Overview
Claude-Red organizes offensive-security knowledge into focused SKILL.md documents that help an authorized reader move from a scoped question to target evidence, causal explanation, reproducible testing, defender-aware reporting, and retesting. The series follows that progression across application, identity, cloud, CI/CD, wireless, device, exploit, fuzzing, and AI boundaries. Its value is structured, domain-specific context—not proof that a target is vulnerable, complete coverage, current technique, permission, or autonomous action; metadata gaps and roadmap limits require provenance checks, human authority, and adaptation.
Learning outcomes
- Claude-Red’s main value is focused, condition-aware context that helps a model and reader frame a security question without loading an undifferentiated catalogue.
- A category or skill can organize coverage and suggest a hypothesis, but only target evidence and causal analysis support a finding.
- Together, reproducibility, defender visibility, reporting, and retesting make observations—from crashes to trust-boundary behavior—more defensible.
- The retained snapshot has uneven discoverability: 28 of 78 skills lack the stated YAML frontmatter, while some manifest descriptions are empty.
Keywords
Preview every chapter · Unlock full text, audio, and video from chapter 4
PlansFAQ
What is Claude-Red useful for?
Claude-Red is useful for supplying focused, condition-aware security context for authorized questions, from scoping and observation through validation, reporting, and retesting. It improves the starting point; it does not observe the target or decide what the evidence proves.
Does Claude-Red prove that a target is vulnerable?
No; Claude-Red cannot prove that a target is vulnerable. Target evidence, causal analysis, and controlled validation are required to establish a finding, so its examples remain possibilities until the tested environment supports them.
How should a reader choose a Claude-Red skill?
Choose the smallest skill or combination that matches the authorized question, system boundary, available identity or device, protocol or state, and permitted evidence. Add context only when observations justify the dependency, and check provenance and metadata.
Can Claude-Red act autonomously on a target?
No; Claude-Red is not an autonomous security operator. Its Markdown and installer do not establish permission, constrain tools, or enforce scope, so human authorization, isolation, least privilege, logging, and approval gates must surround any action.