Anthropic Releases Cybersecurity Framework Comprising Seven Hundred Fifty-Four Structured Agent Skills
May 30, 2026 · Edited by Oleksandr Kuzmenko
Anthropic has open-sourced a collection of seven hundred fifty-four structured cybersecurity skills mapped to industry standards. These schemas allow developers to build secure, standardized agents for vulnerability detection. They are ready to import into local agent environments.
Why it matters
You can now use Anthropic's secure schemas to quickly build standardized, robust vulnerability auditing subagents into your local continuous integration systems.
Key takeaways
- Import the security skill schemas directly into your local agent tools setup to ensure consistent formatting.
- Link these schema definitions to automated code scanners like Semgrep to handle the actual inspection execution.
- Implement the OWASP-mapped schemas in your pre-release loops to systematically block vulnerabilities before deployment.
Securing applications and scanning for system vulnerabilities has traditionally required manual testing or rigid static analysis tools. Anthropic has released a cybersecurity skills dataset containing seven hundred fifty-four structured skill definitions mapped to five major security frameworks. This collection lets developers equip their AI agents with formal, structured definitions of security procedures, enabling safe vulnerability detection directly in developer deployment pipelines.\n\nUnder the hood, these skill schemas are formatted to align with the tool calling specifications expected by frontier models. Each skill is cataloged with its necessary arguments, structural execution parameters, and safety guidelines. By grounding agent interfaces in standardized formats (such as MITRE ATT&CK or OWASP), the schemas reduce model hallucination rates and establish deterministic operational boundaries for active system tasks.\n\nIf you want to build an automated DevSecOps (Development Security Operations) pipeline using local frameworks or the Claude Agent SDK (Software Development Kit), this release provides a comprehensive library. Instead of writing custom system prompts for automated testing, you can import these pre-configured schemas. For example, you can set up a local subagent to automatically audit newly committed code blocks for injection risks or outdated package configurations during local continuous integration cycles.\n\nHowever, these schemas are definition-only. They specify how a skill should be requested and structured, but they do not provide the underlying security scanning engines. Developers must still hook up the schemas to practical security tools like Nmap, Semgrep, or custom testing scripts to perform the actual verification.\n\nThis release sets a new standard for defining complex agent behaviors, establishing a playbook for structured system interactions.
Source: Github ↗