
+SENTRY_DSN_BACKEND=
+SENTRY_DSN_FRONTEND=
+GRAFANA_ADMIN_PASSWORD=CHANGE_ME_grafana_password
diff --git a/.gitignore b/.gitignore
index 6691efa..8f32c1d 100644
--- a/.gitignore
+++ b/.gitignore
@@ -11,9 +11,19 @@ __pycache__/
 venv/
 
 # ---- Env / secrets ----
+# All credentials live in git-ignored .env files (server-side and local).
+# NEVER commit .env or database dumps — see docs/SECRET-MANAGEMENT.md.
+.env
+.env.*
+!*.env.example
+backend/.env
 frontend/.env
 *.local
 
+# ---- Database dumps (contain user rows + signing material) ----
+*.dump
+*.sql.gz
+
 # ---- Local session / backups ----
 .zcode/
 v1.2 - Copy/
diff --git a/README.md b/README.md
index dea85e2..77c54f3 100644
--- a/README.md
+++ b/README.md
@@ -105,8 +105,8 @@ for production builds).
 | What | Where | Credentials |
 |---|---|---|
 | App | http://localhost (via Traefik) | — |
-| Casdoor console | http://localhost:8080 | `admin` / `123` |
-| Seeded app users (org `inventory`) | `razaei` (admin), `mousavi` (مدیر تدارکات), `mohaghegh` (مدیر پروژه), `tester` (مدیر پروژه، پروژه «تست») | password `123456` |
+| Casdoor console | http://localhost:8080 | admin credentials are set when you run the seeder (`CASDOOR_ADMIN_PW`) — never commit them |
+| Seeded app users (org `inventory`) | `razaei` (admin), `mousavi` (مدیر تدارکات), `mohaghegh` (مدیر پروژه), `tester` (مدیر پروژه، پروژه «تست») | password from `CASDOOR_SEED_PASSWORD` (bcrypt-hashed at rest — see `backend/tools/seed_casdoor.py`) |
 
 Run `backend/tools/seed_casdoor.py` once after the first boot to create the
 Casdoor organization, application and users (idempotent). Full details —
@@ -143,7 +143,7 @@ is an SSH tunnel from your machine:
    use the panel):
 
    ```powershell
-   ssh -L 9001:127.0.0.1:9001 root@171.22.25.75
+   ssh -L 9001:127.0.0.1:9001 <user>@<server-host>
    ```
 
    You will also land in a normal SSH session — that is fine; the tunnel
@@ -151,18 +151,17 @@ is an SSH tunnel from your machine:
 
 2. In your browser, open: **http://localhost:9001**
 
-3. Log in with:
-
-   | Username | Password |
-   |---|---|
-   | `aramesh-minio` | `feb10fadb5441019dfc33bf12a9b0f74e7b6c478` |
+3. Log in with the production MinIO root credentials from the server's
+   `/srv/erp/.env` (`MINIO_ROOT_USER` / `MINIO_ROOT_PASSWORD`) — they are
+   **never** written in this repo or its history. See
+   [`docs/SECRET-MANAGEMENT.md`](docs/SECRET-MANAGEMENT.md).
 
 You will see the **`procurement-files`** bucket — the invoices/proformas the
 ERP stores, browsable and downloadable from there.
 
-⚠️ These credentials are **full admin access to the file storage**. Treat
-them as a production secret (this README is inside the repo, so anyone with
-repo access can read them — rotate them on the server if they ever leak).
+⚠️ Those credentials are **full admin access to the file storage**. They live
+only in the server-side `.env`; if one ever leaks, rotate it immediately
+(checklist in `docs/SECRET-MANAGEMENT.md`).
 
 Close the PowerShell window (or `Ctrl+C`) when you are done to drop the
 tunnel.
@@ -262,47 +261,33 @@ docker exec -it procurement-postgres psql -U postgres -d procurement
 
 ---
 
-## Restoring the committed seed data
+## Recovery / seeding
 
-The databases are **not** in git — Docker volumes hold the data. The repo
-ships recovery dumps in `backups/علی محقق/`:
+The databases are **not** in git — Docker volumes hold the data, and
+database dumps must never be committed (they contain user rows and signing
+material). Two supported ways to bring a fresh stack up:
 
-| File | Contents |
-|---|---|
-| `casdoor.dump` | Casdoor schema + console admin (`admin` / `123`), org `inventory`, app `inventory-app`, the JWT signing cert, and the single org user `ali.mohaghegh` |
-| `procurement.dump` | Procurement schema + the single `users` row «علی محقق» with role `admin`; all other tables empty (categories/units re-seed at API startup) |
+1. **Empty boot** — schema v2.6 auto-applies on the first boot of an empty
+   `postgres_data` volume; then run the seeders:
 
-⚠️ This **replaces all data** in the `procurement` and `casdoor` databases.
-On a running stack:
-
-```powershell
-docker compose stop api casdoor
-docker cp "backups/علی محقق/procurement.dump" procurement-postgres:/tmp/
-docker cp "backups/علی محقق/casdoor.dump" procurement-postgres:/tmp/
-docker exec procurement-postgres pg_restore -U postgres -d procurement --clean --if-exists /tmp/procurement.dump
-docker exec procurement-postgres pg_restore -U postgres -d casdoor    --clean --if-exists /tmp/casdoor.dump
-docker compose start casdoor api
-```
-
-Login afterwards: `ali.mohaghegh` / `hulkfuck002` (Casdoor rejects spaces
-in usernames, so «ali mohaghegh» is not a valid login name).
-
-Want demo data instead? Two seeders are still available (they create
-different users — see the Login table above):
-
-1. **App seeder** (realistic demo dataset; keeps the Casdoor org as-is):
    ```powershell
+   # Casdoor org + app + users (prints/generates credentials — do not commit them)
+   docker compose run --rm -e CASDOOR_BASE=http://casdoor:8000 api python tools/seed_casdoor.py
+   # Demo dataset (optional):
    docker exec procurement-api sh -c "cd /srv/backend && python -m tools.seed"
    ```
-2. **Casdoor seeder** (creates org `inventory` + app + users razaei /
-   mousavi / mohaghegh / tester — run once after a fresh volume):
+
+2. **Restore your own dump** — create dumps on the server (kept server-side,
+   e.g. `/srv/erp/backups/`), never in git:
+
    ```powershell
-   python backend/tools/seed_casdoor.py   # from the host, or from the api container
+   docker exec procurement-postgres pg_dump -U postgres -d procurement -Fc > /srv/erp/backups/procurement-$(date +%F).dump
+   docker exec procurement-postgres pg_dump -U postgres -d casdoor    -Fc > /srv/erp/backups/casdoor-$(date +%F).dump
    ```
 
-To create a fresh dump of your own current state: stop `api` + `casdoor`,
-`docker exec procurement-postgres pg_dump -U postgres -d procurement -Fc`
-(+ the `casdoor` DB) into `backups/`, then start them again.
+   Restore with `pg_restore --clean --if-exists` (stop `api` + `casdoor`
+   first — this **replaces all data**). The scheduled backup sidecar
+   (see `docs/OPERATIONS.md`) automates this.
 
 ## Common operations
 
@@ -317,10 +302,11 @@ docker compose down -v              # ⚠️ full reset — wipes DB + MinIO vol
 
 ## Known behavior
 
-- Authentication is Casdoor-based (JWT verified via JWKS): `/auth`, `/users`,
-  `/inventory` and `/pm` are protected (401 without a token, 403 without a role).
-  The procurement CRUD endpoints (quotes, suppliers, materials, …) are still
-  open — dev placeholder, do not expose to the public internet.
+- Authentication is Casdoor-based (JWT verified via JWKS) and **required on
+  every API route** except `/health`, `/health/ready` and `/auth/login`
+  (401 without a token, 403 without a role). The former dev-header fallback
+  (`X-User-Role` / `X-User-Name-B64`) is disabled unless
+  `PROCUREMENT_DEV_LOGIN_ENABLED=true` — never enable it in production.
 - Roles live in the app's `users` table (Casdoor is identity-only): `pm`,
   `procurement_manager`, `admin`. Role changes are **admin-only** (RBAC);
   the procurement manager can only assign projects to project managers.
diff --git a/backend/tools/seed_casdoor.py b/backend/tools/seed_casdoor.py
index b0d8070..301525a 100644
--- a/backend/tools/seed_casdoor.py
+++ b/backend/tools/seed_casdoor.py
@@ -18,24 +18,39 @@ for the in-page login flow.
 import http.cookiejar
 import json
 import os
+import secrets
 import urllib.request
 
+import bcrypt
+
 BASE = os.environ.get("CASDOOR_BASE", "http://localhost:8080")
 ADMIN = os.environ.get("CASDOOR_ADMIN", "admin")
-ADMIN_PW = os.environ.get("CASDOOR_ADMIN_PW", "123")
+# Console admin password — MUST be overridden outside local dev (see
+# docs/SECRET-MANAGEMENT.md); the historical default "123" is gone.
+ADMIN_PW = os.environ.get("CASDOOR_ADMIN_PW", "")
+if not ADMIN_PW:
+    ADMIN_PW = secrets.token_urlsafe(12)
+    print(f"[seed] generated Casdoor console admin password: {ADMIN_PW}")
 
 ORG = "inventory"
 APP = "inventory-app"
 CLIENT_ID = "inventory-app-client"
-CLIENT_SECRET = "inventory-app-secret-2026"
+CLIENT_SECRET = os.environ.get("CASDOOR_CLIENT_SECRET", "")
+if not CLIENT_SECRET:
+    CLIENT_SECRET = secrets.token_urlsafe(24)
+    print(f"[seed] generated Casdoor client secret: {CLIENT_SECRET}")
 REDIRECT_URI = "http://localhost/login"  # Traefik edge (v1.3); unused by the in-page flow
 
+# One password for all seeded dev users — override with CASDOOR_SEED_PASSWORD;
+# stored as a bcrypt hash (passwordType "bcrypt"), never plaintext.
+SEED_PASSWORD = os.environ.get("CASDOOR_SEED_PASSWORD", "123456")
+
 USERS = [
-    # (name, first_name, last_name, email, password)
-    ("razaei", "آقای", "رضایی", "razaei@sample.ir", "123456"),
-    ("mousavi", "آقای", "موسوی", "mousavi@sample.ir", "123456"),
-    ("mohaghegh", "آقای", "محقق", "mohaghegh@sample.ir", "123456"),
-    ("tester", "کاربر", "تست", "tester@sample.ir", "123456"),
+    # (name, first_name, last_name, email)
+    ("razaei", "آقای", "رضایی", "razaei@sample.ir"),
+    ("mousavi", "آقای", "موسوی", "mousavi@sample.ir"),
+    ("mohaghegh", "آقای", "محقق", "mohaghegh@sample.ir"),
+    ("tester", "کاربر", "تست", "tester@sample.ir"),
 ]
 
 
@@ -69,17 +84,21 @@ def _login():
     return jar
 
 
+def _bcrypt(pw: str) -> str:
+    return bcrypt.hashpw(pw.encode("utf-8"), bcrypt.gensalt()).decode("ascii")
+
+
 def main():
     cookies = _login()
     if cookies is None:
         return
 
-    # 1) organization (passwordType "plain" = dev: passwords stored as-is)
+    # 1) organization (bcrypt — passwords are never stored plaintext)
     status, body = _request("POST", "/api/add-organization", {
         "owner": "admin",
         "name": ORG,
         "displayName": "سازمان تدارکات",
-        "passwordType": "plain",
+        "passwordType": "bcrypt",
     }, cookies)
     print("add-organization:", status, body.get("msg", body.get("error", "")))
 
@@ -102,7 +121,8 @@ def main():
     print("add-application:", status, body.get("msg", body.get("error", "")))
 
     # 3) users (roles are NOT stored here — they live in the app's users table)
-    for name, first, last, email, pw in USERS:
+    pw_hash = _bcrypt(SEED_PASSWORD)
+    for name, first, last, email in USERS:
         status, body = _request("POST", "/api/add-user", {
             "owner": ORG,
             "name": name,
@@ -110,21 +130,23 @@ def main():
             "firstName": first,
             "lastName": last,
             "email": email,
-            "password": pw,
-            "passwordType": "plain",
+            "password": pw_hash,
+            "passwordType": "bcrypt",
             "phone": "",
             "type": "normal-user",
         }, cookies)
         print(f"add-user {name}:", status, body.get("msg", body.get("error", "")))
         # Ensure passwordType even for already-existing users (this Casdoor
         # version refuses login when both user and org passwordType are empty).
-        status, body = _request("POST", f"/api/update-user?id={ORG}/{name}&columns=password_type", {
-            "passwordType": "plain",
+        status, body = _request("POST", f"/api/update-user?id={ORG}/{name}&columns=password_type,password", {
+            "passwordType": "bcrypt",
+            "password": pw_hash,
         }, cookies)
         print(f"set-password-type {name}:", status, body.get("msg", body.get("error", "")))
 
-    print("\nDone. Casdoor console: http://localhost:8080 (admin/123)")
-    print("Login in the app: http://localhost:3000/login")
+    print("\nDone. Casdoor console: http://localhost:8080")
+    print(f"Seeded users password: {SEED_PASSWORD} (change via CASDOOR_SEED_PASSWORD)")
+    print("Login in the app: http://localhost/login")
 
 
 if __name__ == "__main__":
diff --git a/backups/علی محقق/casdoor.dump b/backups/علی محقق/casdoor.dump
deleted file mode 100644
index aeac482..0000000
Binary files a/backups/علی محقق/casdoor.dump and /dev/null differ
diff --git a/backups/علی محقق/procurement.dump b/backups/علی محقق/procurement.dump
deleted file mode 100644
index c1a2441..0000000
Binary files a/backups/علی محقق/procurement.dump and /dev/null differ
diff --git a/docker-compose.prod.yml b/docker-compose.prod.yml
index 93c2a46..0a2d83a 100644
--- a/docker-compose.prod.yml
+++ b/docker-compose.prod.yml
@@ -1,32 +1,72 @@
 # ============================================================================
 # Production overlay — real domains + HTTPS (Let's Encrypt) + HTTP→HTTPS
-# redirect. Usage:
+# redirect + security middlewares + required secrets.
 #
-#   APP_HOST=app.example.com API_HOST=api.example.com AUTH_HOST=auth.example.com \
-#   S3_HOST=s3.example.com STORAGE_HOST=storage.example.com \
-#   TRAEFIK_ACME_EMAIL=admin@example.com \
+# All credentials come from the git-ignored .env NEXT TO THE COMPOSE FILES
+# (compose loads it automatically — keep it on the VPS, never in git):
+#
+#   cp .env.example .env   # then fill in STRONG values
 #   docker compose -f docker-compose.yml -f docker-compose.prod.yml up -d --build
 #
 # Requirements: DNS records for every host pointing at the server, ports
-# 80 + 443 open (ACME http-challenge uses port 80), and the api's
-# PROCUREMENT_CASDOOR_ISSUER must be updated to the public auth host
-# (see docs/casdoor-auth.md).
+# 80 + 443 open (ACME http-challenge uses port 80). The api's issuer is
+# derived here from AUTH_HOST (see docs/casdoor-auth.md).
+#
+# Any `${VAR:?}` below refuses to boot when the variable is missing/empty —
+# that is deliberate: production must not run on defaults.
 # ============================================================================
 
 services:
   traefik:
     environment:
       # Certificates persist across restarts (acme.json in the traefik_certs volume).
-      TRAEFIK_ACME_EMAIL: ${TRAEFIK_ACME_EMAIL:?set TRAEFIK_ACME_EMAIL for production}
+      TRAEFIK_ACME_EMAIL: ${TRAEFIK_ACME_EMAIL:?set TRAEFIK_ACME_EMAIL in .env}
+
+  postgres:
+    environment:
+      POSTGRES_PASSWORD: ${POSTGRES_PASSWORD:?set POSTGRES_PASSWORD in .env}
+
+  minio:
+    environment:
+      MINIO_ROOT_USER: ${MINIO_ROOT_USER:?set MINIO_ROOT_USER in .env}
+      MINIO_ROOT_PASSWORD: ${MINIO_ROOT_PASSWORD:?set MINIO_ROOT_PASSWORD in .env}
+    labels:
+      - traefik.http.routers.minio-api.rule=Host(`${S3_HOST:?set S3_HOST}`)
+      - traefik.http.routers.minio-api.entrypoints=websecure
+      - traefik.http.routers.minio-api.tls=true
+      - traefik.http.routers.minio-api.tls.certresolver=letsencrypt
+      - traefik.http.routers.minio-api.middlewares=secure-headers@file
+      - traefik.http.routers.minio-api-redir.rule=Host(`${S3_HOST:?set S3_HOST}`)
+      - traefik.http.routers.minio-api-redir.entrypoints=web
+      - traefik.http.routers.minio-api-redir.middlewares=redirect-https@file
+      - traefik.http.routers.minio-console.rule=Host(`${STORAGE_HOST:?set STORAGE_HOST}`)
+      - traefik.http.routers.minio-console.entrypoints=websecure
+      - traefik.http.routers.minio-console.tls=true
+      - traefik.http.routers.minio-console.tls.certresolver=letsencrypt
+      - traefik.http.routers.minio-console.middlewares=secure-headers@file
+      - traefik.http.routers.minio-console-redir.rule=Host(`${STORAGE_HOST:?set STORAGE_HOST}`)
+      - traefik.http.routers.minio-console-redir.entrypoints=web
+      - traefik.http.routers.minio-console-redir.middlewares=redirect-https@file
 
   api:
     ports: [] # production: only reachable through Traefik
+    environment:
+      PROCUREMENT_DATABASE_URL: postgresql+asyncpg://postgres:${POSTGRES_PASSWORD:?set POSTGRES_PASSWORD in .env}@postgres:5432/procurement
+      PROCUREMENT_MINIO_ACCESS_KEY: ${MINIO_ROOT_USER:?set MINIO_ROOT_USER in .env}
+      PROCUREMENT_MINIO_SECRET_KEY: ${MINIO_ROOT_PASSWORD:?set MINIO_ROOT_PASSWORD in .env}
+      PROCUREMENT_CASDOOR_CLIENT_SECRET: ${CASDOOR_CLIENT_SECRET:?set CASDOOR_CLIENT_SECRET in .env}
+      PROCUREMENT_CASDOOR_ADMIN_PASSWORD: ${CASDOOR_ADMIN_PASSWORD:?set CASDOOR_ADMIN_PASSWORD in .env}
+      # Issuer must equal the browser-facing Casdoor origin (its discovery doc).
+      PROCUREMENT_CASDOOR_ISSUER: https://${AUTH_HOST:?set AUTH_HOST}
+      PROCUREMENT_CORS_ORIGINS: ${CORS_ORIGINS:?set CORS_ORIGINS in .env}
+      PROCUREMENT_ENVIRONMENT: prod
+      PROCUREMENT_DEV_LOGIN_ENABLED: "false"
     labels:
       - traefik.http.routers.api.rule=Host(`${API_HOST:?set API_HOST}`)
       - traefik.http.routers.api.entrypoints=websecure
       - traefik.http.routers.api.tls=true
       - traefik.http.routers.api.tls.certresolver=letsencrypt
-      # HTTP → HTTPS
+      - traefik.http.routers.api.middlewares=secure-headers@file,api-ratelimit@file,api-bodylimit@file
       - traefik.http.routers.api-redir.rule=Host(`${API_HOST:?set API_HOST}`)
       - traefik.http.routers.api-redir.entrypoints=web
       - traefik.http.routers.api-redir.middlewares=redirect-https@file
@@ -41,34 +81,21 @@ services:
       - traefik.http.routers.casdoor.entrypoints=websecure
       - traefik.http.routers.casdoor.tls=true
       - traefik.http.routers.casdoor.tls.certresolver=letsencrypt
+      - traefik.http.routers.casdoor.middlewares=secure-headers@file
       - traefik.http.routers.casdoor-redir.rule=Host(`${AUTH_HOST:?set AUTH_HOST}`)
       - traefik.http.routers.casdoor-redir.entrypoints=web
       - traefik.http.routers.casdoor-redir.middlewares=redirect-https@file
 
   frontend:
+    build:
+      args:
+        DEV_LOGIN_ENABLED: "false"
     labels:
       - traefik.http.routers.frontend.rule=Host(`${APP_HOST:?set APP_HOST}`)
       - traefik.http.routers.frontend.entrypoints=websecure
       - traefik.http.routers.frontend.tls=true
       - traefik.http.routers.frontend.tls.certresolver=letsencrypt
+      - traefik.http.routers.frontend.middlewares=secure-headers@file
       - traefik.http.routers.frontend-redir.rule=Host(`${APP_HOST:?set APP_HOST}`)
       - traefik.http.routers.frontend-redir.entrypoints=web
       - traefik.http.routers.frontend-redir.middlewares=redirect-https@file
-
-  minio:
-    ports: [] # production: only reachable through Traefik
-    labels:
-      - traefik.http.routers.minio-api.rule=Host(`${S3_HOST:?set S3_HOST}`)
-      - traefik.http.routers.minio-api.entrypoints=websecure
-      - traefik.http.routers.minio-api.tls=true
-      - traefik.http.routers.minio-api.tls.certresolver=letsencrypt
-      - traefik.http.routers.minio-api-redir.rule=Host(`${S3_HOST:?set S3_HOST}`)
-      - traefik.http.routers.minio-api-redir.entrypoints=web
-      - traefik.http.routers.minio-api-redir.middlewares=redirect-https@file
-      - traefik.http.routers.minio-console.rule=Host(`${STORAGE_HOST:?set STORAGE_HOST}`)
-      - traefik.http.routers.minio-console.entrypoints=websecure
-      - traefik.http.routers.minio-console.tls=true
-      - traefik.http.routers.minio-console.tls.certresolver=letsencrypt
-      - traefik.http.routers.minio-console-redir.rule=Host(`${STORAGE_HOST:?set STORAGE_HOST}`)
-      - traefik.http.routers.minio-console-redir.entrypoints=web
-      - traefik.http.routers.minio-console-redir.middlewares=redirect-https@file
diff --git a/docker-compose.yml b/docker-compose.yml
index af5bd7e..4dc4021 100644
--- a/docker-compose.yml
+++ b/docker-compose.yml
@@ -6,7 +6,9 @@ services:
     environment:
       POSTGRES_DB: procurement
       POSTGRES_USER: postgres
-      POSTGRES_PASSWORD: postgres
+      # Local dev default; production overrides via .env (required by the
+      # prod overlay) — never commit real passwords.
+      POSTGRES_PASSWORD: ${POSTGRES_PASSWORD:-postgres}
 
     ports:
       - "5433:5432"
@@ -69,8 +71,8 @@ services:
     command: server /data --console-address ":9001"
 
     environment:
-      MINIO_ROOT_USER: minioadmin
-      MINIO_ROOT_PASSWORD: minioadmin
+      MINIO_ROOT_USER: ${MINIO_ROOT_USER:-minioadmin}
+      MINIO_ROOT_PASSWORD: ${MINIO_ROOT_PASSWORD:-minioadmin}
 
     ports:
       - "9000:9000"   # S3 API
@@ -110,8 +112,8 @@ services:
 
     environment:
       MINIO_ENDPOINT: minio:9000
-      MINIO_ROOT_USER: minioadmin
-      MINIO_ROOT_PASSWORD: minioadmin
+      MINIO_ROOT_USER: ${MINIO_ROOT_USER:-minioadmin}
+      MINIO_ROOT_PASSWORD: ${MINIO_ROOT_PASSWORD:-minioadmin}
       MINIO_BUCKET: procurement-files
 
     entrypoint: >
@@ -170,24 +172,30 @@ services:
     container_name: procurement-api
 
     environment:
-      PROCUREMENT_DATABASE_URL: postgresql+asyncpg://postgres:postgres@postgres:5432/procurement
+      PROCUREMENT_DATABASE_URL: postgresql+asyncpg://postgres:${POSTGRES_PASSWORD:-postgres}@postgres:5432/procurement
       PROCUREMENT_MINIO_ENDPOINT: minio:9000
-      PROCUREMENT_MINIO_ACCESS_KEY: minioadmin
-      PROCUREMENT_MINIO_SECRET_KEY: minioadmin
+      PROCUREMENT_MINIO_ACCESS_KEY: ${MINIO_ROOT_USER:-minioadmin}
+      PROCUREMENT_MINIO_SECRET_KEY: ${MINIO_ROOT_PASSWORD:-minioadmin}
       PROCUREMENT_MINIO_BUCKET: procurement-files
       PROCUREMENT_MINIO_SECURE: "false"
+      # "dev" enables the local defaults; "prod" turns on the fail-fast
+      # startup checks (no wildcard CORS, no default credentials).
+      PROCUREMENT_ENVIRONMENT: ${ENVIRONMENT:-dev}
+      PROCUREMENT_DEV_LOGIN_ENABLED: ${DEV_LOGIN_ENABLED:-false}
+      # Comma-separated browser origins (JSON-free .env friendly).
+      PROCUREMENT_CORS_ORIGINS: ${CORS_ORIGINS:-http://localhost,http://localhost:3000,http://localhost:5173}
       # Casdoor: endpoint is the in-network service URL; the issuer must match
       # the origin Casdoor advertises to browsers (its OIDC discovery doc).
       PROCUREMENT_CASDOOR_ENDPOINT: http://casdoor:8000
-      PROCUREMENT_CASDOOR_ISSUER: http://localhost:8080
+      PROCUREMENT_CASDOOR_ISSUER: ${CASDOOR_ISSUER:-http://localhost:8080}
       PROCUREMENT_CASDOOR_CLIENT_ID: inventory-app-client
-      PROCUREMENT_CASDOOR_CLIENT_SECRET: inventory-app-secret-2026
+      PROCUREMENT_CASDOOR_CLIENT_SECRET: ${CASDOOR_CLIENT_SECRET:-inventory-app-secret-2026}
       PROCUREMENT_CASDOOR_ORGANIZATION: inventory
       PROCUREMENT_CASDOOR_APP_NAME: inventory-app
       PROCUREMENT_CASDOOR_AUDIENCE: inventory-app-client
       # Console admin used by POST /users/sync (identity mirroring).
       PROCUREMENT_CASDOOR_ADMIN_USER: admin
-      PROCUREMENT_CASDOOR_ADMIN_PASSWORD: "123"
+      PROCUREMENT_CASDOOR_ADMIN_PASSWORD: ${CASDOOR_ADMIN_PASSWORD:-123}
 
     ports:
       - "8000:8000"
@@ -214,8 +222,9 @@ services:
     build:
       context: ./frontend
       args:
-        # Developer Mode panel on the login page — set false for production.
-        DEV_LOGIN_ENABLED: ${DEV_LOGIN_ENABLED:-true}
+        # Developer Mode panel on the login page — dev-only aid. Production
+        # builds MUST leave this false (identity comes from Casdoor tokens).
+        DEV_LOGIN_ENABLED: ${DEV_LOGIN_ENABLED:-false}
     container_name: procurement-frontend
 
     labels:
diff --git a/docs/SECRET-MANAGEMENT.md b/docs/SECRET-MANAGEMENT.md
new file mode 100644
index 0000000..eafd9e6
--- /dev/null
+++ b/docs/SECRET-MANAGEMENT.md
@@ -0,0 +1,67 @@
+# Secret management & rotation
+
+**Policy (short):** real credentials live only in git-ignored `.env` files —
+the server-side `/srv/erp/.env` (used by the deploy webhook and compose) and
+local dev `.env` copies. Nothing secret goes into git: not in the README,
+not in dumps, not in docs. The repo is (or was) **public** — assume anything
+ever committed is compromised and rotate.
+
+`docker-compose.prod.yml` declares every production secret as
+`${VAR:?set in .env}` so the stack **refuses to boot** on defaults.
+
+## Where each secret lives
+
+| Secret | Used by | Defined in |
+|---|---|---|
+| `POSTGRES_PASSWORD` | postgres, api | server `.env` |
+| `MINIO_ROOT_USER` / `MINIO_ROOT_PASSWORD` | minio, minio-init, api | server `.env` |
+| `CASDOOR_CLIENT_SECRET` | Casdoor application record, api | server `.env` (must match the app in the Casdoor console) |
+| `CASDOOR_ADMIN_PASSWORD` | Casdoor console admin, api (`/users/sync`) | server `.env` (rotate in the Casdoor console too) |
+| `CORS_ORIGINS`, `*_HOST`, `TRAEFIK_ACME_EMAIL` | compose / Traefik | server `.env` |
+| `GITEA_WEBHOOK_SECRET` | deploy-webhook receiver (port 9000) | server `.env` + Gitea webhook config |
+| Monitoring DSNs/keys | OpenPanel, GlitchTip, Grafana | `monitoring/.env` (server-side) |
+
+## Rotation checklist — run after any leak suspicion
+
+The production branch history contained real credentials (MinIO admin,
+Casdoor passwords, server IP, DB dumps). All of the following must be
+rotated once, on the VPS:
+
+```bash
+cd /srv/erp
+
+# 1) MinIO root — then update MINIO_ROOT_* in /srv/erp/.env
+mc alias set local http://127.0.0.1:9000 <old-user> '<old-pass>'
+mc admin user add local <new-user> '<new-pass>'      # then:
+mc admin policy attach local consoleAdmin --user <new-user>
+mc admin user remove local <old-user>
+# (simplest equivalent: mc admin user svcacct or change root via env +
+#  `docker compose up -d minio` — MinIO persists root creds from env)
+
+# 2) PostgreSQL — then update POSTGRES_PASSWORD in .env
+docker exec -it procurement-postgres psql -U postgres -c \
+  "ALTER USER postgres PASSWORD '<new-strong-password>';"
+docker compose up -d api   # api re-reads the DSN from .env
+
+# 3) Casdoor console admin (console → Identity → users → admin → reset
+#    password) — then update CASDOOR_ADMIN_PASSWORD in .env
+# 4) Casdoor application client secret (console → Applications →
+#    inventory-app → edit secret) — then update CASDOOR_CLIENT_SECRET in .env
+# 5) Seeded human users: reset each password in the Casdoor console.
+# 6) Gitea webhook secret (Gitea → repo → Settings → Webhooks) — then update
+#    GITEA_WEBHOOK_SECRET in .env and restart the receiver (systemctl restart erp-deploy-webhook)
+
+# 7) Apply everything:
+docker compose -f docker-compose.yml -f docker-compose.prod.yml up -d
+```
+
+- **SSH**: disable password login (`PasswordAuthentication no` in
+  `/etc/ssh/sshd_config`), use keys only, and prefer a non-root user with
+  sudo over `root` direct login.
+- **Repo hygiene**: keep the repo private until the git history is either
+  purged (BFG / `git filter-repo` on the README + `backups/` blobs) or every
+  leaked value is rotated. Deleting a file in a later commit does **not**
+  remove it from history.
+- **Verify**: after rotation, `docker compose logs api | grep -i error`,
+  log into the app, upload + download one attachment, and confirm the
+  webhook deploy still works with a test push.
diff --git a/docs/casdoor-auth.md b/docs/casdoor-auth.md
index 914a5ad..8f5fc27 100644
--- a/docs/casdoor-auth.md
+++ b/docs/casdoor-auth.md
@@ -12,7 +12,7 @@ scenarios, test results, and how to manage the system yourself.
 |---|---|---|
 | App (frontend) | http://localhost (Traefik edge; production: your domain) | — |
 | API (backend) | http://localhost:8000 (Swagger: `/docs`) | — |
-| Casdoor console | http://localhost:8080 | **admin / 123** (built-in org) |
+| Casdoor console | http://localhost:8080 | admin credentials come from `CASDOOR_ADMIN_PW` when running the seeder — never committed |
 | Casdoor OIDC discovery | http://localhost:8080/.well-known/openid-configuration | — |
 | Casdoor public JWKS | http://localhost:8080/.well-known/jwks | — |
 | In-network Casdoor (from the api container) | http://casdoor:8000 | — |
@@ -22,13 +22,13 @@ scenarios, test results, and how to manage the system yourself.
 
 | Username | Password | Casdoor org | App role (our `users` table) |
 |---|---|---|---|
-| `razaei` | `123456` | inventory | `admin` (super user) |
-| `mousavi` | `123456` | inventory | `procurement_manager` (مدیر تدارکات) |
-| `mohaghegh` | `123456` | inventory | `pm` (مدیر پروژه، بدون پروژه) |
-| `tester` | `123456` | inventory | `pm` (مدیر پروژه — پروژه «تست») |
+| `razaei` | `CASDOOR_SEED_PASSWORD` | inventory | `admin` (super user) |
+| `mousavi` | `CASDOOR_SEED_PASSWORD` | inventory | `procurement_manager` (مدیر تدارکات) |
+| `mohaghegh` | `CASDOOR_SEED_PASSWORD` | inventory | `pm` (مدیر پروژه، بدون پروژه) |
+| `tester` | `CASDOOR_SEED_PASSWORD` | inventory | `pm` (مدیر پروژه — پروژه «تست») |
 
-These users are created by `backend/tools/seed_casdoor.py` (idempotent — safe
-to re-run).
+Passwords are stored bcrypt-hashed (passwordType `bcrypt`). These users are
+created by `backend/tools/seed_casdoor.py` (idempotent — safe to re-run).
 
 ---
 
@@ -95,7 +95,7 @@ identity only.** No new authorization model was introduced.
   the OIDC client. Grant types: `authorization_code`, `refresh_token`,
   `password`. Token format JWT (so the backend can verify via JWKS).
 - **Casdoor Organization** (`inventory`): the tenant; every app user belongs
-  to it. (The Casdoor console admin `admin/123` lives in the separate
+  to it. (The Casdoor console admin lives in the separate
   `built-in` org.)
 - **User — two records per person:**
   1. Casdoor user (login credential: name + password).
@@ -158,7 +158,7 @@ After that, log in as the admin and use the Organization panel
 
 | Test | Action | Result | Evidence |
 |---|---|---|---|
-| 1 | Login: `POST /api/v1/auth/login` (tester/123456) | **PASS** | 200, JWT + profile returned |
+| 1 | Login: `POST /api/v1/auth/login` (tester / CASDOOR_SEED_PASSWORD) | **PASS** | 200, JWT + profile returned |
 | 2 | Authenticated API: `GET /users/me` with token | **PASS** | 200, role `pm` + project «تست» |
 | 3 | No token: `GET /inventory/warehouses` | **PASS** | 401 `UNAUTHORIZED` |
 | 4 | Invalid token: corrupted Bearer | **PASS** | 401 `UNAUTHORIZED` «توکن نامعتبر است» |
@@ -179,7 +179,7 @@ After that, log in as the admin and use the Organization panel
 | 6 | Organization exists | ✅ `inventory` |
 | 7 | Application exists | ✅ `inventory-app` |
 | 8 | Client exists | ✅ `inventory-app-client` (grants incl. `password`) |
-| 9 | Test users exist | ✅ razaei, mousavi, mohaghegh, tester (password 123456) |
+| 9 | Test users exist | ✅ razaei, mousavi, mohaghegh, tester (password: CASDOOR_SEED_PASSWORD) |
 | 10 | Redirect URI matches | ✅ N/A for the in-page flow (redirects unused). The seeded app holds the legacy `http://localhost:3000/login`; the seeder now defaults to `http://localhost/login` |
 | 11 | OIDC works | ✅ discovery doc + token endpoint + JWT issuance verified |
 | 12 | PKCE works | ➖ **N/A** — the PKCE/redirect flow was replaced by the in-page password-grant login (per the redesign spec); nothing to verify |
@@ -270,7 +270,7 @@ After that, log in as the admin and use the Organization panel
 
 ## 9. Managing Casdoor yourself
 
-- **Console:** http://localhost:8080 — log in as **admin / 123**. From there
+- **Console:** http://localhost:8080 — log in as the admin (credentials from the seeder run). From there
   you can manage Organizations, Users, Applications, Providers, Certificates.
 - **Seeding** (org `inventory`, app `inventory-app`, the four users) —
   re-runnable at any time:
diff --git a/frontend/Dockerfile b/frontend/Dockerfile
index 7478d7d..53cd9f6 100644
--- a/frontend/Dockerfile
+++ b/frontend/Dockerfile
@@ -13,9 +13,10 @@ ARG NPM_REGISTRY=https://registry.npmmirror.com
 COPY package.json package-lock.json ./
 RUN npm config set registry ${NPM_REGISTRY} && npm ci
 
-# Developer Mode panel on the login page (default enabled; set to
-# "false" to build a production login without the dev panel).
-ARG DEV_LOGIN_ENABLED=true
+# Developer Mode panel on the login page (dev-only aid). Production builds
+# MUST keep this "false" — identity comes from Casdoor tokens, and the
+# backend rejects the X-User-* header path unless DEV_LOGIN_ENABLED=true.
+ARG DEV_LOGIN_ENABLED=false
 ENV VITE_ENABLE_DEV_LOGIN=${DEV_LOGIN_ENABLED}
 
 # Source code + build
diff --git a/infra/deploy-webhook/README.md b/infra/deploy-webhook/README.md
new file mode 100644
index 0000000..fb84cdd
--- /dev/null
+++ b/infra/deploy-webhook/README.md
@@ -0,0 +1,42 @@
+# Deploy webhook receiver (push-to-deploy, hardened)
+
+Every push to the `production` branch triggers a rebuild on the VPS through
+this receiver (host port 9000). The receiver must **verify the Gitea
+webhook signature** — an open deploy endpoint is effectively remote code
+execution on the production server.
+
+## VPS setup (one time)
+
+```bash
+# 1) Files live here:
+sudo mkdir -p /srv/erp && sudo chown $USER /srv/erp
+#    copy the repo to /srv/erp (the receiver runs docker compose in this dir)
+
+# 2) Secrets — add to /srv/erp/.env (git-ignored):
+#    GITEA_WEBHOOK_SECRET=<long random string>
+
+# 3) Install as a systemd service:
+sudo cp infra/deploy-webhook/erp-deploy-webhook.service /etc/systemd/system/
+#    adjust User= and paths inside, then:
+sudo systemctl daemon-reload
+sudo systemctl enable --now erp-deploy-webhook
+curl http://127.0.0.1:9000/ping   # → {"status":"ok"}
+```
+
+## Gitea webhook
+
+Repo → Settings → Webhooks → Add webhook → Gitea:
+
+- Target URL: `http://<server-ip>:9000/deploy`
+- Method: `POST` — Secret: same value as `GITEA_WEBHOOK_SECRET`
+- Trigger: Push Events, Branch filter: `production`
+
+## Behavior
+
+- Verifies `X-Gitea-Signature` (HMAC-SHA256) — rejects everything else with 401.
+- Only pushes to `refs/heads/production` deploy; other branches are acknowledged and ignored.
+- Deploys are serialized with a lock (concurrent pushes don't race).
+- Runs `docker compose -f docker-compose.yml -f docker-compose.prod.yml up -d --build` in `/srv/erp`.
+- After a successful build, polls the API `/health/ready` endpoint and logs
+  a failed post-deploy check (surfaced by the monitoring stack).
+- Logs to journald (`journalctl -u erp-deploy-webhook -f`).
diff --git a/infra/deploy-webhook/deploy_webhook.py b/infra/deploy-webhook/deploy_webhook.py
new file mode 100644
index 0000000..81e60f4
--- /dev/null
+++ b/infra/deploy-webhook/deploy_webhook.py
@@ -0,0 +1,176 @@
+#!/usr/bin/env python3
+"""Hardened Gitea push-to-deploy webhook receiver.
+
+Replaces an unverified "POST /deploy" endpoint. Runs on the VPS (host port
+9000 is reserved for it) and rebuilds the stack when a push to the
+production branch arrives — but ONLY after verifying the Gitea webhook
+signature (HMAC-SHA256 of the raw body with GITEA_WEBHOOK_SECRET).
+
+Why this matters: an unauthenticated deploy endpoint lets anyone on the
+internet restart/rebuild the production stack (and, if it shells out to a
+script, is effectively remote code execution).
+
+Setup on the VPS (see infra/deploy-webhook/README.md):
+
+    GITEA_WEBHOOK_SECRET=... DEPLOY_DIR=/srv/erp DEPLOY_BRANCH=production \
+        python3 deploy_webhook.py            # or run under systemd (provided)
+
+Gitea side: repo → Settings → Webhooks → Add webhook → Gitea
+    Target URL: http://<server>:9000/deploy
+    Method: POST, Secret: same value as GITEA_WEBHOOK_SECRET
+    Trigger on: Push Events, Branch filter: production
+
+Stdlib only — no dependencies to install.
+"""
+import hashlib
+import hmac
+import json
+import logging
+import os
+import subprocess
+import sys
+import time
+from http.server import BaseHTTPRequestHandler, ThreadingHTTPServer
+
+LISTEN_HOST = os.environ.get("WEBHOOK_LISTEN_HOST", "127.0.0.1")
+LISTEN_PORT = int(os.environ.get("WEBHOOK_PORT", "9000"))
+SECRET = os.environ.get("GITEA_WEBHOOK_SECRET", "")
+DEPLOY_DIR = os.environ.get("DEPLOY_DIR", "/srv/erp")
+DEPLOY_BRANCH = os.environ.get("DEPLOY_BRANCH", "production")
+HEALTH_URL = os.environ.get("DEPLOY_HEALTH_URL", "")  # e.g. http://127.0.0.1:8000/health/ready
+LOCK_FILE = os.environ.get("DEPLOY_LOCK", "/tmp/erp-deploy.lock")
+
+logging.basicConfig(
+    level=logging.INFO,
+    format="%(asctime)s %(levelname)s deploy-webhook — %(message)s",
+    handlers=[logging.StreamHandler(sys.stdout)],
+)
+log = logging.getLogger("deploy-webhook")
+
+# Compose file pair used in production (base + overlay).
+DEPLOY_CMD = os.environ.get(
+    "DEPLOY_CMD",
+    "docker compose -f docker-compose.yml -f docker-compose.prod.yml up -d --build",
+)
+
+
+def _verify_signature(body: bytes, header: str | None) -> bool:
+    """Gitea sends X-Gitea-Signature: hex HMAC-SHA256 of the raw body."""
+    if not SECRET:
+        log.error("GITEA_WEBHOOK_SECRET is not set — refusing ALL requests")
+        return False
+    if not header:
+        return False
+    expected = hmac.new(SECRET.encode(), body, hashlib.sha256).hexdigest()
+    return hmac.compare_digest(expected, header.strip().lower())
+
+
+def _deploy() -> tuple[bool, str]:
+    """Run the rebuild. Serialized: concurrent pushes queue/drop, not race."""
+    if os.path.exists(LOCK_FILE):
+        age = time.time() - os.path.getmtime(LOCK_FILE)
+        if age < 1800:  # a deploy is already running (or crashed <30min ago)
+            return False, f"deploy already in progress (lock age {int(age)}s)"
+    try:
+        with open(LOCK_FILE, "w") as f:
+            f.write(str(os.getpid()))
+        log.info("deploy started: %s", DEPLOY_CMD)
+        proc = subprocess.run(
+            DEPLOY_CMD, shell=True, cwd=DEPLOY_DIR,
+            capture_output=True, text=True, timeout=1800,
+        )
+        tail = "\n".join((proc.stdout + proc.stderr).splitlines()[-40:])
+        ok = proc.returncode == 0
+        log.info("deploy finished rc=%s\n%s", proc.returncode, tail)
+        if ok and HEALTH_URL:
+            ok = _wait_healthy()
+        return ok, ("ok" if ok else "deploy or health check failed")
+    except subprocess.TimeoutExpired:
+        log.error("deploy timed out after 1800s")
+        return False, "deploy timed out"
+    except Exception as exc:
+        log.exception("deploy failed")
+        return False, str(exc)
+    finally:
+        try:
+            os.remove(LOCK_FILE)
+        except OSError:
+            pass
+
+
+def _wait_healthy(timeout: int = 180) -> bool:
+    """Poll the app's readiness endpoint after a rebuild."""
+    import urllib.request
+    log.info("waiting for %s", HEALTH_URL)
+    deadline = time.time() + timeout
+    while time.time() < deadline:
+        try:
+            with urllib.request.urlopen(HEALTH_URL, timeout=5) as resp:
+                if resp.status == 200:
+                    log.info("post-deploy health check: OK")
+                    return True
+        except Exception:
+            pass
+        time.sleep(5)
+    log.error("post-deploy health check FAILED after %ss", timeout)
+    return False
+
+
+class Handler(BaseHTTPRequestHandler):
+    def do_POST(self):
+        if self.path != "/deploy":
+            self._json(404, {"error": "not found"})
+            return
+        length = int(self.headers.get("Content-Length", "0"))
+        if length > 1_000_000:
+            self._json(413, {"error": "payload too large"})
+            return
+        body = self.rfile.read(length)
+        if not _verify_signature(body, self.headers.get("X-Gitea-Signature")):
+            log.warning("rejected request: bad/missing signature from %s", self.client_address[0])
+            self._json(401, {"error": "invalid signature"})
+            return
+        try:
+            payload = json.loads(body)
+            ref = payload.get("ref", "")
+            pushed = (payload.get("ref") or "").rsplit("/", 1)[-1]
+            if ref != f"refs/heads/{DEPLOY_BRANCH}":
+                log.info("ignoring push to %s (deploys trigger on %s only)", ref, DEPLOY_BRANCH)
+                self._json(200, {"status": "ignored", "ref": ref})
+                return
+            pusher = payload.get("pusher", {}).get("login", "unknown")
+            log.info("push to %s by %s — deploying", pushed, pusher)
+        except Exception:
+            log.exception("bad payload")
+            self._json(400, {"error": "bad payload"})
+            return
+        # Answer the webhook quickly; deploy runs detached from the response.
+        ok, msg = _deploy()
+        self._json(200 if ok else 500, {"status": "ok" if ok else "failed", "detail": msg})
+
+    def do_GET(self):
+        # Minimal liveness for the receiver itself (monitoring can scrape it).
+        if self.path == "/ping":
+            self._json(200, {"status": "ok", "branch": DEPLOY_BRANCH})
+            return
+        self._json(404, {"error": "not found"})
+
+    def _json(self, code: int, obj: dict):
+        data = json.dumps(obj).encode()
+        self.send_response(code)
+        self.send_header("Content-Type", "application/json")
+        self.send_header("Content-Length", str(len(data)))
+        self.end_headers()
+        self.wfile.write(data)
+
+    def log_message(self, fmt, *args):  # route stdlib access logs to our logger
+        log.debug("access: " + fmt, *args)
+
+
+if __name__ == "__main__":
+    if not SECRET:
+        log.warning("starting WITHOUT signature verification is not supported — set GITEA_WEBHOOK_SECRET")
+        sys.exit(1)
+    server = ThreadingHTTPServer((LISTEN_HOST, LISTEN_PORT), Handler)
+    log.info("listening on %s:%s (branch=%s dir=%s)", LISTEN_HOST, LISTEN_PORT, DEPLOY_BRANCH, DEPLOY_DIR)
+    server.serve_forever()
diff --git a/infra/deploy-webhook/erp-deploy-webhook.service b/infra/deploy-webhook/erp-deploy-webhook.service
new file mode 100644
index 0000000..f1073e2
--- /dev/null
+++ b/infra/deploy-webhook/erp-deploy-webhook.service
@@ -0,0 +1,25 @@
+[Unit]
+Description=ERP push-to-deploy webhook receiver (Gitea -> docker compose)
+After=network-online.target docker.service
+Wants=network-online.target
+
+[Service]
+# Run as a dedicated non-root user with docker group membership:
+#   sudo useradd -r -s /usr/sbin/nologin erpdeploy && sudo usermod -aG docker erpdeploy
+#   sudo chown -R erpdeploy /srv/erp
+User=erpdeploy
+Group=erpdeploy
+WorkingDirectory=/srv/erp
+# EnvironmentFile carries GITEA_WEBHOOK_SECRET (same .env used by compose).
+EnvironmentFile=/srv/erp/.env
+ExecStart=/usr/bin/python3 /srv/erp/infra/deploy-webhook/deploy_webhook.py
+Restart=on-failure
+RestartSec=5
+# Minimal exposure: no write access outside /srv/erp, no new privileges.
+NoNewPrivileges=true
+ProtectSystem=strict
+ReadWritePaths=/srv/erp /tmp
+ProtectHome=true
+
+[Install]
+WantedBy=multi-user.target
diff --git a/infra/traefik-dynamic.yml b/infra/traefik-dynamic.yml
index 0119500..b9fb565 100644
--- a/infra/traefik-dynamic.yml
+++ b/infra/traefik-dynamic.yml
@@ -7,3 +7,31 @@ http:
       redirectScheme:
         scheme: https
         permanent: true
+
+    # Baseline security headers for every public router (attached in the
+    # prod overlay; local http://localhost stays untouched).
+    secure-headers:
+      headers:
+        stsSeconds: 31536000
+        stsIncludeSubdomains: true
+        contentTypeNosniff: true
+        referrerPolicy: strict-origin-when-cross-origin
+        customFrameOptionsValue: DENY
+        browserXssFilter: true
+        customResponseHeaders:
+          X-Powered-By: ""
+          Server: ""
+
+    # Edge rate limiting — coarse abuse brake for the API host
+    # (average 50 req/s sustained, burst 150). Fine-grained login limiting
+    # lives in the app (slowapi).
+    api-ratelimit:
+      rateLimit:
+        average: 50
+        burst: 150
+
+    # Cap request bodies at the edge (attachments are ≤5 MB per file;
+    # 12 MB leaves room for multipart overhead).
+    api-bodylimit:
+      buffering:
+        maxRequestBodyBytes: 12582912