dreadnode 1.0.6

Dreadnode SDK

Dreadnode Strikes SDK

Strikes is a platform for building, experimenting with, and evaluating AI security agent code.

• Experiment + Tasking + Observability in a single place that's lightweight and scales.
• Track your data with parameters, inputs, and outputs all connected to your tasks.
• Log your artifacts — data, models, files, and folders — to track data of your Dreadnode runs, enabling easy reuse and reproducibility.
• Measure everything with metrics throughout your code and anywhere you need them.
• Scale your code from a single run to thousands.

import dreadnode as dn
import rigging as rg

from .tools import reversing_tools

dn.configure()

@dataclass
class Finding:
    name: str
    severity: str
    description: str
    exploit_code: str

@dn.scorer(name="Score Finding")
async def score_finding(finding: Finding) -> float:
    if finding.severity == "critical":
        return 1.0
    elif finding.severity == "high":
        return 0.8
    else:
        return 0.2

@dn.task(scorers=[score_finding])
@rg.prompt(tools=[reversing_tools])
async def analyze_binary(binary: str) -> list[Finding]:
    """
    Analyze the binary for vulnerabilities.
    """
    ...

with dn.run(tags=["reverse-engineering"]):
    binary = "c2/downloads/service.exe"

    dn.log_params(
        model="gpt-4",
        temperature=0.5,
        binary=binary
    )

    findings = await analyze_binary(binary)

    dn.log_metric("findings", len(findings))

Installation

We publish every version to PyPi:

pip install -U dreadnode

If you want to build from source:

poetry install

See our installation guide for more options.

Getting Started

Read through our introduction guide in the docs.

Examples

Check out dreadnode/example-agents to find your favorite use case.