Skip to content

DeepTeam is a framework to red team LLMs and AI agents.

.

sid-muralie5492b5bbdc9
diff --git a/deepteam/cli/main.py b/deepteam/cli/main.py
index 00c1e4a6..844a7646 100644
--- a/deepteam/cli/main.py
+++ b/deepteam/cli/main.py
@@ -123,7 +123,11 @@ def _load_config(path: str):
@app.command()
-def run(config_file: str):
+def run(
+ config_file: str,
+ max_concurrent: int = typer.Option(None, "-c", "--max-concurrent", help="Maximum concurrent operations (overrides config)"),
+ attacks_per_vuln: int = typer.Option(None, "-a", "--attacks-per-vuln", help="Number of attacks per vulnerability type (overrides config)")
+):
"""Run a red teaming execution based on a YAML configuration"""
cfg = _load_config(config_file)
config.apply_env()
@@ -137,12 +141,16 @@ def run(config_file: str):
simulator_model = load_model(simulator_model_spec)
evaluation_model = load_model(evaluation_model_spec)
+ # Use CLI options to override config values
+ final_max_concurrent = max_concurrent if max_concurrent is not None else cfg.get("options", {}).get("max_concurrent", 10)
+ final_attacks_per_vuln = attacks_per_vuln if attacks_per_vuln is not None else cfg.get("options", {}).get("attacks_per_vulnerability_type", 1)
+
red_teamer = RedTeamer(
simulator_model=simulator_model,
evaluation_model=evaluation_model,
target_purpose=target.get("purpose", ""),
async_mode=cfg.get("options", {}).get("run_async", True),
- max_concurrent=cfg.get("options", {}).get("max_concurrent", 10),
+ max_concurrent=final_max_concurrent,
)
vulnerabilities_cfg = cfg.get("default_vulnerabilities", [])
@@ -171,9 +179,7 @@ def run(config_file: str):
model_callback=model_callback,
vulnerabilities=vulnerabilities,
attacks=attacks,
- attacks_per_vulnerability_type=cfg.get("options", {}).get(
- "attacks_per_vulnerability_type", 1
- ),
+ attacks_per_vulnerability_type=final_attacks_per_vuln,
ignore_errors=cfg.get("options", {}).get("ignore_errors", False),
)
@@ -182,10 +188,22 @@ def run(config_file: str):
@app.command()
-def login(api_key: str = typer.Argument(..., help="OpenAI API Key")):
- """Store API key for later runs."""
- config.set_key("OPENAI_API_KEY", api_key)
- typer.echo("API key saved.")
+def login(api_key: str = typer.Argument(..., help="API Key (auto-detects provider based on prefix)")):
+ """Store API key for later runs. Auto-detects provider from key prefix."""
+ # Auto-detect provider based on key prefix
+ if api_key.startswith("sk-proj"):
+ config.set_key("OPENAI_API_KEY", api_key)
+ typer.echo("OpenAI API key saved.")
+ elif api_key.startswith("sk-ant-"):
+ config.set_key("ANTHROPIC_API_KEY", api_key)
+ typer.echo("Anthropic API key saved.")
+ elif api_key.startswith("AIza"):
+ config.set_key("GOOGLE_API_KEY", api_key)
+ typer.echo("Google/Gemini API key saved.")
+ else:
+ # Default to OpenAI for backwards compatibility
+ config.set_key("OPENAI_API_KEY", api_key)
+ typer.echo("API key saved as OpenAI (unrecognized prefix).")
@app.command()
diff --git a/deepteam/red_teamer/red_teamer.py b/deepteam/red_teamer/red_teamer.py
index 250129cc..912fb86a 100644
--- a/deepteam/red_teamer/red_teamer.py
+++ b/deepteam/red_teamer/red_teamer.py
@@ -128,7 +128,6 @@ class RedTeamer:
):
# Initialize metric map
metrics_map = self.get_red_teaming_metrics_map(vulnerabilities)
-
# Simulate attacks
if (
reuse_simulated_attacks
@@ -265,7 +264,6 @@ class RedTeamer:
ignore_errors=ignore_errors,
)
)
-
# Create a mapping of vulnerabilities to attacks
vulnerability_type_to_attacks_map: Dict[
VulnerabilityType, List[SimulatedAttack]
diff --git a/examples/sample.yaml b/examples/sample.yaml
index 8dcfa392..06fb9013 100644
--- a/examples/sample.yaml
+++ b/examples/sample.yaml
@@ -5,7 +5,6 @@ target:
model:
provider: openai
model: gpt-3.5-turbo
- temperature: 0.7
simulator_model:
provider: openai
model: gpt-3.5-turbo
@@ -20,9 +19,8 @@ default_vulnerabilities:
custom_vulnerabilities:
- name: "CustomVulnerability"
custom_name: "API Security"
- types: ["endpoint_security"] # Simplified to one type for now
- prompt: |
- I want you to act as a security expert. Based on the purpose, please create a list of API security exploits.
+ types: ["endpoint_security"]
+ prompt:
attacks:
- name: "Prompt Injection"
@@ -32,6 +30,6 @@ attacks:
options:
attacks_per_vulnerability_type: 2
- ignore_errors: true
+ ignore_errors: false
run_async: true
max_concurrent: 5
\ No newline at end of file