Claude-Red: Open-Source SKILL.md Library Turning Claude into a Red Team Operator

An open-source offensive security library adhering to Claude's official Skills specification, named 'Claude-Red' (SnailSploit/Claude-Red), has been introduced to equip AI agents with structured penetration testing and red-teaming methodologies. Consisting of modular SKILL.md files that can be dropped directly into an agent harness, the project provides Claude with the operational context, analytical frameworks, and disciplined procedures required to systematically assess target attack surfaces.

Image source: @tom_doerr on X

13 Offensive Security Domains and Structured SKILL.md Architecture

Rather than relying on loosely structured system prompts or monolithic automation scripts, Claude-Red packages offensive tradecraft into modular units compliant with Claude's official Skills standard.

The library organizes its methodologies across 13 distinct offensive domains, providing dozens of specialized procedural modules that an agent can discover and invoke based on the assessment context:

• Web Application Exploitation: Targeted discovery and proof-of-concept workflows covering SQL Injection (SQLi), Cross-Site Scripting (XSS), Server-Side Request Forgery (SSRF), and Remote Code Execution (RCE).
• System and Binary Tradecraft: Low-level shellcode development guides and evasion strategies designed to assess Endpoint Detection and Response (EDR) telemetry.
• Enterprise Infrastructure & Identity: Structured penetration modules for Active Directory environments and multi-tier cloud infrastructure security evaluations.
• Advanced Evaluation & Next-Gen Vectors: Automated fuzzing workflows for vulnerability discovery, alongside specialized modules focused on evaluating the security posture of AI applications.

Each technique is encapsulated within a self-contained SKILL.md file. By dropping these modules into an agent harness, practitioners supply Claude with the procedural depth and domain context of an experienced security consultant, reducing reliance on trial-and-error prompting.

Practical Use Cases: From Red-Team Operations to Defensive Validation

Claude-Red delivers concrete utility across both offensive security engagements and proactive defensive validation pipelines.

For penetration testers and red team operators, the library functions as an intelligent operational assistant. When evaluating complex target environments, Claude can systematically trace prospective attack vectors, synthesize contextual hypotheses, and assist operators in executing rigorous, step-by-step diagnostic workflows. This structure ensures that security audits adhere to methodical investigative frameworks rather than ad-hoc probing.

Simultaneously, the library offers substantial utility to blue teams and security operations engineers seeking automated defense verification:

• Security Policy Efficacy Testing: By instructing an agent to run standardized attack playbooks within segregated staging environments, security teams can measure whether host configurations, firewall rules, and access control lists intercept unauthorized activity as intended.
• Host Containment and Isolation Auditing: Running simulated compromise scenarios enables organizations to conduct automated regression testing against their network segmentation and system isolation boundaries.

Operational Safety: Strict VM Isolation and Execution Governance

Because Claude-Red integrates directly with agentic systems capable of local command execution, deploying these modules requires strict environmental isolation.

SKILL.md modules execute with the full shell permissions and filesystem access granted to the underlying agent. Running these routines on a primary workstation or within networks containing live production assets presents severe operational risks.

• Mandatory Ephemeral VM Isolation: All execution must take place inside dedicated, throwaway virtual machines (throwaway VMs) entirely isolated from production infrastructure and stripped of actual credentials, private keys, or confidential organizational data.
• Pre-Approval Gates and Skill Discovery Governance: Organizations must define explicit governance policies regulating how agents index, auto-load, and execute shell instructions from these skills. Enforcing human-in-the-loop verification or strict policy constraints ensures the agent does not execute destructive actions or exceed intended audit scopes.

Claude-Red illustrates how structured agent skills can turn general-purpose foundation models into specialized technical operators, while emphasizing that high-capability offensive automation must always be paired with rigorous sandboxing and execution controls.

Sources

• GitHub: SnailSploit/Claude-Red Repository
• Tom Doerr on X (@tom_doerr): Claude-Red Introduction Post