How Cyberful Works
Cyberful is an open-source AI-powered red teaming platform designed to automate the discovery, exploitation, verification, and remediation of vulnerabilities. It operates as a multi-agent system that orchestrates 316 cybersecurity tools, including 203 command-line applications such as Nmap, Metasploit, Nuclei, OWASP ZAP, and Ghidra. These tools are integrated into a structured workflow where hypothesis workers map attack surfaces, test exploit paths, and validate the impact of vulnerabilities in isolated environments. A deterministic control plane enforces scope boundaries and preserves evidence, ensuring that only independently reproduced vulnerabilities are reported.
Before initiating a live-target operation, an ephemeral private-network HTTPS canary verifies that essential tools like curl, OpenSSL, Git, and Python’s Requests library can traverse a real ZAP proxy with the engagement’s certificate authority. This step ensures that the environment is properly configured without contacting the target or the internet. Once verified, Cyberful installs and enforces host-scoped rate limits and mandatory public request headers in ZAP before proceeding with target profiling. After each phase, a best-effort passive ZAP checkpoint is recorded, filtered to authorized origins, ensuring compliance with the engagement’s scope.
Cyberful embeds a hardened, agent-controlled Chrome browser fork for isolated browsing tasks, eliminating the need for personal browser profiles. The platform also includes a runtime image with ZAP, Ghidra, Python, Ruby/Bundler, native debugging and fuzzing tools, managed Firefox/Marionette, Xvfb/X11 clipboard testing, and archive extraction capabilities. Docker is used to containerize these tools, but Docker Compose is not required. The system is designed to operate within strict constraints, ensuring that non-web engagements do not trigger unnecessary ZAP checkpoints or block handoffs.
The architecture relies on a deterministic control plane that enforces scope and preserves evidence. This ensures that findings are evidence-backed, independently verifiable, and tied to the authorized engagement. Cyberful does not lower security standards despite its automation capabilities. Instead, it democratizes advanced security work by making it understandable, inspectable, and usable for developers, researchers, and smaller teams. The platform’s design ensures that all actions are traceable, reproducible, and aligned with the engagement’s objectives.
Source launches reuse this snapshot, ensuring consistency without requiring network updates. If no snapshot is available, the system reports DATASETUNAVAILABLE, maintaining operational integrity.
The platform’s configuration and local state management are handled through a settings.yaml file, which defines providers, models, fallback routing, delegation, trusted instructions, and a progressive skill-catalog budget. Provider credentials are stored securely in Cyberful’s protected credential store or named environment variables. Workareas and session logs are stored under work/ and logs/session-logs/, respectively, and must never be committed to version control due to their sensitivity. The first local engagement also prepares a release-pinned CVE Dictionary, which is verified through multiple checks before download to ensure integrity.
Users can start a specific headless workflow with commands like cyberful run --workflow bug-bounty --workarea. The --format json flag generates a redacted, versioned coding-agent stream, while cyberful session inspect allows inspection of durable or attached state. The cyberful session watch command provides a terminal-based user interface (TUI) for supervising attached sessions. Existing sessions can be resumed with cyberful run --continue or cyberful run --session. The platform’s session lifecycle includes steering, approvals, reports, and cleanup, all documented in the official resources.
Cyberful’s skill catalog is extensible, allowing users to integrate custom skills or extend existing ones. The public skills/cyberful Agent Skills package and the SessionInspectionV1 API enable deeper customization and control. The platform’s design ensures that all interactions are logged, auditable, and aligned with the engagement’s scope, providing a robust framework for red teaming, penetration testing, and bug bounty research.
Use Cases
Cyberful is primarily designed for authorized penetration testing, code audits, and bug bounty research. Its multi-agent system and deterministic control plane make it suitable for security teams and researchers who need to systematically identify and validate vulnerabilities. The platform’s ability to orchestrate a wide range of tools—from network scanning (Nmap) to exploit development (Metasploit) and static analysis (Ghidra)—provides comprehensive coverage for various security assessments.
In penetration testing, Cyberful automates the hypothesis-driven exploration of attack surfaces. Hypothesis workers map potential entry points, test exploit paths, and validate the impact of discovered vulnerabilities in isolated environments. This approach reduces the risk of false positives and ensures that only reproducible vulnerabilities are reported. The deterministic control plane enforces scope boundaries, preventing unauthorized or unintended actions while preserving evidence for post-engagement analysis.
For code audits, Cyberful’s integration with tools like Ghidra and OWASP ZAP enables deep static and dynamic analysis of software artifacts. This makes it ideal for identifying vulnerabilities in source code, binaries, or web applications without disrupting ongoing operations. The evidence-backed findings generated by Cyberful are particularly valuable for compliance and remediation efforts.
Bug bounty researchers can leverage Cyberful to streamline their workflows, particularly when dealing with large-scale engagements or complex targets. Researchers can focus on identifying high-impact vulnerabilities while Cyberful handles the orchestration, validation, and reporting processes.
Cyberful’s design also supports integration with other security tools and workflows. For example, the cyberful session inspect and cyberful session watch commands allow users to monitor ongoing sessions in real time, while the --format json flag enables programmatic interaction with the platform’s output. This flexibility makes Cyberful adaptable to various security operations, from large-scale enterprise assessments to niche research projects. The platform’s open-source nature and AGPL 3.0 license encourage collaboration and customization, ensuring that it remains relevant to evolving security challenges.
Key Features and Capabilities
Cyberful’s core strength lies in its ability to orchestrate a diverse set of cybersecurity tools within a structured and controlled environment. The platform integrates 316 tools, including 203 command-line applications, to cover a broad spectrum of security testing needs. These tools are managed through a multi-agent system where hypothesis workers generate and test attack scenarios, while the deterministic control plane enforces scope and preserves evidence. This dual-layer approach ensures that engagements are both thorough and compliant with authorization boundaries.
One of Cyberful’s standout features is its use of ephemeral private-network HTTPS canaries to verify the integrity of the testing environment before live-target operations begin. This step ensures that essential tools like curl, OpenSSL, Git, and Python’s Requests library are properly configured to traverse the ZAP proxy without contacting the target or the internet. Once verified, Cyberful enforces host-scoped rate limits and mandatory public request headers, further ensuring that engagements adhere to predefined constraints. After each phase, a passive ZAP checkpoint is recorded, capturing evidence from authorized origins only.
Cyberful’s support for isolated browsing is another key capability. The platform embeds a hardened, agent-controlled Chrome browser fork, eliminating the need for personal browser profiles and reducing the risk of cross-contamination between engagements. This isolation extends to other components, such as the runtime image, which includes tools like ZAP, Ghidra, Python, Ruby/Bundler, and native debugging and fuzzing utilities. The use of Docker for containerization ensures that these tools operate in a controlled and reproducible manner, while the absence of Docker Compose requirements simplifies deployment.
This snapshot is reused across sessions, ensuring consistency without requiring network updates. If no snapshot is available, the system defaults to reporting DATASETUNAVAILABLE, maintaining operational integrity.
Cyberful’s configuration and state management are designed to balance flexibility and security. The settings.yaml file allows users to define providers, models, fallback routing, and trusted instructions, while provider credentials are stored securely in the platform’s protected credential store or environment variables. Workareas and session logs are stored locally and must never be committed to version control due to their sensitivity. The first local engagement also prepares a release-pinned CVE Dictionary, which is verified through multiple integrity checks before download, ensuring that the data used for vulnerability assessment is accurate and up-to-date.
The platform’s command-line interface (CLI) provides users with fine-grained control over workflows and sessions. Commands like cyberful run --workflow bug-bounty --workarea initiate specific engagements, while cyberful session inspect and cyberful session watch enable real-time monitoring of ongoing sessions. The --format json flag generates structured output for programmatic use, and existing sessions can be resumed with cyberful run --continue or cyberful run --session. This CLI-driven approach ensures that users can tailor Cyberful’s capabilities to their specific needs, whether for penetration testing, code audits, or bug bounty research.
The public skills/cyberful Agent Skills package and the SessionInspectionV1 API allow users to integrate custom skills or extend existing ones.
The platform’s recent updates, such as the introduction of SessionInspectionV1 and cyberful session inspect/watch commands in release v0.4.0, further enhance its usability and control. Changes to the JSONL wire format ensure that output is versioned and structured, while improvements to workflow completion guarantee that reports are delivered accurately and the TUI returns to a ready state.