AI Security Arsenal

AI Security Arsenal

A collection of skills for automated security testing, penetration testing, and security assessment workflows with coding agents like Claude Code, Codex, OpenCode, and others.

Each skill is structured as a thin router (SKILL.md, 150-250 lines) with lazy-loaded sibling content (workflows/, methodology/, payloads/, references/, examples/, schemas/). This lets agents load only what a sub-task needs instead of parsing a monolithic document every turn — leaving more context for the actual target code, traffic, and tool output.

Last validated: 2026-04 against frontier coding agents (Claude Opus 4.x, GPT-5.x class).

Skills Overview

Skill Architecture

Every skill follows the same layout so agents can navigate them predictably:

Every SKILL.md includes:

- When NOT to use — explicit pointers to sibling skills for overlap cases
- Decision tree — how to pick the right workflow
- Parallelism hints — which operations run concurrently vs sequentially
- Sub-agent delegation — when to spawn parallel sub-agents (e.g., one per domain, one per tool, one per auth context)
- Reasoning budget — where extended thinking pays off vs where to execute without it
- Multimodal hooks — screenshots, diagrams, image-based evidence (where relevant)
- Output schema reference — points at schemas/finding.json

Findings validate against the per-skill JSON Schema, so output can be piped directly into SARIF converters, Jira, DefectDojo, or custom triage tooling.

Headline workflows (new in 2026-04)

Mobile Security

Android Pentest

Comprehensive Android mobile application penetration testing with full ADB shell access to rooted devices.

Capabilities:
- Static & dynamic analysis of Android applications
- Runtime manipulation with Frida (SSL pinning bypass, root detection bypass,…