Release v0.6.2

.github/workflows/docker.yml

@@ -132,7 +132,7 @@ jobs:
         uses: actions/checkout@v5
 
       - name: Set up Node.js for Frontend build
-        uses: actions/setup-node@v5
+        uses: actions/setup-node@v6
         with:
           node-version: '24'
           cache: 'npm'

.github/workflows/fe-installation-test.yml

@@ -27,7 +27,7 @@ jobs:
         uses: actions/checkout@v5
 
       - name: Set up Node.js environment
-        uses: actions/setup-node@v5
+        uses: actions/setup-node@v6
         with:
           node-version: "24"
 

.github/workflows/lint-frontend.yml

@@ -29,7 +29,7 @@ jobs:
         uses: actions/checkout@v5
 
       - name: Set up Node.js
-        uses: actions/setup-node@v5
+        uses: actions/setup-node@v6
         with:
             node-version: 24
 

CHANGELOG.md

@@ -1,3 +1,10 @@
+# Version: v0.6.2
+
+* [#125](https://github.com/SAP/STARS/pull/125): Re organize backend
+* [#136](https://github.com/SAP/STARS/pull/136): Bump actions/setup-node from 5 to 6
+* [#137](https://github.com/SAP/STARS/pull/137): Bump the js-dependencies group in /frontend with 17 updates
+
+
 # Version: v0.6.1
 
 * [#112](https://github.com/SAP/STARS/pull/112): Bump sentence-transformers from 5.1.0 to 5.1.1 in /backend-agent

backend-agent/Dockerfile

@@ -1,4 +1,4 @@
-FROM astral/uv:python3.11-trixie-slim AS builder
+FROM astral/uv:python3.12-trixie-slim AS builder
 
 # Install build dependencies with minimal footprint
 RUN apt-get update && apt-get install -y --no-install-recommends \
@@ -38,7 +38,8 @@ RUN . ~/.cargo/env && \
     # Remove test files and documentation from packages (keeping runtime libs)
     find /app/.venv -type d -name "tests" -exec rm -rf {} + 2>/dev/null || true && \
     find /app/.venv -type d -name "test" -exec rm -rf {} + 2>/dev/null || true && \
-    find /app/.venv -type d -name "docs" -exec rm -rf {} + 2>/dev/null || true && \
+    # Do not remove docs folders as they may contain relevant files
+    # find /app/.venv -type d -name "docs" -exec rm -rf {} + 2>/dev/null || true && \
     # Strip debug symbols from shared libraries to reduce size
     find /app/.venv -name "*.so" -exec strip {} + 2>/dev/null || true && \
     # Aggressive cache and temporary file cleanup
@@ -59,7 +60,7 @@ RUN . ~/.cargo/env && \
 
 # ----------------------------------------
 
-FROM python:3.11-slim-trixie AS runtime
+FROM python:3.12-slim-trixie AS runtime
 
 # Install minimal runtime dependencies
 RUN apt-get update && apt-get install -y --no-install-recommends \

backend-agent/app/routes.py

@@ -0,0 +1,245 @@
+import json
+import os
+
+from flask import request, jsonify, send_file, abort
+from sqlalchemy import select
+
+from app.db.models import Attack, ModelAttackScore, TargetModel, db
+from app.utils import send_intro, verify_api_key
+from attack_result import SuiteResult
+from services import run_all_attacks
+from status import status
+
+
+def register_routes(app, sock, agent=None, callbacks=None):
+    # ----------------------
+    # Health endpoints
+    # ----------------------
+    @app.route("/health")
+    def check_health():
+        """
+        Health route is used in the CI to test that the installation was
+        successful.
+        """
+        return jsonify({'status': 'ok'})
+
+    # ----------------------
+    # Attacks endpoints
+    # ----------------------
+    @app.route('/run_all', methods=['POST'])
+    def execute_all_attacks():
+        """
+        Run all attacks. Used for automation.
+        Expected JSON body:
+        {
+        "target": "string"
+        }
+        """
+        verify_api_key()
+        data = request.get_json()
+        target_model = data.get('target') if data else None
+        if not target_model:
+            return jsonify({'error': 'target parameter is required'}), 400
+        # Call the service to run all attacks
+        result = run_all_attacks(
+            target=target_model
+        )
+        return jsonify(result), 200 if result.get('success') else 500
+
+    @app.route('/api/attacks', methods=['GET'])
+    def get_attacks():
+        """
+        Endpoint to retrieve all attacks with their weights.
+        Returns a JSON object with attack names and their weights.
+        """
+        try:
+            attacks = db.session.query(Attack).all()
+            attack_list = [
+                {'name': attack.name, 'weight': attack.weight}
+                for attack in attacks
+            ]
+            return jsonify(attack_list), 200
+        except Exception as e:
+            return jsonify({'error': str(e)}), 500
+
+    @app.route('/api/attacks', methods=['PUT'])
+    def update_attack_weights():
+        """
+        Update weights for multiple attacks.
+        Expects a JSON object like: {"artPrompt": 2, "codeAttack": 1, ...}
+        """
+        verify_api_key()
+        try:
+            weights = request.get_json()
+            if not isinstance(weights, dict):
+                return jsonify({'error': 'Invalid payload format'}), 400
+
+            for name, weight in weights.items():
+                attack = db.session.query(Attack).filter_by(name=name).first()
+                if attack:
+                    attack.weight = float(weight)
+                else:
+                    return jsonify({'error': f'Attack not found: {name}'}), 404
+
+            db.session.commit()
+            return jsonify({'message': 'Weights updated successfully'}), 200
+
+        except Exception as e:
+            db.session.rollback()
+            return jsonify({'error': str(e)}), 500
+
+    # ----------------------
+    # Reports endpoints
+    # ----------------------
+    @app.route('/download_report')
+    def download_report():
+        """
+        This route allows to download attack suite reports by specifying
+        their name.
+        """
+        name = request.args.get('name')
+        format = request.args.get('format', 'md')
+
+        # Ensure that a name is provided
+        if not name:
+            abort(400)
+        # Ensure that only allowed chars are in the filename
+        # (e.g. no path traversal)
+        if not all([c in SuiteResult.FILENAME_ALLOWED_CHARS for c in name]):
+            abort(400)
+
+        results = SuiteResult.load_from_name(name)
+
+        generated_name = name + '_generated'
+        path = os.path.join(SuiteResult.DEFAULT_OUTPUT_PATH, generated_name)
+        result_path = results.to_file(path, format)
+        return send_file(
+            result_path,
+            mimetype=SuiteResult.get_mime_type(format)
+        )
+
+    @app.route('/api/heatmap', methods=['GET'])
+    def get_heatmap():
+        """
+        Endpoint to retrieve heatmap data showing model score
+        against various attacks.
+
+        Queries the database for total attacks and successes per target model
+        and attack combination.
+        Calculates attack success rate and returns structured data for
+        visualization.
+
+        Returns:
+            JSON response with:
+                - models: List of target models and their attack success rate
+                per attack.
+                - attacks: List of attack names and their associated weights.
+
+        HTTP Status Codes:
+            200: Data successfully retrieved.
+            500: Internal server error during query execution.
+        """
+        try:
+            query = (
+                select(
+                    ModelAttackScore.total_number_of_attack,
+                    ModelAttackScore.total_success,
+                    TargetModel.name.label('attack_model_name'),
+                    Attack.name.label('attack_name'),
+                    Attack.weight.label('attack_weight')
+                )
+                .join(TargetModel, ModelAttackScore.target_model_id == TargetModel.id)  # noqa: E501
+                .join(Attack, ModelAttackScore.attack_id == Attack.id)
+            )
+
+            scores = db.session.execute(query).all()
+            all_models = {}
+            all_attacks = {}
+
+            for score in scores:
+                model_name = score.attack_model_name
+                attack_name = score.attack_name
+
+                if attack_name not in all_attacks:
+                    all_attacks[attack_name] = score.attack_weight
+
+                if model_name not in all_models:
+                    all_models[model_name] = {
+                        'name': model_name,
+                        'scores': {},
+                    }
+
+                # Compute attack success rate for this model/attack
+                success_ratio = (
+                    round((score.total_success / score.total_number_of_attack) * 100)  # noqa: E501
+                    if score.total_number_of_attack else 0
+                )
+
+                all_models[model_name]['scores'][attack_name] = success_ratio
+
+            return jsonify({
+                'models': list(all_models.values()),
+                'attacks': [
+                    {'name': name, 'weight': weight}
+                    for name, weight in sorted(all_attacks.items())
+                ]
+            })
+        except Exception as e:
+            return jsonify({'error': str(e)}), 500
+
+    # ----------------------
+    # WebSocket endpoints
+    # ----------------------
+    @sock.route('/agent')
+    def query_agent(sock):
+        """
+        Websocket route for the frontend to send prompts to the agent and
+        receive responses as well as status updates.
+
+        Messages received are in this JSON format:
+        {
+            "type":"message",
+            "data":"Start the vulnerability scan",
+            "key":"secretapikey"
+        }
+        """
+        # Verify API key from headers before establishing session
+        verify_api_key()
+        if not agent:
+            sock.send(json.dumps({
+                'type': 'message',
+                'data': 'Agent is disabled on this deployment.'
+            }))
+            return
+        status.sock = sock
+        # Intro is sent after connecting successfully
+        send_intro(sock)
+        while True:
+            try:
+                data_raw = sock.receive()
+                data = json.loads(data_raw)
+                assert 'data' in data
+                query = data['data']
+                status.clear_report()
+                response = agent.invoke(
+                    {'input': query},
+                    config=callbacks or {}
+                )
+                ai_response = response['output']
+                formatted_output = {
+                    'type': 'message',
+                    'data': (
+                        f'{ai_response}'
+                    )
+                }
+                sock.send(json.dumps(formatted_output))
+            except json.JSONDecodeError:
+                sock.send(json.dumps({
+                    'type': 'error',
+                    'data': 'Invalid JSON format'
+                }))
+            except Exception as e:
+                sock.send(json.dumps({
+                    'type': 'error',
+                    'data': f'Error: {str(e)}'
+                }))

backend-agent/app/utils.py

@@ -0,0 +1,34 @@
+import json
+import os
+
+from flask import request, abort
+
+
+def send_intro(sock):
+    """
+    Sends the intro via the websocket connection.
+
+    The intro is meant as a short tutorial on how to use the agent.
+    Also it includes meaningful suggestions for prompts that should
+    result in predictable behavior for the agent, e.g.
+    "Start the vulnerability scan".
+    """
+    intro_file = 'data/intro.txt'
+    try:
+        with open(intro_file, 'r') as f:
+            intro = f.read()
+    except FileNotFoundError:
+        intro = "Welcome! (intro file missing)"
+    sock.send(json.dumps({'type': 'message', 'data': intro}))
+
+
+def verify_api_key():
+    """
+    Verifies the API key from the request headers against the env variable.
+    If no API key is configured, access is allowed.
+    If API key is configured but missing/invalid, request is rejected.
+    """
+    if os.getenv('API_KEY'):
+        provided_key = request.headers.get('X-API-Key')
+        if provided_key != os.getenv('API_KEY'):
+            abort(403)

backend-agent/cli.py

@@ -3,7 +3,6 @@
 import os
 import sys
 from argparse import ArgumentParser, Namespace
-from pathlib import Path
 from typing import Callable
 
 from attack import AttackSpecification, AttackSuite
@@ -17,6 +16,7 @@
     test as test_textattack,
 )
 from llm import LLM
+from services import run_all_attacks
 from status import Trace
 
 # Library-free Subcommand utilities from
@@ -385,31 +385,9 @@ def run(args):
         required=True),
 ])
 def run_all(args):
-    """Run all LLM attacks with specified target and evaluation models."""
-    default_spec_path = Path('data/all/default.json')
-    try:
-        with default_spec_path.open("r") as f:
-            spec = json.load(f)
-    except FileNotFoundError:
-        print(f'File not found: {args.file}', file=sys.stderr)
-        return
-    except json.JSONDecodeError as e:
-        print(f'Invalid JSON format: {e}', file=sys.stderr)
-        return
-    except PermissionError:
-        print(f'Permission denied reading file: {args.file}', file=sys.stderr)
-        return
-    if 'attacks' in spec:
-        suite = AttackSuite.from_dict(spec)
-        suite.set_target(args.target)
-        results = suite.run()
-        result_return = {'success': True, 'results': results}
-    else:
-        result_return = {
-            'success': False,
-            'error': 'JSON is invalid. No attacks run.'
-        }
-    return result_return
+    """Run all LLM attacks with specified target."""
+    result = run_all_attacks(target=args.target)
+    return result
 
 
 @subcommand()