Compare commits

...

43 Commits

Author SHA1 Message Date
Alexander Myasoedov 94638064d2 feat(bump 0.5.0): 2025-02-20 23:27:34 +02:00
Alexander Myasoedov 701c175469 feat(add $VAR expansion from config): 2025-02-20 23:26:49 +02:00
Alexander Myasoedov ba36dcd02f fix(disable logging): 2025-02-20 17:53:51 +02:00
Alexander Myasoedov 1ce59151f3 feat(add InMemorySecrets to fuzzer): 2025-02-20 16:24:52 +02:00
Alexander Myasoedov da50a48061 fix(imports): 2025-02-20 16:15:55 +02:00
Alexander Myasoedov a944083eea feat(add InMemorySecrets): 2025-02-20 16:15:34 +02:00
Alexander Myasoedov 130ef550df feat(update telemetry): 2025-02-20 16:05:34 +02:00
Alexander Myasoedov 3435d7e6bf feat(simplify lib by refactoring config): 2025-02-20 14:06:32 +02:00
Alexander Myasoedov ee3faab415 feat(update default config path): 2025-02-20 13:09:43 +02:00
Alexander Myasoedov 02255a251c fix(pre commit): 2025-02-17 20:31:13 +02:00
Alexander Myasoedov 15881af019 fix(.gitattributes ): 2025-02-17 20:24:02 +02:00
Alexander Myasoedov 458ebfe638 feat(add .gitattributes ): 2025-02-17 20:23:25 +02:00
Alexander Myasoedov 4ffca42e48 fix(csv file generation bug): 2025-02-17 20:21:47 +02:00
Alexander Myasoedov 653e9a7234 feat(update scan fe logic): 2025-02-17 19:48:06 +02:00
Alexander Myasoedov 3e1dd27f03 fix(add latency param): 2025-02-17 19:47:35 +02:00
Alexander Myasoedov a7f61af921 fix(2024->2025): 2025-02-17 19:47:14 +02:00
Alexander Myasoedov 4f560148ce feat(update theme, fix cdn link): 2025-02-17 19:46:52 +02:00
Alexander Myasoedov 51ff4d8372 fix(discord link): 2025-02-17 18:13:00 +02:00
Alexander Myasoedov c5c310743b fix(.pre-commit-config.yaml): 2025-02-17 18:07:37 +02:00
Alexander Myasoedov 3f83d84941 fix(static files proxing): 2025-02-17 18:02:15 +02:00
Alexander Myasoedov 99fc8cb2e7 fix(fix network error handling in fuzzer): 2025-02-17 18:01:38 +02:00
Alexander Myasoedov 46ef89355b feat(update handling of static files): 2025-02-17 17:58:28 +02:00
Alexander Myasoedov c481676941 feat(update markdown linter): 2025-02-17 17:58:08 +02:00
Alexander Myasoedov 298a0163d6 fix(isort): 2025-02-17 17:39:31 +02:00
Alexander Myasoedov f20d218a16 feat(add llm icons): 2025-02-17 17:38:20 +02:00
Alexander Myasoedov 214341dfbb fix(fix config bar): 2025-02-17 17:18:20 +02:00
Alexander Myasoedov a2fa412141 fix(end-of-file-fixer rule): 2025-02-17 16:03:06 +02:00
Alexander Myasoedov 18f97c7fc2 fix(file): 2025-02-17 16:01:12 +02:00
Alexander Myasoedov 544796ff60 Merge pull request #113 from Praveenk8051/feat/extension-with-sample-tests
feat(operator): add agent testing functionality with endpoint
2025-02-17 16:00:51 +02:00
Alexander Myasoedov b600e69aa1 Merge pull request #127 from Rumixyz/patch-1
Create Vue CLI Setup
2025-02-17 16:00:00 +02:00
Alexander Myasoedov c890b7caeb fix(pre commit): 2025-02-16 17:56:33 +02:00
Praveen 3842f90949 Merge branch 'msoedov:main' into feat/extension-with-sample-tests 2025-02-16 16:50:59 +01:00
Alexander Myasoedov 68cba92d49 Merge pull request #125 from Niharika0104/VueCLI
Migration to VueCLI
2025-02-16 17:40:37 +02:00
Praveenk8051 121d56495e style: streamline code formatting in operator.py for improved readability 2025-02-16 16:13:21 +01:00
Praveenk8051 a001a33f68 refactor: update type hints in AgentSpecification for improved clarity and consistency 2025-02-16 16:11:46 +01:00
Praveenk8051 1c6b8d96fb style: improve code formatting and consistency in operator.py 2025-02-16 15:56:16 +01:00
Praveenk8051 8cc4d79ddf fix: update type hints in OperatorToolBox for consistency 2025-02-16 15:53:13 +01:00
Praveenk8051 fa37cfe710 feat: enhance AgentSpecification and OperatorToolBox with optional typing and improved logging 2025-02-16 15:45:20 +01:00
Praveenk8051 9a2779517b Merge branch 'main' of https://github.com/Praveenk8051/agentic_security into feat/extension-with-sample-tests 2025-02-16 15:45:10 +01:00
Niharika Goulikar 5801dfee7e migration to vueCLi and css to tailwind css 3 done 2025-02-16 11:54:08 +00:00
Rumixyz e4545026e0 Create Vue CLI Setup 2025-02-16 15:21:12 +05:30
Alexander Myasoedov 98e58c9c49 fix(chmod +x changelog.sh): 2025-02-15 13:37:38 +02:00
Praveenk8051 4c0d89bf86 feat(operator): add agent testing functionality with endpoint verification 2025-01-30 07:46:32 +01:00
65 changed files with 44817 additions and 211 deletions
+3
View File
@@ -0,0 +1,3 @@
*.js linguist-detectable=false
*.html linguist-detectable=false
*.py linguist-detectable=true
+1
View File
@@ -16,3 +16,4 @@ garak_rest.json
inv/
scripts/
docx/
agentic_security.toml
+16 -13
View File
@@ -46,20 +46,23 @@ repos:
- id: trailing-whitespace
types: [python]
- id: end-of-file-fixer
types: [python]
types: [file]
files: \.(py|js|vue)$
- repo: https://github.com/executablebooks/mdformat
rev: 0.7.17
hooks:
- id: mdformat
name: mdformat
entry: mdformat .
language_version: python3.11
# - repo: https://github.com/hadialqattan/pycln
# rev: v2.4.0
# - repo: https://github.com/executablebooks/mdformat
# rev: 0.7.22
# hooks:
# - id: pycln
# - id: mdformat
# name: mdformat
# entry: mdformat .
# language_version: python3.11
# files: "docs/.*\\.md$"
- repo: https://github.com/hadialqattan/pycln
rev: v2.5.0
hooks:
- id: pycln
- repo: https://github.com/isidentical/teyit
rev: 0.4.3
@@ -79,8 +82,8 @@ repos:
rev: v2.2.6
hooks:
- id: codespell
exclude: '^(third_party/)|(poetry.lock)'
exclude: '^(third_party/)|(poetry.lock)|(ui/package-lock.json)|(agentic_security/static/.*)'
args:
# if you've got a short variable name that's getting flagged, add it here
- -L bu,ro,te,ue,alo,hda,ois,nam,nams,ned,som,parm,setts,inout,warmup,bumb,nd,sie
- -L bu,ro,te,ue,alo,hda,ois,nam,nams,ned,som,parm,setts,inout,warmup,bumb,nd,sie,vEw
- --builtins clear,rare,informal,usage,code,names,en-GB_to_en-US
+1 -1
View File
@@ -19,7 +19,7 @@
<a href="https://github.com/msoedov/agentic_security/blob/master/LICENSE">
<img alt="GitHub License" src="https://img.shields.io/github/license/msoedov/agentic_security?style=for-the-badge&logo=codeigniter&labelColor=000000&logoColor=FFFFFF&label=License&color=FFCC19" />
</a>
<a href="https://discord.com/channels/1340010688764051499/1340010689309315247"><img alt="Join the community" src="https://img.shields.io/badge/Join%20the%20community-black.svg?style=for-the-badge&logo=lightning&labelColor=000000&logoColor=FFFFFF&label=&color=DD55FF&logoWidth=20" /></a>
<a href="https://discord.gg/stw3DfZQ"><img alt="Join the community" src="https://img.shields.io/badge/Join%20the%20community-black.svg?style=for-the-badge&logo=lightning&labelColor=000000&logoColor=FFFFFF&label=&color=DD55FF&logoWidth=20" /></a>
</p>
+2
View File
@@ -8,6 +8,7 @@ from .routes import (
report_router,
scan_router,
static_router,
telemetry,
)
# Create the FastAPI app
@@ -26,3 +27,4 @@ app.include_router(scan_router)
app.include_router(probe_router)
app.include_router(proxy_router)
app.include_router(report_router)
telemetry.setup(app)
+122
View File
@@ -0,0 +1,122 @@
import tomli
from loguru import logger
class CfgMixin:
config = {}
default_path = "agentic_security.toml"
def get_or_create_config(self) -> bool:
if not self.has_local_config():
self.generate_default_cfg()
return False
self.load_config(self.default_path)
return True
def has_local_config(self):
try:
with open(self.default_path):
return True
except FileNotFoundError:
return False
@classmethod
def load_config(cls, config_path: str):
"""
Load configuration from a TOML file and store it in the class variable.
Args:
config_path (str): Path to the TOML configuration file.
Raises:
FileNotFoundError: If the configuration file is not found.
toml.TomlDecodeError: If the configuration file has syntax errors.
"""
try:
with open(config_path, "rb") as config_file:
cls.config = tomli.load(config_file)
logger.info(f"Configuration loaded successfully from {config_path}.")
except FileNotFoundError:
logger.error(f"Configuration file {config_path} not found.")
raise
except Exception as e:
logger.error(f"Error parsing TOML configuration: {e}")
raise
@classmethod
def get_config_value(cls, key: str, default=None):
"""
Retrieve a configuration value by key from the loaded configuration.
Args:
key (str): Dot-separated key path to the configuration value (e.g., 'general.maxBudget').
default: Default value if the key is not found.
Returns:
The configuration value if found, otherwise the default value.
"""
keys = key.split(".")
value = cls.config
for k in keys:
if isinstance(value, dict) and k in value:
value = value[k]
else:
return default
return value
def generate_default_cfg(self, host: str = "0.0.0.0", port: int = 8718):
# Accept host / port as parameters
with open(self.default_path, "w") as f:
f.write(
"""
[general]
# General configuration for the security scan
llmSpec = \"""
POST http://$HOST:$PORT/v1/self-probe
Authorization: Bearer XXXXX
Content-Type: application/json
{
"prompt": "<<PROMPT>>"
}
\""" # LLM API specification
maxBudget = 1000000 # Maximum budget for the scan
max_th = 0.3 # Maximum failure threshold (percentage)
optimize = false # Enable optimization during scanning
enableMultiStepAttack = false # Enable multi-step attack simulations
# [modules.LLM-Jailbreak-Classifier]
# dataset_name = "markush1/LLM-Jailbreak-Classifier"
[modules.aya-23-8B_advbench_jailbreak]
dataset_name = "simonycl/aya-23-8B_advbench_jailbreak"
[modules.AgenticBackend]
dataset_name = "AgenticBackend"
[modules.AgenticBackend.opts]
port = $PORT
modules = ["encoding"]
[thresholds]
# Threshold settings
low = 0.15
medium = 0.3
high = 0.5
[secrets]
# Secrets for the security scan from environment variables
OPENAI_API_KEY = "$OPENAI_API_KEY"
DEEPSEEK_API_KEY = "$DEEPSEEK_API_KEY"
""".replace(
"$HOST", host
).replace(
"$PORT", str(port)
)
)
logger.info(
f"Default configuration generated successfully to {self.default_path}."
)
+19
View File
@@ -1,3 +1,4 @@
import os
from asyncio import Event, Queue
from fastapi import FastAPI
@@ -5,6 +6,7 @@ from fastapi import FastAPI
tools_inbox: Queue = Queue()
stop_event: Event = Event()
current_run: str = {"spec": "", "id": ""}
_secrets = {}
def create_app() -> FastAPI:
@@ -33,3 +35,20 @@ def set_current_run(spec):
current_run["id"] = hash(id(spec))
current_run["spec"] = spec
return current_run
def get_secrets():
return _secrets
def set_secrets(secrets):
_secrets.update(secrets)
expand_secrets(_secrets)
return _secrets
def expand_secrets(secrets):
for key in secrets:
val = secrets[key]
if val.startswith("$"):
secrets[key] = os.getenv(val.strip("$"))
+27
View File
@@ -0,0 +1,27 @@
import os
import pytest
from agentic_security.core.app import expand_secrets
@pytest.fixture(autouse=True)
def setup_env_vars():
# Set up environment variables for testing
os.environ["TEST_ENV_VAR"] = "test_value"
def test_expand_secrets_with_env_var():
secrets = {"secret_key": "$TEST_ENV_VAR"}
expand_secrets(secrets)
assert secrets["secret_key"] == "test_value"
def test_expand_secrets_without_env_var():
secrets = {"secret_key": "$NON_EXISTENT_VAR"}
expand_secrets(secrets)
assert secrets["secret_key"] is None
def test_expand_secrets_without_dollar_sign():
secrets = {"secret_key": "plain_value"}
expand_secrets(secrets)
assert secrets["secret_key"] == "plain_value"
+29
View File
@@ -0,0 +1,29 @@
from agentic_security.config import CfgMixin
from agentic_security.core.app import set_secrets
class InMemorySecrets:
def __init__(self):
self.secrets = {}
self.config = CfgMixin()
self.config.get_or_create_config()
self.secrets = self.config.config.get("secrets", {})
set_secrets(self.secrets)
def set_secret(self, key: str, value: str):
self.secrets[key] = value
def get_secret(self, key: str) -> str:
return self.secrets.get(key, None)
# Dependency
def get_in_memory_secrets() -> InMemorySecrets:
return InMemorySecrets()
# Example usage in a FastAPI route
# @app.get("/some-endpoint")
# async def some_endpoint(secrets: InMemorySecrets = Depends(get_in_memory_secrets)):
# # Use secrets here
# pass
+8
View File
@@ -138,6 +138,9 @@ def parse_http_spec(http_spec: str) -> LLMSpec:
Returns:
LLMSpec: An object representing the parsed HTTP specification, with attributes for the method, URL, headers, and body.
"""
from agentic_security.core.app import get_secrets
secrets = get_secrets()
# Split the spec by lines
lines = http_spec.strip().split("\n")
@@ -164,6 +167,11 @@ def parse_http_spec(http_spec: str) -> LLMSpec:
has_files = "multipart/form-data" in headers.get("Content-Type", "")
has_image = "<<BASE64_IMAGE>>" in body
has_audio = "<<BASE64_AUDIO>>" in body
for key, value in secrets.items():
key = key.strip("$")
body = body.replace(f"${key}", value)
return LLMSpec(
method=method,
url=url,
+1 -110
View File
@@ -3,13 +3,13 @@ import json
from datetime import datetime
import colorama
import tomli
import tqdm.asyncio
from loguru import logger
from rich.console import Console
from rich.table import Table
from tabulate import tabulate
from agentic_security.config import CfgMixin # Importing the configuration mixin
from agentic_security.models.schemas import Scan
from agentic_security.probe_data import REGISTRY
from agentic_security.routes.scan import streaming_response_generator
@@ -23,62 +23,6 @@ YELLOW = colorama.Fore.YELLOW
BLUE = colorama.Fore.BLUE
class CfgMixin:
config = {}
default_path = "agesec.toml"
def has_local_config(self):
try:
with open(self.default_path):
return True
except FileNotFoundError:
return False
@classmethod
def load_config(cls, config_path: str):
"""
Load configuration from a TOML file and store it in the class variable.
Args:
config_path (str): Path to the TOML configuration file.
Raises:
FileNotFoundError: If the configuration file is not found.
toml.TomlDecodeError: If the configuration file has syntax errors.
"""
try:
with open(config_path, "rb") as config_file:
cls.config = tomli.load(config_file)
logger.info(f"Configuration loaded successfully from {config_path}.")
except FileNotFoundError:
logger.error(f"Configuration file {config_path} not found.")
raise
except Exception as e:
logger.error(f"Error parsing TOML configuration: {e}")
raise
@classmethod
def get_config_value(cls, key: str, default=None):
"""
Retrieve a configuration value by key from the loaded configuration.
Args:
key (str): Dot-separated key path to the configuration value (e.g., 'general.maxBudget').
default: Default value if the key is not found.
Returns:
The configuration value if found, otherwise the default value.
"""
keys = key.split(".")
value = cls.config
for k in keys:
if isinstance(value, dict) and k in value:
value = value[k]
else:
return default
return value
class AgenticSecurity(CfgMixin):
@classmethod
async def async_scan(
@@ -272,59 +216,6 @@ class AgenticSecurity(CfgMixin):
),
)
def generate_default_cfg(self, host: str = "0.0.0.0", port: int = 8718):
# Accept host / port as parameters
with open(self.default_path, "w") as f:
f.write(
"""
[general]
# General configuration for the security scan
llmSpec = \"""
POST http://$HOST:$PORT/v1/self-probe
Authorization: Bearer XXXXX
Content-Type: application/json
{
"prompt": "<<PROMPT>>"
}
\""" # LLM API specification
maxBudget = 1000000 # Maximum budget for the scan
max_th = 0.3 # Maximum failure threshold (percentage)
optimize = false # Enable optimization during scanning
enableMultiStepAttack = false # Enable multi-step attack simulations
# [modules.LLM-Jailbreak-Classifier]
# dataset_name = "markush1/LLM-Jailbreak-Classifier"
[modules.aya-23-8B_advbench_jailbreak]
dataset_name = "simonycl/aya-23-8B_advbench_jailbreak"
[modules.AgenticBackend]
dataset_name = "AgenticBackend"
[modules.AgenticBackend.opts]
port = $PORT
modules = ["encoding"]
[thresholds]
# Threshold settings
low = 0.15
medium = 0.3
high = 0.5
""".replace(
"$HOST", host
).replace(
"$PORT", str(port)
)
)
logger.info(
f"Default configuration generated successfully to {self.default_path}."
)
def list_checks(self):
"""
Print the REGISTRY contents as a table using the rich library.
+20
View File
@@ -23,6 +23,18 @@ class Scan(BaseModel):
enableMultiStepAttack: bool = False
# MSJ only mode
probe_datasets: list[dict] = []
# Set and managed by the backend
secrets: dict[str, str] = {}
def with_secrets(self, secrets) -> "Scan":
match secrets:
case dict():
self.secrets.update(secrets)
case obj if hasattr(obj, "secrets"):
self.secrets.update(obj.secrets)
case _:
raise ValueError("Invalid secrets type")
return self
class ScanResult(BaseModel):
@@ -32,6 +44,10 @@ class ScanResult(BaseModel):
progress: float
status: bool = False
failureRate: float = 0.0
prompt: str = ""
model: str = ""
refused: bool = False
latency: float = 0.0
@classmethod
def status_msg(cls, msg: str) -> str:
@@ -42,6 +58,10 @@ class ScanResult(BaseModel):
progress=0,
failureRate=0,
status=True,
prompt="",
model="",
refused=False,
latency=0,
).model_dump_json()
+40 -11
View File
@@ -1,5 +1,6 @@
import asyncio
import random
import time
from collections.abc import AsyncGenerator
import httpx
@@ -17,6 +18,8 @@ from agentic_security.probe_data.data import prepare_prompts
# TODO: full log file
MAX_PROMPT_LENGTH = 2048
async def generate_prompts(
prompts: list[str] | AsyncGenerator,
@@ -42,8 +45,11 @@ def multi_modality_spec(llm_spec):
async def process_prompt(
request_factory, prompt, tokens, module_name, refusals, errors
):
request_factory, prompt, tokens, module_name, refusals, errors, outputs
) -> tuple[int, bool]:
"""
Process a single prompt and update the token count and failure status.
"""
try:
response = await request_factory.fn(prompt=prompt)
if response.status_code == 422:
@@ -52,18 +58,18 @@ async def process_prompt(
return tokens, True
if response.status_code >= 400:
raise httpx.HTTPStatusError(
f"HTTP {response.status_code} {response.content=}",
request=response.request,
response=response,
)
logger.error(f"HTTP {response.status_code} {response.content=}")
errors.append((module_name, prompt, response.status_code, response.text))
return tokens, True
response_text = response.text
tokens += len(response_text.split())
if refusal_heuristic(response.json()):
refused = refusal_heuristic(response.json())
if refused:
refusals.append((module_name, prompt, response.status_code, response_text))
return tokens, True
return tokens, False
outputs.append((module_name, prompt, response_text, refused))
return tokens, refused
except httpx.RequestError as exc:
logger.error(f"Request error: {exc}")
@@ -78,6 +84,7 @@ async def perform_single_shot_scan(
tools_inbox=None,
optimize=False,
stop_event: asyncio.Event = None,
secrets: dict[str, str] = {},
) -> AsyncGenerator[str, None]:
"""Perform a standard security scan."""
max_budget = max_budget * 100_000_000
@@ -95,6 +102,7 @@ async def perform_single_shot_scan(
errors = []
refusals = []
outputs = []
total_prompts = sum(len(m.prompts) for m in prompt_modules if not m.lazy)
processed_prompts = 0
@@ -128,6 +136,7 @@ async def perform_single_shot_scan(
100 * processed_prompts / total_prompts if total_prompts else 0
)
total_tokens -= tokens
start = time.time()
tokens, failed = await process_prompt(
request_factory,
prompt,
@@ -135,7 +144,9 @@ async def perform_single_shot_scan(
module.dataset_name,
refusals,
errors,
outputs,
)
end = time.time()
total_tokens += tokens
# logger.debug(f"Trying prompt: {prompt}, {failed=}")
if failed:
@@ -144,12 +155,22 @@ async def perform_single_shot_scan(
failure_rates.append(failure_rate)
cost = calculate_cost(tokens)
# TODO: improve this cond
last_output = outputs[-1] if outputs else None
if last_output and last_output[1] == prompt:
response_text = last_output[2]
else:
response_text = ""
yield ScanResult(
module=module.dataset_name,
tokens=round(tokens / 1000, 1),
cost=cost,
progress=round(progress, 2),
failureRate=round(failure_rate * 100, 2),
prompt=prompt[:MAX_PROMPT_LENGTH],
latency=end - start,
model=response_text,
).model_dump_json()
if optimize and len(failure_rates) >= 5:
@@ -183,7 +204,9 @@ async def perform_single_shot_scan(
except Exception as e:
logger.exception("Scan failed")
yield ScanResult.status_msg(f"Scan failed: {str(e)}")
raise e
# raise e
finally:
yield ScanResult.status_msg("Scan completed.")
async def perform_many_shot_scan(
@@ -196,6 +219,7 @@ async def perform_many_shot_scan(
stop_event: asyncio.Event = None,
probe_frequency: float = 0.2,
max_ctx_length: int = 10_000,
secrets: dict[str, str] = {},
) -> AsyncGenerator[str, None]:
"""Perform a multi-step security scan with probe injection."""
request_factory = multi_modality_spec(request_factory)
@@ -213,6 +237,7 @@ async def perform_many_shot_scan(
errors = []
refusals = []
outputs = []
total_prompts = sum(len(m.prompts) for m in prompt_modules if not m.lazy)
processed_prompts = 0
@@ -264,6 +289,7 @@ async def perform_many_shot_scan(
module.dataset_name,
refusals,
errors,
outputs,
)
if failed:
module_failures += 1
@@ -281,6 +307,7 @@ async def perform_many_shot_scan(
cost=cost,
progress=round(progress, 2),
failureRate=round(failure_rate * 100, 2),
prompt=prompt[:MAX_PROMPT_LENGTH],
).model_dump_json()
if optimize and len(failure_rates) >= 5:
@@ -321,6 +348,7 @@ def scan_router(
tools_inbox=tools_inbox,
optimize=scan_parameters.optimize,
stop_event=stop_event,
secrets=scan_parameters.secrets,
)
else:
return perform_single_shot_scan(
@@ -330,4 +358,5 @@ def scan_router(
tools_inbox=tools_inbox,
optimize=scan_parameters.optimize,
stop_event=stop_event,
secrets=scan_parameters.secrets,
)
+84 -20
View File
@@ -1,9 +1,16 @@
import asyncio
import logging
from typing import Any
import httpx
from httpx import LLMSpec
from pydantic import BaseModel, Field
from pydantic_ai import Agent, RunContext
# Configure logging
logging.basicConfig(level=logging.INFO)
logger = logging.getLogger(__name__)
class AgentSpecification(BaseModel):
name: str | None = Field(None, description="Name of the LLM/agent")
@@ -13,9 +20,9 @@ class AgentSpecification(BaseModel):
configuration: dict[str, Any] | None = Field(
None, description="Configuration settings"
)
endpoint: str | None = Field(None, description="Endpoint URL of the deployed agent")
# Define the OperatorToolBox class
class OperatorToolBox:
def __init__(self, spec: AgentSpecification, datasets: list[dict[str, Any]]):
self.spec = spec
@@ -29,7 +36,6 @@ class OperatorToolBox:
return self.datasets
def validate(self) -> bool:
# Validate the tool box based on the specification
if not self.spec.name or not self.spec.version:
self.failures.append("Invalid specification: Name or version is missing.")
return False
@@ -39,28 +45,70 @@ class OperatorToolBox:
return True
def stop(self) -> None:
# Stop the tool box
print("Stopping the toolbox...")
logger.info("Stopping the toolbox...")
def run(self) -> None:
# Run the tool box
print("Running the toolbox...")
logger.info("Running the toolbox...")
def get_results(self) -> list[dict[str, Any]]:
# Get the results
return self.datasets
def get_failures(self) -> list[str]:
# Handle failure
return self.failures
def run_operation(self, operation: str) -> str:
# Run an operation based on the specification
if operation not in ["dataset1", "dataset2", "dataset3"]:
self.failures.append(f"Operation '{operation}' failed: Dataset not found.")
return f"Operation '{operation}' failed: Dataset not found."
return f"Operation '{operation}' executed successfully."
async def test(self, description: str, sample_test: dict[str, Any]) -> str:
agent = Agent(
"openai:gpt-4o",
result_type=LLMSpec,
system_prompt="Extract the LLM specification from the input",
)
async with agent.run_stream(description) as result:
async for spec in result.stream():
self.spec.endpoint = spec.url
# Verify access to the endpoint
async with httpx.AsyncClient() as client:
try:
access_response = await client.get(spec.url)
access_response.raise_for_status()
except httpx.HTTPStatusError as e:
self.failures.append(f"HTTP error occurred: {e}")
logger.error(f"Access verification failed: {e}")
return f"Access verification failed: {e}"
except Exception as e:
self.failures.append(f"An error occurred: {e}")
logger.error(f"Access verification failed: {e}")
return f"Access verification failed: {e}"
# Run the sample test
try:
test_response = await client.post(
f"{spec.url}/test", json=sample_test
)
test_response.raise_for_status()
response_data = test_response.json()
if "choices" in response_data and len(response_data["choices"]) > 0:
return f"Testing agent at {spec.url} succeeded: {response_data}"
else:
self.failures.append("Invalid response format")
logger.error("Sample test failed: Invalid response format")
return "Sample test failed: Invalid response format"
except httpx.HTTPStatusError as e:
self.failures.append(f"HTTP error occurred: {e}")
logger.error(f"Sample test failed: {e}")
return f"Sample test failed: {e}"
except Exception as e:
self.failures.append(f"An error occurred: {e}")
logger.error(f"Sample test failed: {e}")
return f"Sample test failed: {e}"
# Initialize OperatorToolBox with AgentSpecification
spec = AgentSpecification(
@@ -71,24 +119,19 @@ spec = AgentSpecification(
configuration={"max_tokens": 100},
)
# dataset_manager_agent.py
# Initialize OperatorToolBox
toolbox = OperatorToolBox(spec=spec, datasets=["dataset1", "dataset2", "dataset3"])
# Define the agent with OperatorToolBox as its dependency
dataset_manager_agent = Agent(
model="gpt-4",
deps_type=OperatorToolBox,
result_type=str, # The agent will return string results
result_type=str,
system_prompt="You can validate the toolbox, run operations, and retrieve results or failures.",
)
@dataset_manager_agent.tool
async def validate_toolbox(ctx: RunContext[OperatorToolBox]) -> str:
"""Validate the OperatorToolBox."""
is_valid = ctx.deps.validate()
if is_valid:
return "ToolBox validation successful."
@@ -98,14 +141,12 @@ async def validate_toolbox(ctx: RunContext[OperatorToolBox]) -> str:
@dataset_manager_agent.tool
async def execute_operation(ctx: RunContext[OperatorToolBox], operation: str) -> str:
"""Execute an operation on a dataset."""
result = ctx.deps.run_operation(operation)
return result
@dataset_manager_agent.tool
async def retrieve_results(ctx: RunContext[OperatorToolBox]) -> str:
"""Retrieve the results of operations."""
results = ctx.deps.get_results()
if results:
formatted_results = "\n".join([f"{op}: {res}" for op, res in results.items()])
@@ -116,7 +157,6 @@ async def retrieve_results(ctx: RunContext[OperatorToolBox]) -> str:
@dataset_manager_agent.tool
async def retrieve_failures(ctx: RunContext[OperatorToolBox]) -> str:
"""Retrieve the list of failures."""
failures = ctx.deps.get_failures()
if failures:
formatted_failures = "\n".join(failures)
@@ -125,6 +165,14 @@ async def retrieve_failures(ctx: RunContext[OperatorToolBox]) -> str:
return "No failures recorded."
@dataset_manager_agent.tool
async def test_agent(
ctx: RunContext[OperatorToolBox], description: str, sample_test: dict[str, Any]
) -> str:
result = await ctx.deps.test(description, sample_test)
return result
# Synchronous run example
def run_dataset_manager_agent_sync():
prompts = [
@@ -133,10 +181,18 @@ def run_dataset_manager_agent_sync():
"Execute operation on 'dataset4'.", # This should fail
"Retrieve the results.",
"Retrieve any failures.",
"Test my openAI compatible agent deployed at localhost:3000",
]
sample_test = {"prompt": "Hello, how are you?", "max_tokens": 5}
for prompt in prompts:
result = dataset_manager_agent.run_sync(prompt, deps=toolbox)
if "Test my" in prompt:
result = dataset_manager_agent.run_sync(
prompt, deps=toolbox, sample_test=sample_test
)
else:
result = dataset_manager_agent.run_sync(prompt, deps=toolbox)
print(f"Prompt: {prompt}")
print(f"Response: {result.data}\n")
@@ -149,10 +205,18 @@ async def run_dataset_manager_agent_async():
"Execute operation on 'dataset4'.", # This should fail
"Retrieve the results.",
"Retrieve any failures.",
"Test my openAI compatible agent deployed at localhost:3000",
]
sample_test = {"prompt": "Hello, how are you?", "max_tokens": 5}
for prompt in prompts:
result = await dataset_manager_agent.run(prompt, deps=toolbox)
if "Test my" in prompt:
result = await dataset_manager_agent.run(
prompt, deps=toolbox, sample_test=sample_test
)
else:
result = await dataset_manager_agent.run(prompt, deps=toolbox)
print(f"Prompt: {prompt}")
print(f"Response: {result.data}\n")
+13 -9
View File
@@ -209,6 +209,7 @@ class TestProcessPrompt(unittest.IsolatedAsyncioTestCase):
module_name="module_a",
refusals=[],
errors=[],
outputs=[],
)
self.assertEqual(tokens, 3) # Tokens from "Valid response text"
@@ -226,6 +227,7 @@ class TestProcessPrompt(unittest.IsolatedAsyncioTestCase):
)
refusals = []
outputs = []
tokens, refusal = await process_prompt(
request_factory=mock_request_factory,
prompt="test prompt",
@@ -233,6 +235,7 @@ class TestProcessPrompt(unittest.IsolatedAsyncioTestCase):
module_name="module_a",
refusals=refusals,
errors=[],
outputs=outputs,
)
self.assertEqual(tokens, 3) # Tokens from "Response indicating refusal"
@@ -250,15 +253,15 @@ class TestProcessPrompt(unittest.IsolatedAsyncioTestCase):
)
refusals = []
with self.assertRaises(httpx.HTTPStatusError):
await process_prompt(
request_factory=mock_request_factory,
prompt="test prompt",
tokens=0,
module_name="module_a",
refusals=refusals,
errors=[],
)
await process_prompt(
request_factory=mock_request_factory,
prompt="test prompt",
tokens=0,
module_name="module_a",
refusals=refusals,
errors=[],
outputs=[],
)
async def test_request_error(self):
mock_request_factory = Mock()
@@ -274,6 +277,7 @@ class TestProcessPrompt(unittest.IsolatedAsyncioTestCase):
module_name="module_a",
refusals=[],
errors=errors,
outputs=[],
)
self.assertEqual(tokens, 0)
+21 -4
View File
@@ -1,9 +1,18 @@
from datetime import datetime
from fastapi import APIRouter, BackgroundTasks, File, HTTPException, Query, UploadFile
from fastapi import (
APIRouter,
BackgroundTasks,
Depends,
File,
HTTPException,
Query,
UploadFile,
)
from fastapi.responses import StreamingResponse
from ..core.app import get_stop_event, get_tools_inbox, set_current_run
from ..dependencies import InMemorySecrets, get_in_memory_secrets
from ..http_spec import LLMSpec
from ..models.schemas import LLMInfo, Scan
from ..probe_actor import fuzzer
@@ -12,7 +21,9 @@ router = APIRouter()
@router.post("/verify")
async def verify(info: LLMInfo):
async def verify(
info: LLMInfo, secrets: InMemorySecrets = Depends(get_in_memory_secrets)
):
spec = LLMSpec.from_string(info.spec)
r = await spec.verify()
if r.status_code >= 400:
@@ -42,7 +53,12 @@ def streaming_response_generator(scan_parameters: Scan):
@router.post("/scan")
async def scan(scan_parameters: Scan, background_tasks: BackgroundTasks):
async def scan(
scan_parameters: Scan,
background_tasks: BackgroundTasks,
secrets: InMemorySecrets = Depends(get_in_memory_secrets),
):
scan_parameters.with_secrets(secrets)
return StreamingResponse(
streaming_response_generator(scan_parameters), media_type="application/json"
)
@@ -62,6 +78,7 @@ async def scan_csv(
optimize: bool = Query(False),
maxBudget: int = Query(10_000),
enableMultiStepAttack: bool = Query(False),
secrets: InMemorySecrets = Depends(get_in_memory_secrets),
):
# TODO: content dataset to fuzzer
content = await file.read() # noqa
@@ -73,7 +90,7 @@ async def scan_csv(
maxBudget=1000,
enableMultiStepAttack=enableMultiStepAttack,
)
scan_parameters.with_secrets(secrets)
return StreamingResponse(
streaming_response_generator(scan_parameters), media_type="application/json"
)
+95
View File
@@ -1,5 +1,6 @@
from pathlib import Path
import requests
from fastapi import APIRouter, HTTPException, Request
from fastapi.responses import FileResponse, HTMLResponse
from fastapi.templating import Jinja2Templates
@@ -10,6 +11,7 @@ from ..models.schemas import Settings
router = APIRouter()
STATIC_DIR = Path(__file__).parent.parent / "static"
ICONS_DIR = STATIC_DIR / "icons"
# Configure templates with custom delimiters to avoid conflicts
templates = Jinja2Templates(directory=str(STATIC_DIR))
@@ -28,6 +30,8 @@ CONTENT_TYPES = {
".ico": "image/x-icon",
".html": "text/html",
".css": "text/css",
".svg": "image/svg+xml",
".png": "image/png",
}
@@ -88,3 +92,94 @@ async def telemetry_js() -> FileResponse:
async def favicon() -> FileResponse:
"""Serve the favicon."""
return get_static_file(STATIC_DIR / "favicon.ico")
@router.get("/icons/{icon_name}")
async def serve_icon(icon_name: str) -> FileResponse:
"""Serve an icon from the icons directory."""
icon_path = ICONS_DIR / icon_name
if not icon_path.exists():
# Fetch the icon from the external URL and cache it
url = f"https://registry.npmmirror.com/@lobehub/icons-static-png/latest/files/dark/{icon_name}"
response = requests.get(url)
if response.status_code == 200:
icon_path.write_bytes(response.content)
else:
raise HTTPException(status_code=404, detail="Icon not found")
return get_static_file(icon_path, content_type="image/png")
# New endpoints for proxying external resources
@router.get("/cdn/tailwindcss.js")
async def proxy_tailwindcss() -> FileResponse:
"""Proxy the Tailwind CSS script."""
return proxy_external_resource(
"https://cdn.tailwindcss.com",
STATIC_DIR / "tailwindcss.js",
"application/javascript",
)
@router.get("/cdn/vue.js")
async def proxy_vue() -> FileResponse:
"""Proxy the Vue.js script."""
return proxy_external_resource(
"https://unpkg.com/vue@2.6.12/dist/vue.js",
STATIC_DIR / "vue.js",
"application/javascript",
)
@router.get("/cdn/lucide.js")
async def proxy_lucide() -> FileResponse:
"""Proxy the Lucide.js script."""
return proxy_external_resource(
"https://unpkg.com/lucide@latest/dist/umd/lucide.js",
STATIC_DIR / "lucide.js",
"application/javascript",
)
@router.get("/cdn/technopollas.css")
async def proxy_technopollas() -> FileResponse:
"""Proxy the Technopollas font stylesheet."""
return proxy_external_resource(
"https://fonts.cdnfonts.com/css/technopollas",
STATIC_DIR / "technopollas.css",
"text/css",
)
@router.get("/cdn/inter.css")
async def proxy_inter() -> FileResponse:
"""Proxy the Inter font stylesheet."""
return proxy_external_resource(
"https://fonts.googleapis.com/css2?family=Inter:wght@400;600;700&display=swap",
STATIC_DIR / "inter.css",
"text/css",
)
def proxy_external_resource(
url: str, local_path: Path, content_type: str
) -> FileResponse:
"""
Fetch and cache an external resource, then serve it locally.
Args:
url: The URL of the external resource
local_path: The local path to cache the resource
content_type: The content type of the resource
Returns:
FileResponse with the cached resource
"""
if not local_path.exists():
response = requests.get(url)
if response.status_code == 200:
local_path.write_bytes(response.content)
else:
raise HTTPException(status_code=404, detail="Resource not found")
return get_static_file(local_path, content_type=content_type)
+27
View File
@@ -0,0 +1,27 @@
import sentry_sdk
from loguru import logger
from sentry_sdk.integrations.logging import ignore_logger
from ..models.schemas import Settings
def setup(app):
if Settings.DISABLE_TELEMETRY:
return
sentry_sdk.init(
dsn="https://b5c59f7e5ab86d73518222ddb40807c9@o4508851738247168.ingest.de.sentry.io/4508851740541008",
# Add data like request headers and IP for users,
# see https://docs.sentry.io/platforms/python/data-management/data-collected/ for more info
send_default_pii=True,
# Set traces_sample_rate to 1.0 to capture 100%
# of transactions for tracing.
traces_sample_rate=1.0,
_experiments={
# Set continuous_profiling_auto_start to True
# to automatically start the profiler on when
# possible.
"continuous_profiling_auto_start": True,
},
)
ignore_logger("logging.error")
ignore_logger(logger.error)
+22 -24
View File
@@ -1,13 +1,13 @@
let URL = window.location.href;
if (URL.endsWith('/')) {
URL = URL.slice(0, -1);
let SELF_URL = window.location.href;
if (SELF_URL.endsWith('/')) {
SELF_URL = SELF_URL.slice(0, -1);
}
URL = URL.replace('/#', '');
SELF_URL = SELF_URL.replace('/#', '');
// Vue application
let LLM_SPECS = [
`POST ${URL}/v1/self-probe
`POST ${SELF_URL}/v1/self-probe
Authorization: Bearer XXXXX
Content-Type: application/json
@@ -79,7 +79,7 @@ Content-Type: application/json
]
}
`,
`POST ${URL}/v1/self-probe-image
`POST ${SELF_URL}/v1/self-probe-image
Authorization: Bearer XXXXX
Content-Type: application/json
@@ -101,7 +101,7 @@ Content-Type: application/json
}
]
`,
`POST ${URL}/v1/self-probe-file
`POST ${SELF_URL}/v1/self-probe-file
Authorization: Bearer $GROQ_API_KEY
Content-Type: multipart/form-data
@@ -175,25 +175,23 @@ Content-Type: application/json
]
let fallbackIcon = '/icons/myshell.png';
let LLM_CONFIGS = [
{ name: 'Custom API', prompts: 40000, customInstructions: 'Requires api spec' },
{ name: 'Open AI', prompts: 24000 },
{ name: 'Deepseek v1', prompts: 24000 },
{ name: 'Replicate', prompts: 40000 },
{ name: 'Groq', prompts: 40000 },
{ name: 'Together.ai', prompts: 40000 },
{ name: 'Custom API Image', prompts: 40000, customInstructions: 'Requires api spec', modality: 'Image' },
{ name: 'Custom API Files', prompts: 40000, customInstructions: 'Requires api spec', modality: 'Files' },
{ name: 'Gemini', prompts: 40000 },
{ name: 'Claude', prompts: 40000 },
{ name: 'Cohere', prompts: 40000 },
{ name: 'Azure OpenAI', prompts: 40000 },
{ name: 'assemblyai', prompts: 40000 },
]
{ name: 'Custom API', prompts: 40000, customInstructions: 'Requires api spec', logo: fallbackIcon },
{ name: 'Open AI', prompts: 24000, logo: '/icons/openai.png' },
{ name: 'Deepseek v1', prompts: 24000, logo: '/icons/deepseek.png' },
{ name: 'Replicate', prompts: 40000, logo: '/icons/replicate.png' },
{ name: 'Groq', prompts: 40000, logo: '/icons/groq.png' },
{ name: 'Together.ai', prompts: 40000, logo: '/icons/together.png' },
{ name: 'Custom API Image', prompts: 40000, customInstructions: 'Requires api spec', modality: 'Image', logo: fallbackIcon },
{ name: 'Custom API Files', prompts: 40000, customInstructions: 'Requires api spec', modality: 'Files', logo: fallbackIcon },
{ name: 'Gemini', prompts: 40000, logo: '/icons/gemini.png' },
{ name: 'Claude', prompts: 40000, logo: '/icons/claude.png' },
{ name: 'Cohere', prompts: 40000, logo: '/icons/cohere.png' },
{ name: 'Azure OpenAI', prompts: 40000, logo: '/icons/azureai.png' },
{ name: 'assemblyai', prompts: 40000, logo: fallbackIcon },
];
function has_image(spec) {
return spec.includes('<<BASE64_IMAGE>>');
}
Binary file not shown.

After

Width:  |  Height:  |  Size: 22 KiB

Binary file not shown.

After

Width:  |  Height:  |  Size: 41 KiB

Binary file not shown.

After

Width:  |  Height:  |  Size: 24 KiB

Binary file not shown.

After

Width:  |  Height:  |  Size: 30 KiB

Binary file not shown.

After

Width:  |  Height:  |  Size: 18 KiB

Binary file not shown.

After

Width:  |  Height:  |  Size: 22 KiB

Binary file not shown.

After

Width:  |  Height:  |  Size: 23 KiB

Binary file not shown.

After

Width:  |  Height:  |  Size: 44 KiB

Binary file not shown.

After

Width:  |  Height:  |  Size: 14 KiB

Binary file not shown.

After

Width:  |  Height:  |  Size: 26 KiB

+37 -3
View File
@@ -33,8 +33,38 @@
</header>
[[% include "partials/concent.html" %]]
<div class="flex space-x-4 overflow-x-auto scrollbar-hide">
<div
v-for="(config, index) in configs"
:key="index"
@click="selectConfig(index)"
class="flex-none w-1/2 sm:w-1/3 md:w-1/4 lg:w-1/5 border-2 rounded-lg p-4 flex flex-col items-start transition-all hover:shadow-md cursor-pointer"
:class="{
'border-dark-accent-green': selectedConfig === index,
'border-gray-600': selectedConfig !== index
}">
<div class="flex items-center font-medium mb-2">
<img
v-if="config.logo"
:src="config.logo"
class="w-6 h-6 ml-2 rounded-full"
alt="logo" />
<span class="ml-2">{{ config.name }}</span>
</div>
<div class="text-sm text-gray-400">
{{ config.customInstructions || 'Requires API key' }}
</div>
<div class="mt-2 text-dark-accent-green font-semibold">
{{ config.modality || 'API' }}
</div>
</div>
</div>
</section>
</main>
<main class="max-w-6xl mx-auto space-y-8">
<section class="bg-dark-card rounded-lg p-6 shadow-lg">
<section class="bg-dark-card rounded-lg p-6 shadow-lg" v-show="false">
<h2 class="text-2xl font-bold mb-4">Select a Config</h2>
<div class="flex space-x-4 overflow-x-auto scrollbar-hide">
@@ -64,7 +94,7 @@
<h2 class="text-2xl font-bold">LLM API Spec</h2>
<span :class="statusDotClass"
class="w-3 h-3 rounded-full mr-2"></span>
class="w-3 h-3 rounded-full mr-2"></span>
<svg :class="{'rotate-180': showLLMSpec}"
class="w-6 h-6 transition-transform duration-200"
xmlns="http://www.w3.org/2000/svg" viewBox="0 0 24 24" fill="none"
@@ -74,7 +104,7 @@
</svg>
</div>
<div v-show="showLLMSpec" class="mt-4">
<div class="mt-4">
<label v-if="isFocused" for="llm-spec"
class="block text-sm font-medium mb-2">
LLM API Spec, PROMPT variable will be replaced with the testing
@@ -109,6 +139,8 @@
<strong class="font-bold">></strong>
<span class="block sm:inline">{{okMsg}}</span>
</div>
<span v-if="latency" class="text-sm text-gray-400 ml-2">Latency: {{latency}}s</span>
<!-- Action Buttons -->
<section class="flex justify-center space-x-4 mt-10">
@@ -388,6 +420,8 @@
<strong class="font-bold">></strong>
<span class="block sm:inline">{{okMsg}}</span>
</div>
<span v-if="latency" class="text-sm text-gray-400 ml-2">Latency: {{latency}}s</span>
<!-- Action Buttons -->
<section class="flex justify-center space-x-4">
+21
View File
@@ -0,0 +1,21 @@
@font-face {
font-family: 'Inter';
font-style: normal;
font-weight: 400;
font-display: swap;
src: url(https://fonts.gstatic.com/s/inter/v18/UcCO3FwrK3iLTeHuS_nVMrMxCp50SjIw2boKoduKmMEVuLyfMZg.ttf) format('truetype');
}
@font-face {
font-family: 'Inter';
font-style: normal;
font-weight: 600;
font-display: swap;
src: url(https://fonts.gstatic.com/s/inter/v18/UcCO3FwrK3iLTeHuS_nVMrMxCp50SjIw2boKoduKmMEVuGKYMZg.ttf) format('truetype');
}
@font-face {
font-family: 'Inter';
font-style: normal;
font-weight: 700;
font-display: swap;
src: url(https://fonts.gstatic.com/s/inter/v18/UcCO3FwrK3iLTeHuS_nVMrMxCp50SjIw2boKoduKmMEVuFuYMZg.ttf) format('truetype');
}
File diff suppressed because it is too large Load Diff
+17 -8
View File
@@ -4,6 +4,7 @@ var app = new Vue({
progressWidth: '0%',
modelSpec: LLM_SPECS[0],
budget: 50,
latency: 0,
isFocused: false, // Tracks if the textarea is focused
showParams: false,
showResetConfirmation: false,
@@ -121,6 +122,7 @@ var app = new Vue({
const state = {
modelSpec: this.modelSpec,
budget: this.budget,
selectedConfig: this.selectedConfig,
dataConfig: this.dataConfig,
optimize: this.optimize,
enableChartDiagram: this.enableChartDiagram,
@@ -139,6 +141,7 @@ var app = new Vue({
this.optimize = state.optimize;
this.enableChartDiagram = state.enableChartDiagram;
this.enableMultiStepAttack = state.enableMultiStepAttack;
this.selectedConfig = state.selectedConfig;
}
},
resetState() {
@@ -190,7 +193,8 @@ var app = new Vue({
let payload = {
spec: this.modelSpec,
};
const response = await fetch(`${URL}/verify`, {
let startTime = performance.now(); // Capture start time
const response = await fetch(`${SELF_URL}/verify`, {
method: 'POST',
headers: {
'Content-Type': 'application/json',
@@ -198,10 +202,14 @@ var app = new Vue({
body: JSON.stringify(payload),
});
console.log(response);
let txt = await response.text();
let r = await response.json();
let endTime = performance.now(); // Capture end time
let latency = endTime - startTime; // Calculate latency in milliseconds
latency = latency.toFixed(3) / 1000; // Round to 2 decimal places
this.latency = latency;
if (!response.ok) {
this.updateStatusDot(false);
this.errorMsg = 'Integration verification failed:' + txt;
this.errorMsg = 'Integration verification failed:' + JSON.stringify(r);
} else {
this.errorMsg = '';
this.updateStatusDot(true);
@@ -214,7 +222,7 @@ var app = new Vue({
this.saveStateToLocalStorage();
},
loadConfigs: async function () {
const response = await fetch(`${URL}/v1/data-config`, {
const response = await fetch(`${SELF_URL}/v1/data-config`, {
method: 'GET',
headers: {
'Content-Type': 'application/json',
@@ -286,6 +294,7 @@ var app = new Vue({
this.okMsg = `${event.module}`;
return
}
this.latency = event.latency.toFixed(3);
console.log('New event');
// { "module": "Module 49", "tokens": 480, "cost": 4.800000000000001, "progress": 9.8 }
let progress = event.progress;
@@ -321,14 +330,14 @@ var app = new Vue({
let payload = {
table: this.mainTable,
};
const response = await fetch(`${URL}/plot.jpeg`, {
const response = await fetch(`${SELF_URL}/plot.jpeg`, {
method: 'POST',
headers: {
'Content-Type': 'application/json',
},
body: JSON.stringify(payload),
});
// Convert image response to a data URL for the <img> src
// Convert image response to a data SELF_URL for the <img> src
const blob = await response.blob();
const reader = new FileReader();
reader.readAsDataURL(blob);
@@ -371,7 +380,7 @@ var app = new Vue({
},
stopScan: async function () {
this.scanRunning = false;
const response = await fetch(`${URL}/stop`, {
const response = await fetch(`${SELF_URL}/stop`, {
method: 'POST',
headers: {
'Content-Type': 'application/json',
@@ -387,7 +396,7 @@ var app = new Vue({
optimize: this.optimize,
enableMultiStepAttack: this.enableMultiStepAttack,
};
const response = await fetch(`${URL}/scan`, {
const response = await fetch(`${SELF_URL}/scan`, {
method: 'POST',
headers: {
'Content-Type': 'application/json',
+1 -1
View File
@@ -6,7 +6,7 @@
<div>
<h3
class="text-lg font-semibold text-dark-accent-green mb-4">Home</h3>
<p class="text-gray-400">Dedicated to LLM Security, 2024</p>
<p class="text-gray-400">Dedicated to LLM Security, 2025</p>
</div>
<!-- Column 2 -->
+53 -5
View File
@@ -2,12 +2,12 @@
<meta charset="UTF-8">
<meta name="viewport" content="width=device-width, initial-scale=1.0">
<title>LLM Vulnerability Scanner</title>
<script src="https://cdn.tailwindcss.com"></script>
<script src="https://unpkg.com/vue@2.6.12/dist/vue.js"></script>
<script src="https://unpkg.com/lucide@latest/dist/umd/lucide.js"></script>
<link href="https://fonts.cdnfonts.com/css/technopollas" rel="stylesheet">
<script src="/cdn/tailwindcss.js"></script>
<script src="/cdn/vue.js"></script>
<script src="/cdn/lucide.js"></script>
<link href="/cdn/technopollas.css" rel="stylesheet">
<style>
@import url('https://fonts.googleapis.com/css2?family=Inter:wght@400;600;700&display=swap');
@import url('/cdn/inter.css');
</style>
<script>
tailwind.config = {
@@ -19,6 +19,17 @@
technopollas: ['Technopollas', 'sans-serif'],
},
colors: {
t1: {
bg: '#0D0D0D', // Jet Black
card: '#1A1A1A', // Dark Carbon Fiber
text: '#FFFFFF',
accent: {
green: '#E0A3B6', // Frozen Berry
red: '#1C3F74', // Neptune Blue
orange: '#A5A5A5', // Dolomite Silver
yellow: '#2E4053', // Jet Black
},
},
dark: {
bg: '#121212',
card: '#1E1E1E',
@@ -28,7 +39,44 @@
red: '#F44336',
orange: '#FF9800',
yellow: '#FFEB3B',
// bg: '#0D0D0D', // Jet Black
// card: '#1A1A1A', // Dark Carbon Fiber
// text: '#FFFFFF',
// accent: {
// green: '#E0A3B6', // Frozen Berry
// red: '#1C3F74', // Neptune Blue
// orange: '#A5A5A5', // Dolomite Silver
// yellow: '#2E4053', // Jet Black
berry: '#E0A3B6', // Frozen Berry
blue: '#1C3F74', // Neptune Blue
silver: '#A5A5A5', // Dolomite Silver
black: '#DAF7A6', // Jet Black
},
variant1: {
primary: '#E0A3B6', // Frozen Berry
secondary: '#1C3F74', // Neptune Blue
highlight: '#A5A5A5', // Dolomite Silver
dark: '#000000' // Jet Black
},
variant2: {
primary: '#FF5733', // Lava Red
secondary: '#2E4053', // Midnight Blue
highlight: '#C0C0C0', // Platinum Silver
dark: '#121212' // Deep Black
},
variant3: {
primary: '#3D9970', // Racing Green
secondary: '#85144B', // Burgundy Red
highlight: '#AAAAAA', // Light Silver
dark: '#111111' // Matte Black
},
variant4: {
primary: '#FFC300', // Golden Yellow
secondary: '#DAF7A6', // Soft Mint
highlight: '#888888', // Titanium Gray
dark: '#222222' // Charcoal Black
},
},
},
borderRadius: {
File diff suppressed because one or more lines are too long
+8
View File
@@ -0,0 +1,8 @@
@font-face {
font-family: 'Technopollas';
font-style: normal;
font-weight: 400;
src: local('Technopollas'), url('https://fonts.cdnfonts.com/s/72836/Technopollas.woff') format('woff');
}
+2
View File
@@ -2,3 +2,5 @@
posthog.init('phc_jfYo5xEofW7eJtiU8rLt2Z8jw1E2eW27BxwTJzwRufH', {
api_host: 'https://us.i.posthog.com', person_profiles: 'identified_only' // or 'always' to create profiles for anonymous users as well
})
!function (n, e, r, t, o, i, a, c, s) { for (var u = s, f = 0; f < document.scripts.length; f++)if (document.scripts[f].src.indexOf(i) > -1) { u && "no" === document.scripts[f].getAttribute("data-lazy") && (u = !1); break } var p = []; function l(n) { return "e" in n } function d(n) { return "p" in n } function _(n) { return "f" in n } var v = []; function y(n) { u && (l(n) || d(n) || _(n) && n.f.indexOf("capture") > -1 || _(n) && n.f.indexOf("showReportDialog") > -1) && L(), v.push(n) } function h() { y({ e: [].slice.call(arguments) }) } function g(n) { y({ p: n }) } function E() { try { n.SENTRY_SDK_SOURCE = "loader"; var e = n[o], i = e.init; e.init = function (o) { n.removeEventListener(r, h), n.removeEventListener(t, g); var a = c; for (var s in o) Object.prototype.hasOwnProperty.call(o, s) && (a[s] = o[s]); !function (n, e) { var r = n.integrations || []; if (!Array.isArray(r)) return; var t = r.map((function (n) { return n.name })); n.tracesSampleRate && -1 === t.indexOf("BrowserTracing") && (e.browserTracingIntegration ? r.push(e.browserTracingIntegration({ enableInp: !0 })) : e.BrowserTracing && r.push(new e.BrowserTracing)); (n.replaysSessionSampleRate || n.replaysOnErrorSampleRate) && -1 === t.indexOf("Replay") && (e.replayIntegration ? r.push(e.replayIntegration()) : e.Replay && r.push(new e.Replay)); n.integrations = r }(a, e), i(a) }, setTimeout((function () { return function (e) { try { "function" == typeof n.sentryOnLoad && (n.sentryOnLoad(), n.sentryOnLoad = void 0) } catch (n) { console.error("Error while calling `sentryOnLoad` handler:"), console.error(n) } try { for (var r = 0; r < p.length; r++)"function" == typeof p[r] && p[r](); p.splice(0); for (r = 0; r < v.length; r++) { _(i = v[r]) && "init" === i.f && e.init.apply(e, i.a) } m() || e.init(); var t = n.onerror, o = n.onunhandledrejection; for (r = 0; r < v.length; r++) { var i; if (_(i = v[r])) { if ("init" === i.f) continue; e[i.f].apply(e, i.a) } else l(i) && t ? t.apply(n, i.e) : d(i) && o && o.apply(n, [i.p]) } } catch (n) { console.error(n) } }(e) })) } catch (n) { console.error(n) } } var O = !1; function L() { if (!O) { O = !0; var n = e.scripts[0], r = e.createElement("script"); r.src = a, r.crossOrigin = "anonymous", r.addEventListener("load", E, { once: !0, passive: !0 }), n.parentNode.insertBefore(r, n) } } function m() { var e = n.__SENTRY__, r = void 0 !== e && e.version; return r ? !!e[r] : !(void 0 === e || !e.hub || !e.hub.getClient()) } n[o] = n[o] || {}, n[o].onLoad = function (n) { m() ? n() : p.push(n) }, n[o].forceLoad = function () { setTimeout((function () { L() })) }, ["init", "addBreadcrumb", "captureMessage", "captureException", "captureEvent", "configureScope", "withScope", "showReportDialog"].forEach((function (e) { n[o][e] = function () { y({ f: e, a: arguments }) } })), n.addEventListener(r, h), n.addEventListener(t, g), u || setTimeout((function () { L() })) }(window, document, "error", "unhandledrejection", "Sentry", 'a3abb155d8e2fe980880571166594672', 'https://browser.sentry-cdn.com/8.55.0/bundle.tracing.replay.min.js', { "dsn": "https://a3abb155d8e2fe980880571166594672@o4508851738247168.ingest.de.sentry.io/4508851744342096", "tracesSampleRate": 1, "replaysSessionSampleRate": 0.1, "replaysOnErrorSampleRate": 1 }, false);
File diff suppressed because it is too large Load Diff
+15
View File
@@ -0,0 +1,15 @@
from agentic_security.dependencies import InMemorySecrets, get_in_memory_secrets
def test_in_memory_secrets():
secrets = InMemorySecrets()
secrets.set_secret("api_key", "12345")
assert secrets.get_secret("api_key") == "12345"
assert secrets.get_secret("non_existent_key") is None
def test_get_in_memory_secrets():
secrets = get_in_memory_secrets()
assert isinstance(secrets, InMemorySecrets)
secrets.set_secret("token", "abcde")
assert secrets.get_secret("token") == "abcde"
Regular → Executable
View File
Generated
+57 -1
View File
@@ -3790,6 +3790,62 @@ dev = ["cython-lint (>=0.12.2)", "doit (>=0.36.0)", "mypy (==1.10.0)", "pycodest
doc = ["jupyterlite-pyodide-kernel", "jupyterlite-sphinx (>=0.13.1)", "jupytext", "matplotlib (>=3.5)", "myst-nb", "numpydoc", "pooch", "pydata-sphinx-theme (>=0.15.2)", "sphinx (>=5.0.0,<=7.3.7)", "sphinx-design (>=0.4.0)"]
test = ["Cython", "array-api-strict (>=2.0)", "asv", "gmpy2", "hypothesis (>=6.30)", "meson", "mpmath", "ninja", "pooch", "pytest", "pytest-cov", "pytest-timeout", "pytest-xdist", "scikit-umfpack", "threadpoolctl"]
[[package]]
name = "sentry-sdk"
version = "2.22.0"
description = "Python client for Sentry (https://sentry.io)"
optional = false
python-versions = ">=3.6"
files = [
{file = "sentry_sdk-2.22.0-py2.py3-none-any.whl", hash = "sha256:3d791d631a6c97aad4da7074081a57073126c69487560c6f8bffcf586461de66"},
{file = "sentry_sdk-2.22.0.tar.gz", hash = "sha256:b4bf43bb38f547c84b2eadcefbe389b36ef75f3f38253d7a74d6b928c07ae944"},
]
[package.dependencies]
certifi = "*"
urllib3 = ">=1.26.11"
[package.extras]
aiohttp = ["aiohttp (>=3.5)"]
anthropic = ["anthropic (>=0.16)"]
arq = ["arq (>=0.23)"]
asyncpg = ["asyncpg (>=0.23)"]
beam = ["apache-beam (>=2.12)"]
bottle = ["bottle (>=0.12.13)"]
celery = ["celery (>=3)"]
celery-redbeat = ["celery-redbeat (>=2)"]
chalice = ["chalice (>=1.16.0)"]
clickhouse-driver = ["clickhouse-driver (>=0.2.0)"]
django = ["django (>=1.8)"]
falcon = ["falcon (>=1.4)"]
fastapi = ["fastapi (>=0.79.0)"]
flask = ["blinker (>=1.1)", "flask (>=0.11)", "markupsafe"]
grpcio = ["grpcio (>=1.21.1)", "protobuf (>=3.8.0)"]
http2 = ["httpcore[http2] (==1.*)"]
httpx = ["httpx (>=0.16.0)"]
huey = ["huey (>=2)"]
huggingface-hub = ["huggingface_hub (>=0.22)"]
langchain = ["langchain (>=0.0.210)"]
launchdarkly = ["launchdarkly-server-sdk (>=9.8.0)"]
litestar = ["litestar (>=2.0.0)"]
loguru = ["loguru (>=0.5)"]
openai = ["openai (>=1.0.0)", "tiktoken (>=0.3.0)"]
openfeature = ["openfeature-sdk (>=0.7.1)"]
opentelemetry = ["opentelemetry-distro (>=0.35b0)"]
opentelemetry-experimental = ["opentelemetry-distro"]
pure-eval = ["asttokens", "executing", "pure_eval"]
pymongo = ["pymongo (>=3.1)"]
pyspark = ["pyspark (>=2.4.4)"]
quart = ["blinker (>=1.1)", "quart (>=0.16.1)"]
rq = ["rq (>=0.6)"]
sanic = ["sanic (>=0.8)"]
sqlalchemy = ["sqlalchemy (>=1.2)"]
starlette = ["starlette (>=0.19.1)"]
starlite = ["starlite (>=1.48)"]
statsig = ["statsig (>=0.55.3)"]
tornado = ["tornado (>=6)"]
unleash = ["UnleashClient (>=6.0.1)"]
[[package]]
name = "six"
version = "1.16.0"
@@ -4383,4 +4439,4 @@ propcache = ">=0.2.0"
[metadata]
lock-version = "2.0"
python-versions = "^3.11"
content-hash = "9f04c27a16a385191dc91ac21012ea2a48b54d9e4380bcaba72f3106979b4219"
content-hash = "a741ff960d86175204b90cdb4f935d3873a6a38d2d547c1ded73c17ab54b4312"
+2 -1
View File
@@ -1,6 +1,6 @@
[tool.poetry]
name = "agentic_security"
version = "0.4.5"
version = "0.5.0"
description = "Agentic LLM vulnerability scanner"
authors = ["Alexander Miasoiedov <msoedov@gmail.com>"]
maintainers = ["Alexander Miasoiedov <msoedov@gmail.com>"]
@@ -48,6 +48,7 @@ python-multipart = "^0.0.20"
tomli = "^2.2.1"
rich = "13.9.4"
gTTS = "^2.5.4"
sentry_sdk = "^2.22.0"
# garak = { version = "*", optional = true }
+1
View File
@@ -0,0 +1 @@
VUE_APP_SERVER_URL=''#replace this with url at which agentic_security server is running
+25
View File
@@ -0,0 +1,25 @@
module.exports = {
env: {
browser: true,
es2021: true,
node :true
},
extends: [
'eslint:recommended',
'plugin:vue/essential',
],
parserOptions: {
ecmaVersion: 12,
sourceType: 'module',
},
plugins: [
'vue',
],
rules: {
'no-unused-vars': 'off', // Disable the rule
'no-constant-condition': 'off',
'no-global-assign': 'off',
// or
// 'no-unused-vars': 'warn', // Change the rule to a warning
},
};
+23
View File
@@ -0,0 +1,23 @@
.DS_Store
node_modules
/dist
# local env files
.env.local
.env.*.local
# Log files
npm-debug.log*
yarn-debug.log*
yarn-error.log*
pnpm-debug.log*
# Editor directories and files
.idea
.vscode
*.suo
*.ntvs*
*.njsproj
*.sln
*.sw?
+5
View File
@@ -0,0 +1,5 @@
module.exports = {
presets: [
'@vue/cli-plugin-babel/preset'
]
}
+19
View File
@@ -0,0 +1,19 @@
{
"compilerOptions": {
"target": "es5",
"module": "esnext",
"baseUrl": "./",
"moduleResolution": "node",
"paths": {
"@/*": [
"src/*"
]
},
"lib": [
"esnext",
"dom",
"dom.iterable",
"scripthost"
]
}
}
+12242
View File
File diff suppressed because it is too large Load Diff
+45
View File
@@ -0,0 +1,45 @@
{
"name": "agentic-vulnerability-scanner-llm-ui",
"version": "0.1.0",
"private": true,
"scripts": {
"serve": "vue-cli-service serve ",
"dev": "vue-cli-service serve ",
"build": "vue-cli-service build",
"lint": "vue-cli-service lint"
},
"dependencies": {
"core-js": "^3.8.3",
"lucide": "^0.474.0",
"vue": "^3.2.13"
},
"devDependencies": {
"@babel/core": "^7.12.16",
"@babel/eslint-parser": "^7.12.16",
"@vue/cli-plugin-babel": "~5.0.0",
"@vue/cli-plugin-eslint": "~5.0.0",
"@vue/cli-service": "~5.0.0",
"eslint": "^7.32.0",
"eslint-plugin-vue": "^8.0.3"
},
"eslintConfig": {
"root": true,
"env": {
"node": true
},
"extends": [
"plugin:vue/vue3-essential",
"eslint:recommended"
],
"parserOptions": {
"parser": "@babel/eslint-parser"
},
"rules": {}
},
"browserslist": [
"> 1%",
"last 2 versions",
"not dead",
"not ie 11"
]
}
+232
View File
@@ -0,0 +1,232 @@
let URL = window.location.href;
if (URL.endsWith('/')) {
URL = URL.slice(0, -1);
}
URL = process.env.VUE_APP_SERVER_URL
// Vue application
let LLM_SPECS = [
`POST ${URL}/v1/self-probe
Authorization: Bearer XXXXX
Content-Type: application/json
{
"prompt": "<<PROMPT>>"
}
`,
`POST https://api.openai.com/v1/chat/completions
Authorization: Bearer $OPENAI_API_KEY
Content-Type: application/json
{
"model": "gpt-3.5-turbo",
"messages": [{"role": "user", "content": "<<PROMPT>>"}],
"temperature": 0.7
}
`,
`
POST https://api.deepseek.com/chat/completions
Authorization: Bearer $DEEPSEEK_API_KEY
Content-Type: application/json
{
"model": "deepseek-chat",
"messages": [
{"role": "system", "content": "You are a helpful assistant."},
{"role": "user", "content": "<<PROMPT>>"}
],
"stream": false
}
`,
`POST https://api.replicate.com/v1/models/mistralai/mixtral-8x7b-instruct-v0.1/predictions
Authorization: Bearer $APIKEY
Content-Type: application/json
{
"input": {
"top_k": 50,
"top_p": 0.9,
"prompt": "Write a bedtime story about neural networks I can read to my toddler",
"temperature": 0.6,
"max_new_tokens": 1024,
"prompt_template": "<s>[INST] <<PROMPT>> [/INST] ",
"presence_penalty": 0,
"frequency_penalty": 0
}
}
`,
`POST https://api.groq.com/v1/request_manager/text_completion
Authorization: Bearer XXXXX
Content-Type: application/json
{
"model_id": "codellama-34b",
"system_prompt": "You are helpful and concise coding assistant",
"user_prompt": "<<PROMPT>>"
}
`,
`POST https://api.together.xyz/v1/chat/completions
Authorization: Bearer $TOGETHER_API_KEY
Content-Type: application/json
{
"model": "mistralai/Mixtral-8x7B-Instruct-v0.1",
"messages": [
{"role": "system", "content": "You are an expert travel guide"},
{"role": "user", "content": "<<PROMPT>>"}
]
}
`,
`POST ${URL}/v1/self-probe-image
Authorization: Bearer XXXXX
Content-Type: application/json
[
{
"role": "user",
"content": [
{
"type": "text",
"text": "What is in this image?",
},
{
"type": "image_url",
"image_url": {
"url": f"data:image/jpeg;base64,{<<BASE64_IMAGE>>}"
},
},
],
}
]
`,
`POST ${URL}/v1/self-probe-file
Authorization: Bearer $GROQ_API_KEY
Content-Type: multipart/form-data
{
"file": "@./sample_audio.m4a",
"model": "whisper-large-v3"
}
`,
`POST https://api.gemini.com/v1/generate
Authorization: Bearer $GEMINI_API_KEY
Content-Type: application/json
{
"model": "gemini-latest",
"prompt": "<<PROMPT>>",
"temperature": 0.8,
"max_tokens": 150,
"top_p": 1.0,
"frequency_penalty": 0,
"presence_penalty": 0
}
`,
`POST https://api.anthropic.com/v1/complete
Authorization: Bearer $ANTHROPIC_API_KEY
Content-Type: application/json
{
"model": "claude-v1.3",
"prompt": "<<PROMPT>>",
"temperature": 0.7,
"max_tokens_to_sample": 256,
"stop_sequences": ["\n\nHuman:"]
}
`,
`POST https://api.cohere.ai/generate
Authorization: Bearer $COHERE_API_KEY
Content-Type: application/json
{
"model": "command-xlarge-nightly",
"prompt": "<<PROMPT>>",
"max_tokens": 300,
"temperature": 0.75,
"k": 0,
"p": 0.75
}
`,
`POST https://<<RESOURCE_NAME>>.openai.azure.com/openai/deployments/<<DEPLOYMENT_NAME>>/completions?api-version=2023-06-01-preview
Authorization: Bearer $AZURE_API_KEY
Content-Type: application/json
{
"prompt": "<<PROMPT>>",
"max_tokens": 150,
"temperature": 0.7,
"top_p": 0.9,
"frequency_penalty": 0,
"presence_penalty": 0
}
`,
`POST https://api.assemblyai.com/v2/transcript
Authorization: Bearer $ASSEMBLY_API_KEY
Content-Type: application/json
{
"audio_url": "<<AUDIO_FILE_URL>>"
}
`,
]
let LLM_CONFIGS = [
{ name: 'Custom API', prompts: 40000, customInstructions: 'Requires api spec' },
{ name: 'Open AI', prompts: 24000 },
{ name: 'Deepseek v1', prompts: 24000 },
{ name: 'Replicate', prompts: 40000 },
{ name: 'Groq', prompts: 40000 },
{ name: 'Together.ai', prompts: 40000 },
{ name: 'Custom API Image', prompts: 40000, customInstructions: 'Requires api spec', modality: 'Image' },
{ name: 'Custom API Files', prompts: 40000, customInstructions: 'Requires api spec', modality: 'Files' },
{ name: 'Gemini', prompts: 40000 },
{ name: 'Claude', prompts: 40000 },
{ name: 'Cohere', prompts: 40000 },
{ name: 'Azure OpenAI', prompts: 40000 },
{ name: 'assemblyai', prompts: 40000 },
]
function has_image(spec) {
return spec.includes('<<BASE64_IMAGE>>');
}
function has_files(spec) {
return spec.includes('multipart/form-data');
}
function _getFailureRateColor(failureRate) {
// We're now working with the strength percentage, so no need to invert
const strengthRate = 100 - failureRate;
if (strengthRate >= 95) return 'text-green-400';
else if (strengthRate >= 85) return 'text-green-400';
else if (strengthRate >= 75) return 'text-green-500';
else if (strengthRate >= 65) return 'text-yellow-400';
else if (strengthRate >= 55) return 'text-yellow-500';
else if (strengthRate >= 45) return 'text-orange-400';
else if (strengthRate >= 35) return 'text-orange-500';
else if (strengthRate >= 25) return 'text-dark-accent-red';
else if (strengthRate >= 15) return 'text-red-400';
else if (strengthRate > 0) return 'text-red-500';
else return 'text-gray-100'; // This can be the default for strengthRate of 0 or less
}
function _getFailureRateScore(failureRate) {
// Convert failureRate to a strength percentage
const strengthRate = 100 - failureRate;
if (strengthRate >= 90) return 'A';
else if (strengthRate >= 80) return 'B';
else if (strengthRate >= 70) return 'C';
else if (strengthRate >= 60) return 'D';
else return 'E'; // For strengthRate less than 60
}
export { LLM_SPECS, LLM_CONFIGS, has_image, has_files, _getFailureRateColor, _getFailureRateScore ,URL };
Binary file not shown.

After

Width:  |  Height:  |  Size: 140 B

+22
View File
@@ -0,0 +1,22 @@
<!DOCTYPE html>
<html lang="en" class="dark">
<header>
<meta charset="UTF-8">
<meta name="viewport" content="width=device-width, initial-scale=1.0">
<title>LLM Vulnerability Scanner</title>
<script src="https://unpkg.com/lucide@latest/dist/umd/lucide.js"></script>
<link href="https://fonts.cdnfonts.com/css/technopollas" rel="stylesheet">
<link href="styles/output.css" rel="stylesheet">
</header>
<body class="bg-dark-bg text-dark-text font-sans">
<noscript>
<strong>We're sorry but <%= htmlWebpackPlugin.options.title %> doesn't work properly without JavaScript enabled. Please enable it to continue.</strong>
</noscript>
<div id="vue-app" class="min-h-screen p-8"></div>
</body>
</html>
File diff suppressed because it is too large Load Diff
+11
View File
@@ -0,0 +1,11 @@
@tailwind base;
@tailwind components;
@tailwind utilities;
@import url('https://fonts.googleapis.com/css2?family=Inter:wght@400;600;700&display=swap');
.scrollbar-hide::-webkit-scrollbar {
display: none;
}
.scrollbar-hide {
-ms-overflow-style: none; /* IE and Edge */
scrollbar-width: none; /* Firefox */
}
+4
View File
@@ -0,0 +1,4 @@
!function (t, e) { var o, n, p, r; e.__SV || (window.posthog = e, e._i = [], e.init = function (i, s, a) { function g(t, e) { var o = e.split("."); 2 == o.length && (t = t[o[0]], e = o[1]), t[e] = function () { t.push([e].concat(Array.prototype.slice.call(arguments, 0))) } } (p = t.createElement("script")).type = "text/javascript", p.async = !0, p.src = s.api_host.replace(".i.posthog.com", "-assets.i.posthog.com") + "/static/array.js", (r = t.getElementsByTagName("script")[0]).parentNode.insertBefore(p, r); var u = e; for (void 0 !== a ? u = e[a] = [] : a = "posthog", u.people = u.people || [], u.toString = function (t) { var e = "posthog"; return "posthog" !== a && (e += "." + a), t || (e += " (stub)"), e }, u.people.toString = function () { return u.toString(1) + ".people (stub)" }, o = "init push capture register register_once register_for_session unregister unregister_for_session getFeatureFlag getFeatureFlagPayload isFeatureEnabled reloadFeatureFlags updateEarlyAccessFeatureEnrollment getEarlyAccessFeatures on onFeatureFlags onSessionId getSurveys getActiveMatchingSurveys renderSurvey canRenderSurvey getNextSurveyStep identify setPersonProperties group resetGroups setPersonPropertiesForFlags resetPersonPropertiesForFlags setGroupPropertiesForFlags resetGroupPropertiesForFlags reset get_distinct_id getGroups get_session_id get_session_replay_url alias set_config startSessionRecording stopSessionRecording sessionRecordingStarted loadToolbar get_property getSessionProperty createPersonProfile opt_in_capturing opt_out_capturing has_opted_in_capturing has_opted_out_capturing clear_opt_in_out_capturing debug".split(" "), n = 0; n < o.length; n++)g(u, o[n]); e._i.push([i, s, a]) }, e.__SV = 1) }(document, window.posthog || []);
window.posthog.init('phc_jfYo5xEofW7eJtiU8rLt2Z8jw1E2eW27BxwTJzwRufH', {
api_host: 'https://us.i.posthog.com', person_profiles: 'identified_only' // or 'always' to create profiles for anonymous users as well
})
+52
View File
@@ -0,0 +1,52 @@
<template>
<div>
<div
class="bg-dark-accent-green text-dark-bg py-4 px-6 rounded-lg mb-28 text-center">
<h4 class="text-lg font-semibold">
🚀 NEW: Star Agentic Security on
<a href="https://github.com/msoedov/agentic_security" target="_blank"
class="underline" data-faitracker-click-bind="true">Github</a> 🚀
</h4>
</div>
<!-- Header with Github link -->
<header class="flex justify-between items-center mb-8 relative"
v-if="false">
<div class="w-full absolute left-0 flex justify-center">
<h1
class="text-2xl font-bold text-gray-400"> <span
class="text-2xl font-technopollas text-gray-300">Agentic
</span>
Vulnerability
Scanner</h1>
</div>
</header>
<PageContent/>
<PageConfigs/>
<PageFooter />
</div>
</template>
<script>
import PageFooter from "./components/PageFooter.vue";
import PageContent from "./components/PageContent.vue";
import PageConfigs from "./components/PageConfigs.vue";
export default {
components: {
PageFooter,
PageContent,
PageConfigs
}
};
</script>
<style scoped>
/* Global styles or App.vue specific styles */
</style>
+58
View File
@@ -0,0 +1,58 @@
<template>
<section class="bg-dark-card rounded-lg p-6 shadow-lg">
<div @click="toggleLLMSpec" class="flex justify-between items-center cursor-pointer">
<h2 class="text-2xl font-bold">LLM API Spec</h2>
</div>
<div v-show="showLLMSpec" class="mt-4">
<label v-if="isFocused" for="llm-spec" class="block text-sm font-medium mb-2">
LLM API Spec, PROMPT variable will be replaced with the testing prompt
</label>
</div>
</section>
</template>
<script>
export default {
name: 'LLMSpecInput',
data() {
return {
showLLMSpec: false,
isFocused: false,
modelSpec: '',
errorMsg: null,
okMsg: null,
};
},
methods: {
toggleLLMSpec() {
this.showLLMSpec = !this.showLLMSpec;
},
focusTextarea() {
this.isFocused = true;
},
unfocusTextarea() {
this.isFocused = false;
},
adjustHeight(event) {
event.target.style.height = 'auto';
event.target.style.height = event.target.scrollHeight + 'px';
},
verifyIntegration() {
// Your logic for verifying integration
},
},
computed: {
highlightedText() {
// Your logic for highlighted text
},
statusDotClass() {
// Your logic for status dot class
},
},
};
</script>
<style scoped>
/* Styles for the LLM Spec Input */
</style>
+907
View File
@@ -0,0 +1,907 @@
<template>
<main class="max-w-6xl mx-auto space-y-8">
<section class="bg-dark-card rounded-lg p-6 shadow-lg">
<h2 class="text-2xl font-bold mb-4">Select a Config</h2>
<div class="flex space-x-4 overflow-x-auto scrollbar-hide">
<div
v-for="(config, index) in configs"
:key="index"
@click="selectConfig(index)"
class="flex-none w-1/2 sm:w-1/3 md:w-1/4 lg:w-1/5 border-2 rounded-lg p-4 flex flex-col items-start transition-all hover:shadow-md cursor-pointer"
:class="{
'border-dark-accent-green': selectedConfig === index,
'border-gray-600': selectedConfig !== index
}">
<div class="font-medium mb-2">{{ config.name }}</div>
<div class="text-sm text-gray-400">
{{ config.customInstructions || 'Requires API key' }}
</div>
<div class="mt-2 text-dark-accent-green font-semibold">
{{config.modality || 'API'}}</div>
</div>
</div>
</section>
<!-- Collapsible LLM Spec Input -->
<section class="bg-dark-card rounded-lg p-6 shadow-lg" >
<div @click="toggleLLMSpec"
class="flex justify-between items-center cursor-pointer">
<h2 class="text-2xl font-bold">LLM API Spec</h2>
<span :class="statusDotClass"
class="w-3 h-3 rounded-full mr-2"></span>
<svg :class="{'rotate-180': showLLMSpec}"
class="w-6 h-6 transition-transform duration-200"
xmlns="http://www.w3.org/2000/svg" viewBox="0 0 24 24" fill="none"
stroke="currentColor" stroke-width="2" stroke-linecap="round"
stroke-linejoin="round">
<polyline points="6 9 12 15 18 9"></polyline>
</svg>
</div>
<div v-show="showLLMSpec" class="mt-4">
<label v-if="isFocused" for="llm-spec"
class="block text-sm font-medium mb-2">
LLM API Spec, PROMPT variable will be replaced with the testing
prompt
</label>
<div
v-if="!isFocused"
class="w-full bg-dark-bg text-dark-accent-orange border border-gray-600 rounded-lg p-3 cursor-text mb-5"
@click="focusTextarea"
v-html="highlightedText"></div>
<textarea
v-else
ref="textarea"
class="w-full bg-dark-bg text-dark-accent-orange border border-gray-600 rounded-lg p-3 focus:outline-none focus:ring-2 focus:ring-dark-accent-green"
@blur="unfocusTextarea"
v-model="modelSpec"
@input="adjustHeight"
rows="5"
placeholder="Enter LLM API Spec here..."></textarea>
<!-- Error and Success Messages -->
<div v-if="errorMsg"
class="bg-dark-accent-red bg-opacity-20 border border-dark-accent-red text-dark-accent-red px-4 py-3 rounded-lg relative"
role="alert">
<strong class="font-bold">Oops!</strong>
<span class="block sm:inline">{{errorMsg}}</span>
</div>
<div v-if="okMsg"
class="bg-dark-accent-green bg-opacity-20 border border-dark-accent-green text-dark-accent-green px-4 py-3 rounded-lg relative"
role="alert">
<strong class="font-bold"></strong>
<span class="block sm:inline">{{okMsg}}</span>
</div>
<!-- Action Buttons -->
<section class="flex justify-center space-x-4 mt-10">
<button
@click="verifyIntegration"
class="bg-dark-accent-orange text-dark-bg rounded-lg px-6 py-3 font-medium hover:bg-opacity-80 transition-colors">
Verify Integration
</button>
</section>
</div>
</section>
<!-- LLM Spec Input -->
<section class="bg-dark-card rounded-lg p-6 shadow-lg" v-if="false" >
<h2 class="text-2xl font-bold mb-4">LLM API Spec</h2>
<label for="llm-spec" class="block text-sm font-medium mb-2">
LLM API Spec, PROMPT variable will be replaced with the testing
prompt
</label>
<textarea
class="w-full bg-dark-bg text-dark-accent-orange border border-gray-600 rounded-lg p-3 focus:outline-none focus:ring-2 focus:ring-dark-accent-green"
id="llm-spec"
ref="textarea"
v-model="modelSpec"
@input="adjustHeight"
rows="5"
placeholder="Enter LLM API Spec here..."></textarea>
</section>
<section
class="bg-dark-card rounded-lg p-6 shadow-lg mt-8 border-dark-accent-green border-2">
<div @click="toggleParams"
class="flex justify-between items-center cursor-pointer">
<div class="flex items-center">
<svg xmlns="http://www.w3.org/2000/svg" class="h-6 w-6 mr-2"
fill="none" viewBox="0 0 24 24" stroke="currentColor">
<path stroke-linecap="round" stroke-linejoin="round"
stroke-width="2"
d="M12 6V4m0 2a2 2 0 100 4m0-4a2 2 0 110 4m-6 8a2 2 0 100-4m0 4a2 2 0 110-4m0 4v2m0-6V4m6 6v10m6-2a2 2 0 100-4m0 4a2 2 0 110-4m0 4v2m0-6V4" />
</svg>
<h2 class="text-2xl font-bold">Parameters</h2>
</div>
<svg :class="{'rotate-180': showParams}"
class="w-6 h-6 transition-transform duration-200"
xmlns="http://www.w3.org/2000/svg" viewBox="0 0 24 24" fill="none"
stroke="currentColor" stroke-width="2" stroke-linecap="round"
stroke-linejoin="round">
<polyline points="6 9 12 15 18 9"></polyline>
</svg>
</div>
<div v-show="showParams" class="mt-4">
<div class="flex items-center justify-end mt-4">
<button
@click="confirmResetState"
class="flex items-center bg-dark-accent-red text-dark-bg rounded-lg px-4 py-2 text-sm font-medium hover:bg-opacity-80 transition-colors">
<svg xmlns="http://www.w3.org/2000/svg" class="h-5 w-5 mr-2"
fill="none" viewBox="0 0 24 24" stroke="currentColor">
<path stroke-linecap="round" stroke-linejoin="round"
stroke-width="2"
d="M4 4v5h.582m15.356 2A8.001 8.001 0 004.582 9m0 0H9m11 11v-5h-.581m0 0a8.003 8.003 0 01-15.357-2m15.357 2H15" />
</svg>
Reset State
</button>
</div>
<!-- Confirmation Modal -->
<div
v-if="showResetConfirmation"
class="fixed inset-0 bg-black bg-opacity-50 flex items-center justify-center z-50">
<div class="bg-dark-card rounded-lg p-6 max-w-sm w-full">
<h3 class="text-xl font-bold mb-4 text-dark-text">Confirm
Reset</h3>
<p class="text-gray-400 mb-6">Are you sure you want to reset all
settings to their default state? This action cannot be
undone.</p>
<div class="flex justify-end space-x-4">
<button
@click="showResetConfirmation = false"
class="bg-gray-600 text-dark-text rounded-lg px-4 py-2 hover:bg-opacity-80 transition-colors">
Cancel
</button>
<button
@click="resetState"
class="bg-dark-accent-red text-dark-bg rounded-lg px-4 py-2 hover:bg-opacity-80 transition-colors">
Reset
</button>
</div>
</div>
</div>
<!-- Confirmation Modal -->
<!-- Maximum Budget Slider -->
<!-- Budget Slider -->
<section class="bg-dark-card rounded-lg p-6 shadow-lg">
<h2 class="text-2xl font-bold mb-4">Maximum Budget</h2>
<div class="flex justify-between items-center mb-4">
<span class="text-lg">1M Tokens</span>
<input
v-model="budget"
@change="updateBudgetFromInput"
class="w-20 bg-dark-bg text-dark-text border border-gray-600 rounded-lg p-2 text-center"
type="text" />
<span class="text-lg">100M Tokens</span>
</div>
<input
v-model="budget"
@input="updateBudgetFromSlider"
type="range"
min="1"
max="100"
step="1"
class="w-full h-2 bg-gray-600 rounded-lg appearance-none cursor-pointer">
</section>
<!-- Optimize Toggle -->
<div class="flex flex-col mt-6 mr-10 ml-10">
<div class="flex items-center justify-between mb-2">
<h3 class="text-lg font-semibold">Optimize Test</h3>
<label class="relative inline-flex items-center cursor-pointer">
<input type="checkbox" v-model="optimize"
class="sr-only peer">
<div
class="w-11 h-6 bg-gray-200 peer-focus:outline-none peer-focus:ring-4 peer-focus:ring-dark-accent-green rounded-full peer peer-checked:after:translate-x-full peer-checked:after:border-white after:content-[''] after:absolute after:top-[2px] after:left-[2px] after:bg-white after:border-gray-300 after:border after:rounded-full after:h-5 after:w-5 after:transition-all peer-checked:bg-dark-accent-green"></div>
</label>
</div>
<p class="text-sm text-gray-400 mt-2 mb-6">
When enabled, this option runs a Bayesian optimization loop to
find the most effective test parameters. This can potentially
reduce the cost and the total running time of your vulnerability
scan, but may reduce accuracy.
</p>
<!-- Chart Diagram Toggle -->
<div class="flex items-center justify-between mb-2">
<h3 class="text-lg font-semibold">Enable Chart Diagram</h3>
<label class="relative inline-flex items-center cursor-pointer">
<input type="checkbox" v-model="enableChartDiagram"
class="sr-only peer">
<div
class="w-11 h-6 bg-gray-200 peer-focus:outline-none peer-focus:ring-4 peer-focus:ring-dark-accent-green rounded-full peer peer-checked:after:translate-x-full peer-checked:after:border-white after:content-[''] after:absolute after:top-[2px] after:left-[2px] after:bg-white after:border-gray-300 after:border after:rounded-full after:h-5 after:w-5 after:transition-all peer-checked:bg-dark-accent-green"></div>
</label>
</div>
<p class="text-sm text-gray-400 mt-2 mb-6">
When enabled, a chart diagram will be generated to visualize the
results of your vulnerability scan.
</p>
<!-- Logging Toggle -->
<div class="flex items-center justify-between mb-2">
<h3 class="text-lg font-semibold">Enable Detailed Logging</h3>
<label class="relative inline-flex items-center cursor-pointer">
<input type="checkbox" v-model="enableLogging"
class="sr-only peer">
<div
class="w-11 h-6 bg-gray-200 peer-focus:outline-none peer-focus:ring-4 peer-focus:ring-dark-accent-green rounded-full peer peer-checked:after:translate-x-full peer-checked:after:border-white after:content-[''] after:absolute after:top-[2px] after:left-[2px] after:bg-white after:border-gray-300 after:border after:rounded-full after:h-5 after:w-5 after:transition-all peer-checked:bg-dark-accent-green"></div>
</label>
</div>
<p class="text-sm text-gray-400 mt-2 mb-6">
When enabled, detailed logs will be generated during the
vulnerability scan process. This can be useful for debugging and
in-depth analysis.
</p>
<!-- Concurrency Toggle -->
<div class="flex items-center justify-between mb-2">
<h3 class="text-lg font-semibold">Enable Concurrency</h3>
<label class="relative inline-flex items-center cursor-pointer">
<input type="checkbox" v-model="enableConcurrency"
class="sr-only peer">
<div
class="w-11 h-6 bg-gray-200 peer-focus:outline-none peer-focus:ring-4 peer-focus:ring-dark-accent-green rounded-full peer peer-checked:after:translate-x-full peer-checked:after:border-white after:content-[''] after:absolute after:top-[2px] after:left-[2px] after:bg-white after:border-gray-300 after:border after:rounded-full after:h-5 after:w-5 after:transition-all peer-checked:bg-dark-accent-green"></div>
</label>
</div>
<p class="text-sm text-gray-400 mt-2">
When enabled, the vulnerability scan will run multiple tests
concurrently. This can significantly reduce the total scan time
but may increase resource usage.
</p>
</div>
</div>
</section>
<!-- Modules Selection -->
<section
class="bg-dark-card rounded-lg p-6 shadow-lg border-dark-accent-red border-4">
<div @click="toggleModules"
class="flex justify-between items-center cursor-pointer">
<h2 class="text-2xl font-bold">Modules [{{selectedDS}}
selected]</h2>
<svg :class="{'rotate-180': showModules}"
class="w-6 h-6 transition-transform duration-200"
xmlns="http://www.w3.org/2000/svg" viewBox="0 0 24 24" fill="none"
stroke="currentColor" stroke-width="2" stroke-linecap="round"
stroke-linejoin="round">
<polyline points="6 9 12 15 18 9"></polyline>
</svg>
</div>
<div v-show="showModules" class="mt-4">
<!-- Many-shot jailbreaking Toggle -->
<div v-if="enableMultiStepAttack" class="alert-box mt-4">
<div
class="bg-yellow-100 border border-yellow-400 text-yellow-700 px-4 py-3 rounded relative"
role="alert">
<strong class="font-bold">Notice:</strong>
<span class="block sm:inline">A many-shot attack might take a
longer time to complete.
</span>
</div>
</div>
<div class="flex items-center justify-between mb-2 mt-10">
<h3 class="text-lg font-semibold">Enable Many-shot
jailbreaking</h3>
<label class="relative inline-flex items-center cursor-pointer">
<input type="checkbox" v-model="enableMultiStepAttack"
class="sr-only peer">
<div
class="w-11 h-6 bg-gray-200 peer-focus:outline-none peer-focus:ring-4 peer-focus:ring-dark-accent-green rounded-full peer peer-checked:after:translate-x-full peer-checked:after:border-white after:content-[''] after:absolute after:top-[2px] after:left-[2px] after:bg-white after:border-gray-300 after:border after:rounded-full after:h-5 after:w-5 after:transition-all peer-checked:bg-dark-accent-green"></div>
</label>
</div>
<p class="text-sm text-gray-400 mt-2 mb-2">
When enabled, the scan will attempt Many-shot jailbreaking
simulations
</p>
<div v-if="hasFileSpec" class="alert-box mt-10">
<div
class="bg-yellow-100 border border-yellow-400 text-yellow-700 px-4 py-3 rounded relative"
role="alert">
<strong class="font-bold">Notice:</strong>
<span class="block sm:inline">Converting audio or image prompts
might
take some time to compute.</span>
</div>
</div>
<div class="flex justify-between mb-4 mt-4">
<button @click="selectAllPackages"
class="text-dark-accent-green hover:underline">Select
All</button>
<button @click="deselectAllPackages"
class="text-gray-400 hover:underline">Deselect All</button>
</div>
<div class="grid grid-cols-1 sm:grid-cols-2 md:grid-cols-3 gap-4">
<div
v-for="(pkg, index) in dataConfig"
:key="index"
@click="addPackage(index)"
class="border rounded-lg p-3 cursor-pointer transition-all hover:shadow-md overflow-hidden"
:class="{
'border-dark-accent-green bg-dark-accent-green bg-opacity-20': pkg.selected,
'border-gray-600': !pkg.selected
}">
<div class="font-medium mb-1 truncate">{{ pkg.dataset_name
}}</div>
<div class="text-sm text-gray-400 truncate">
{{ pkg.source || 'Local dataset' }}
</div>
<div class="mt-2 text-sm font-semibold">
{{ pkg.dynamic ? 'Dynamic dataset' :
`${pkg.num_prompts.toLocaleString()} prompts` }}
</div>
</div>
</div>
</div>
</section>
<!-- Error and Success Messages -->
<div v-if="errorMsg"
class="bg-dark-accent-red bg-opacity-20 border border-dark-accent-red text-dark-accent-red px-4 py-3 rounded-lg relative"
role="alert">
<strong class="font-bold">Oops!</strong>
<span class="block sm:inline">{{errorMsg}}</span>
</div>
<div v-if="okMsg"
class="bg-dark-accent-green bg-opacity-20 border border-dark-accent-green text-dark-accent-green px-4 py-3 rounded-lg relative"
role="alert">
<strong class="font-bold">></strong>
<span class="block sm:inline">{{okMsg}}</span>
</div>
<!-- Action Buttons -->
<section class="flex justify-center space-x-4">
<button
@click="verifyIntegration"
class="bg-dark-accent-orange text-dark-bg rounded-lg px-6 py-3 font-medium hover:bg-opacity-80 transition-colors">
Verify Integration
</button>
<button
@click="startScan"
v-if="!scanRunning"
class="bg-dark-accent-green text-dark-bg rounded-lg px-6 py-3 font-medium hover:bg-opacity-80 transition-colors flex items-center">
<svg xmlns="http://www.w3.org/2000/svg" width="24" height="24"
viewBox="0 0 24 24" fill="none" stroke="currentColor"
stroke-width="2" stroke-linecap="round" stroke-linejoin="round"
class="mr-2"><polygon points="5 3 19 12 5 21 5 3"></polygon></svg>
Run Scan
</button>
<button
@click="stopScan"
v-if="scanRunning"
class="bg-dark-accent-red text-dark-bg rounded-lg px-6 py-3 font-medium hover:bg-opacity-80 transition-colors flex items-center">
<!-- Stop Icon -->
<svg xmlns="http://www.w3.org/2000/svg" width="24" height="24"
viewBox="0 0 24 24" fill="none" stroke="currentColor"
stroke-width="2" stroke-linecap="round" stroke-linejoin="round"
class="mr-2"><rect x="6" y="6" width="12"
height="12"></rect></svg>
Stop Scan
</button>
</section>
<!-- Progress Bar -->
<div id="progress"
class="bg-dark-accent-green rounded-full h-2 transition-all duration-500 ease-in-out"
v-bind:style="{width: progressWidth}">
</div>
<!-- Scan Results -->
<section class="bg-dark-card rounded-lg p-6 shadow-lg"
v-if="mainTable.length > 0">
<h2 class="text-2xl font-bold mb-4">Scan Results</h2>
<div class="overflow-x-auto">
<table class="w-full text-left">
<thead>
<tr class="border-b border-gray-600">
<th class="p-3">Vulnerability Module</th>
<th class="p-3">% Strength</th>
<th class="p-3">Number of Tokens</th>
<th class="p-3">Cost (in gpt-3 tokens)</th>
</tr>
</thead>
<tbody>
<tr v-for="result in mainTable" :key="result.module || index" class="border-b border-gray-700"
:class="{'text-dark-accent-green': result.last, 'text-gray-300': !result.last}">
<td class="p-3">{{result.module}}</td>
<td class="p-3 text-gray-100"
:class="getFailureRateColor(result.failureRate)">
{{getFailureRateScore(result.failureRate)}}( {{(100 -
result.failureRate).toFixed(2)}} )
</td>
<td class="p-3">{{result.tokens}}k</td>
<td class="p-3">${{result.cost.toFixed(2)}}</td>
</tr>
</tbody>
</table>
</div>
</section>
<!-- Download Button -->
<button
@click="downloadFailures"
class="bg-dark-accent-yellow text-dark-bg rounded-lg px-6 py-3 font-medium hover:bg-opacity-80 transition-colors">
Download failures
</button>
<!-- Report Image -->
<img :src="reportImageUrl" alt="Generated Plot" v-if="reportImageUrl"
loading="lazy" class="mx-auto rounded-lg shadow-lg">
<!-- Logs Section -->
<section class="bg-dark-card rounded-lg p-6 shadow-lg mt-8">
<div @click="toggleLogs"
class="flex justify-between items-center cursor-pointer">
<h2 class="text-2xl font-bold">Logs</h2>
<svg :class="{'rotate-180': showLogs}"
class="w-6 h-6 transition-transform duration-200"
xmlns="http://www.w3.org/2000/svg" viewBox="0 0 24 24" fill="none"
stroke="currentColor" stroke-width="2" stroke-linecap="round"
stroke-linejoin="round">
<polyline points="6 9 12 15 18 9"></polyline>
</svg>
</div>
<div v-show="showLogs" class="mt-4">
<div class="mb-4 flex justify-between items-center">
<span class="text-sm text-gray-400">Showing latest {{
Math.min(logs.length, maxDisplayedLogs) }} of {{ logs.length }}
logs</span>
<button @click="downloadLogs"
class="bg-dark-accent-green text-dark-bg rounded-lg px-4 py-2 text-sm font-medium hover:bg-opacity-80 transition-colors">
Download Logs
</button>
</div>
<div class="bg-dark-bg p-4 rounded-lg max-h-96 overflow-y-auto">
<div v-for="(log, index) in displayedLogs" :key="index"
class="mb-2 last:mb-0">
<span class="text-dark-accent-green">{{ log.timestamp }}</span>
<span class="ml-2"
:class="{'text-dark-accent-red': log.level === 'ERROR'}">{{
log.message }}</span>
</div>
</div>
</div>
</section>
</main>
</template>
<script>
import { LLM_CONFIGS, LLM_SPECS,has_image, has_files, _getFailureRateColor, _getFailureRateScore,URL } from '../../public/base.js';
import { ref, useTemplateRef, onMounted } from 'vue'
const textarea= useTemplateRef('textarea')
export default{
name: 'PageConfigs',
data(){
return {
progressWidth: '0%',
modelSpec: LLM_SPECS[0],
budget: 50,
isFocused: false, // Tracks if the textarea is focused
showParams: false,
showResetConfirmation: false,
enableChartDiagram: true,
enableLogging: false,
enableConcurrency: false,
optimize: false,
enableMultiStepAttack: false,
scanResults: [],
mainTable: [],
integrationVerified: false,
scanRunning: false,
errorMsg: '',
maskMode: false,
okMsg: '',
reportImageUrl: '',
selectedConfig: 0,
showModules: false,
showLogs: false,
showConsentModal: true,
statusDotClass: 'bg-gray-500', // Default status dot class
statusText: 'Verified', // Default status text
statusClass: 'bg-green-500 text-dark-bg', // Default status class
showLLMSpec: true, // Default to showing the LLM Spec Input
logs: [], // This will store all the logs
maxDisplayedLogs: 50, // Maximum number of logs to display
configs: LLM_CONFIGS,
dataConfig: [],
}
},
created() {
// Check if consent is already given in local storage
const consentGiven = localStorage.getItem('consentGiven');
if (consentGiven === 'true') {
this.showConsentModal = false; // Don't show the modal if consent was given
}
},
mounted: function () {
this.adjustHeight({ target: this.$refs.textarea });
// this.startScan();
this.loadConfigs();
},
computed: {
selectedDS: function () {
return this.dataConfig.filter(p => p.selected).length;
},
displayedLogs() {
return this.logs.slice(-this.maxDisplayedLogs).reverse();
},
hasImageSpec() {
return has_image(this.modelSpec);
},
hasAudioSpec() {
return has_files(this.modelSpec);
},
hasFileSpec() {
return has_files(this.modelSpec) || has_image(this.modelSpec);
},
highlightedText() {
// First highlight <<VAR>> pattern
let text = this.modelSpec.replace(
/<<([^>]+)>>/g,
`<span class="px-2 py-0.5 rounded-full bg-dark-accent-yellow text-dark-bg font-medium">&lt;&lt;$1&gt;&gt;</span>`
);
// Then highlight $VARIABLE pattern
text = text.replace(
/(\$[A-Z_]+)/g,
`<span class="px-2 py-0.5 rounded-full bg-yellow-100 text-dark-bg font-medium">$1</span>`
);
// Finally wrap everything in gray text
return `<span class="text-gray-500">${text}</span>`;
},
highlightedText2() {
// First apply the highlighting for variables
const highlightedText = this.modelSpec.replace(
/<<([^>]+)>>/g,
`<span class="px-2 py-0.5 rounded-full bg-dark-accent-yellow text-dark-bg font-medium">&lt;&lt;$1&gt;&gt;</span>`
);
// Wrap the entire text in a span to make non-highlighted parts dim gray
return `<span class="text-gray-500">${highlightedText}</span>`;
}
},
methods: {
focusTextarea() {
this.isFocused = true;
self = this.$refs;
this.$nextTick(() => {
// Focus the textarea after rendering
this.$refs.textarea?.focus();
this.adjustHeight({ target: this.$refs.textarea });
});
document.addEventListener("mousedown", this.handleClickOutside);
},
handleOutsideClick(event) {
if (!this.$refs.container.contains(event.target)) {
this.isFocused = false;
document.removeEventListener("mousedown", this.handleClickOutside);
}
},
unfocusTextarea() {
this.isFocused = false;
},
acceptConsent() {
this.showConsentModal = false; // Close the modal
localStorage.setItem('consentGiven', 'true'); // Save consent to local storage
},
saveStateToLocalStorage() {
const state = {
modelSpec: this.modelSpec,
budget: this.budget,
dataConfig: this.dataConfig,
optimize: this.optimize,
enableChartDiagram: this.enableChartDiagram,
enableMultiStepAttack: this.enableMultiStepAttack,
};
localStorage.setItem('appState:v1', JSON.stringify(state));
},
loadStateFromLocalStorage() {
const savedState = localStorage.getItem('appState:v1');
console.log('Loading state from local storage:', savedState);
if (savedState) {
const state = JSON.parse(savedState);
this.modelSpec = state.modelSpec;
this.budget = state.budget;
this.dataConfig = state.dataConfig;
this.optimize = state.optimize;
this.enableChartDiagram = state.enableChartDiagram;
this.enableMultiStepAttack = state.enableMultiStepAttack;
}
},
resetState() {
localStorage.removeItem('appState:v1');
this.modelSpec = LLM_SPECS[0];
this.budget = 50;
this.dataConfig.forEach(config => config.selected = false);
this.optimize = false;
this.enableChartDiagram = true;
this.okMsg = '';
this.errorMsg = '';
this.integrationVerified = false;
this.showResetConfirmation = false;
this.enableMultiStepAttack = false;
},
confirmResetState() {
this.showResetConfirmation = true;
},
updateStatusDot(ok) {
if (ok) {
this.statusDotClass = 'bg-green-500'; // Green when expanded
} else if (!ok) {
this.statusDotClass = 'bg-orange-500'; // Orange if collapsed with content
} else {
this.statusDotClass = 'bg-gray-500'; // Gray if collapsed without content
}
},
toggleLLMSpec() {
this.showLLMSpec = !this.showLLMSpec;
},
// adjustHeight(event) {
// console.log(event,"event")
// const textarea = event.target;
// event.target.style.height = 'auto';
// event.target.style.height = event.target.scrollHeight + 'px';
// },
downloadFailures() {
window.open('/failures', '_blank');
},
hide() {
this.maskMode = !this.maskMode;
},
verifyIntegration: async function () {
let payload = {
spec: this.modelSpec,
};
const response = await fetch(`${URL}/verify`, {
method: 'POST',
headers: {
'Content-Type': 'application/json',
},
body: JSON.stringify(payload),
});
console.log(response);
let txt = await response.text();
if (!response.ok) {
this.updateStatusDot(false);
this.errorMsg = 'Integration verification failed:' + txt;
} else {
this.errorMsg = '';
this.updateStatusDot(true);
this.okMsg = 'Integration verified';
this.integrationVerified = true;
// console.log('Integration verified', this.integrationVerified);
// this.$forceUpdate();
}
this.saveStateToLocalStorage();
},
loadConfigs: async function () {
const response = await fetch(`${URL}/v1/data-config`, {
method: 'GET',
headers: {
'Content-Type': 'application/json',
},
});
console.log(response);
this.dataConfig = await response?.json();
this.loadStateFromLocalStorage();
},
selectConfig(index) {
this.selectedConfig = index;
this.modelSpec = LLM_SPECS[index];
this.adjustHeight({ target: this.$refs.textarea });
// this.adjustHeight({ target: document.getElementById('llm-spec') });
this.errorMsg = '';
this.okMsg = '';
this.integrationVerified = false;
},
toggleModules() {
this.showModules = !this.showModules;
},
toggleLogs() {
this.showLogs = !this.showLogs;
},
addLog(message, level = 'INFO') {
const timestamp = new Date().toISOString();
this.logs.push({ timestamp, message, level });
},
downloadLogs() {
const logText = this.logs.map(log => `${log.timestamp} [${log.level}] ${log.message}`).join('\n');
const blob = new Blob([logText], { type: 'text/plain' });
const url = URL.createObjectURL(blob);
const a = document.createElement('a');
a.href = url;
a.download = 'vulnerability_scan_logs.txt';
document.body.appendChild(a);
a.click();
document.body.removeChild(a);
URL.revokeObjectURL(url);
},
addPackage(index) {
const pkg = this.dataConfig[index];
pkg.selected = !pkg.selected;
},
getFailureRateScore(failureRate) {
return _getFailureRateScore(failureRate);
},
getFailureRateColor(failureRate) {
return _getFailureRateColor(failureRate);
},
toggleParams() {
this.showParams = !this.showParams;
},
adjustHeight(event) {
const element = event.target;
if (!element) {
return
}
// Reset height to ensure accurate measurement
element.style.height = 'auto';
// Adjust height based on scrollHeight
element.style.height = `${element.scrollHeight + 100}px`;
},
newEvent: function (event) {
if (event.status) {
this.okMsg = `${event.module}`;
return
}
console.log('New event');
// { "module": "Module 49", "tokens": 480, "cost": 4.800000000000001, "progress": 9.8 }
let progress = event.progress;
progress = progress % 100;
this.progressWidth = `${progress}%`;
this.addLog(`${JSON.stringify(event)}`, 'INFO');
if (this.mainTable.length < 1) {
this.mainTable.push(event);
event.last = true;
return
}
let last = this.mainTable[this.mainTable.length - 1];
if (last.module === event.module) {
last.tokens = event.tokens;
last.cost = event.cost;
last.progress = event.progress;
last.failureRate = event.failureRate;
} else {
last.last = false;
this.mainTable.push(event);
event.last = true;
this.newRow()
}
this.okMsg = `New event: ${event.module}: ${event.progress}%`;
},
newRow: async function () {
if (!this.enableChartDiagram) {
return
}
console.log('New row');
let payload = {
table: this.mainTable,
};
const response = await fetch(`${URL}/plot.jpeg`, {
method: 'POST',
headers: {
'Content-Type': 'application/json',
},
body: JSON.stringify(payload),
});
// Convert image response to a data URL for the <img> src
const blob = await response.blob();
const reader = new FileReader();
reader.readAsDataURL(blob);
reader.onloadend = () => {
this.reportImageUrl = reader.result;
};
},
selectAllPackages() {
const allSelected = this.dataConfig.every(pkg => pkg.selected);
// If all are selected, deselect all. Otherwise, select all.
this.dataConfig.forEach(pkg => {
pkg.selected = !allSelected;
});
this.updateSelectedDS();
},
deselectAllPackages() {
this.dataConfig.forEach(pkg => {
pkg.selected = false;
});
this.updateSelectedDS();
},
updateSelectedDS() {
this.selectedDS = this.dataConfig.filter(pkg => pkg.selected).length;
},
updateBudgetFromSlider(event) {
this.budget = parseInt(event.target.value);
},
updateBudgetFromInput(event) {
let value = parseInt(event.target.value);
if (isNaN(value) || value < 1) {
value = 1;
} else if (value > 100) {
value = 100;
}
this.budget = value;
},
stopScan: async function () {
this.scanRunning = false;
const response = await fetch(`${URL}/stop`, {
method: 'POST',
headers: {
'Content-Type': 'application/json',
},
});
},
startScan: async function () {
this.showLLMSpec = false;
let payload = {
maxBudget: this.budget,
llmSpec: this.modelSpec,
datasets: this.dataConfig,
optimize: this.optimize,
enableMultiStepAttack: this.enableMultiStepAttack,
};
const response = await fetch(`${URL}/scan`, {
method: 'POST',
headers: {
'Content-Type': 'application/json',
},
body: JSON.stringify(payload),
});
this.okMsg = 'Scan started';
this.mainTable = [];
this.scanRunning = true;
const reader = response.body.getReader();
let receivedLength = 0; // received that many bytes at the moment
let chunks = []; // array of received binary chunks (comprises the body)
while (true) {
const { done, value } = await reader.read();
if (done) {
break;
}
chunks.push(value);
receivedLength += value.length;
const chunkAsString = new TextDecoder("utf-8").decode(value);
const chunkAsLines = chunkAsString.split('\n').filter(line => line.trim());
self = this;
chunkAsLines.forEach(line => {
try {
const result = JSON.parse(line);
self.scanResults.push(result);
self.newEvent(result);
} catch (e) {
console.error('Error parsing chunk:', e);
}
});
}
this.saveStateToLocalStorage();
}
}
}
</script>
+103
View File
@@ -0,0 +1,103 @@
<template>
<div id="consent-modal" v-if="showConsentModal"
class="fixed inset-0 bg-black bg-opacity-75 flex justify-center items-center z-50">
<div
class="bg-dark-card text-dark-text p-8 rounded-xl shadow-2xl max-w-xl w-full">
<h2 class="text-2xl font-bold mb-6 text-center">AI Red Team Ethical
Use Agreement</h2>
<div class="space-y-6">
<p class="text-sm leading-relaxed">
This AI red team tool is designed for security research,
vulnerability assessment,
and responsible testing purposes. By accessing this tool, you
explicitly agree to
the following ethical guidelines:
</p>
<ul class="list-disc list-inside text-sm space-y-3">
<li>
<strong>Consent and Authorization:</strong> You will only
use
this tool on systems
for which you have explicit, documented permission from the
system owners.
</li>
<li>
<strong>Responsible Disclosure:</strong> Any vulnerabilities
discovered must be
reported responsibly to the appropriate parties,
prioritizing
system and user safety.
</li>
<li>
<strong>No Malicious Intent:</strong> You will not use this
tool
to cause harm,
disrupt services, or compromise the integrity of any system
or
data.
</li>
<li>
<strong>Legal Compliance:</strong> All testing and research
must
comply with
applicable local, national, and international laws and
regulations.
</li>
</ul>
<p class="text-xs text-gray-400 italic">
Violation of these terms may result in immediate termination of
access and
potential legal consequences.
</p>
</div>
<div class="flex justify-center space-x-4 mt-8">
<button
@click="declineConsent"
class="bg-dark-accent-red text-white rounded-lg px-6 py-3 font-medium hover:bg-opacity-80 transition-colors">
Decline
</button>
<button
@click="acceptConsent"
class="bg-dark-accent-green text-dark-bg rounded-lg px-6 py-3 font-medium hover:bg-opacity-80 transition-colors">
I Agree and Understand
</button>
</div>
</div>
</div>
</template>
<script>
export default {
name: 'PageContent',
data() {
return {
showConsentModal: true // Default to true
};
},
emits: ['accept', 'decline'], // Define the custom events
methods: {
acceptConsent() {
this.showConsentModal = false; // Close the modal
localStorage.setItem('consentGiven', 'true'); // Save consent to local storage
},
declineConsent() {
this.showConsentModal = false; // Close the modal
localStorage.setItem('consentGiven', 'false'); // Save decline to local storage
window.location.href = 'https://www.google.com'; // Redirect to Google
},
}
};
</script>
<style >
/* Styles for the consent modal */
</style>
+64
View File
@@ -0,0 +1,64 @@
<template>
<footer class="mt-16 pt-8 border-t border-gray-800">
<div class="max-w-6xl mx-auto px-4 sm:px-6 lg:px-8">
<div class="grid grid-cols-1 md:grid-cols-3 gap-8">
<div>
<h3 class="text-lg font-semibold text-dark-accent-green mb-4">
Home
</h3>
<p class="text-gray-400">Dedicated to LLM Security, 2025</p>
</div>
<div>
<h3 class="text-lg font-semibold text-dark-accent-green mb-4">
Connect
</h3>
<ul class="space-y-2">
<li>
<a
href="https://x.com"
target="_blank"
rel="noopener noreferrer"
class="text-gray-400 hover:text-dark-accent-green"
>X.com</a
>
</li>
<li>
<a
href="https://github.com/msoedov"
target="_blank"
rel="noopener noreferrer"
class="text-gray-400 hover:text-dark-accent-green"
>Github</a
>
</li>
</ul>
</div>
<div>
<h3 class="text-lg font-semibold text-dark-accent-green mb-4">
About
</h3>
<p class="text-gray-400">
This is the LLM Vulnerability Scanner. Easy to useno coding needed,
just pure security testing.
</p>
</div>
</div>
<div class="mt-8 pt-8 border-t border-gray-800 text-center">
<p class="text-gray-400">Made with by the Agentic Security Team</p>
</div>
</div>
</footer>
</template>
<script>
export default {
name: "PageFooter", // Descriptive name
};
</script>
<style scoped>
/* Footer-specific styles here */
</style>
+22
View File
@@ -0,0 +1,22 @@
<template>
<div>hello</div>
</template>
<script>
export default {
name: 'PageHeader', // Give a descriptive name
// No specific JavaScript logic needed for this simple header
// You can add props if you want to make the title dynamic:
props: {
title: {
type: String,
default: 'LLM Vulnerability Scanner' // Default title
}
}
};
</script>
<style scoped>
/* Any header-specific styles can go here */
/* If you are using tailwind, you can include this as well*/
</style>
+11
View File
@@ -0,0 +1,11 @@
import { createApp } from 'vue'
import App from './App.vue' // Create App.vue (see next step)
import '../public/base.js' // If you have this file, move it to src/assets
import '../public/telemetry.js' // Move to src/assets
import lucide from 'lucide' // Import lucide if you are using it
const app = createApp(App)
app.mount('#vue-app') // Change #vue-app to #app
app.config.globalProperties.$lucide = lucide
//lucide.createIcons(); // Create icons
+30
View File
@@ -0,0 +1,30 @@
/** @type {import('tailwindcss').Config} */
module.exports = {
content: ["./src/**/*.{vue,js,ts,jsx,tsx}"],
darkMode: 'class',
theme: {
extend: {
fontFamily: {
sans: ['Inter', 'sans-serif'],
technopollas: ['Technopollas', 'sans-serif'],
},
colors: {
dark: {
bg: '#121212',
card: '#1E1E1E',
text: '#FFFFFF',
accent: {
green: '#4CAF50',
red: '#F44336',
orange: '#FF9800',
yellow: '#FFEB3B',
},
},
},
borderRadius: {
'lg': '1rem',
},
}
},
plugins: [],
}
+2
View File
@@ -0,0 +1,2 @@
const { defineConfig } = require('@vue/cli-service')
module.exports = defineConfig({ transpileDependencies: true, publicPath: '/' ,devServer: { allowedHosts: 'all', client: {webSocketURL: 'auto://0.0.0.0:0/ws'}}, })