Trust Is the Attack Surface

Claude-Red’s identity, cloud, and CI/CD skills all begin with a question that is easy to skip: which identity already exists, what can it influence, and through which trust boundary? This chapter focuses on that relationship model. It shows how Active Directory, cloud accounts, and build pipelines can be understood as graphs of authority without collapsing their provider-specific mechanisms into one generic recipe. The reader should leave able to map a principal, a resource, a permission edge, and the evidence needed to show that a path is real.

Identity is the initial condition

In an application review, a request often begins with a user and an object. In an identity system, the starting point is broader: a person, service account, machine, role, token, certificate, or pipeline process may already carry authority. That authority can be direct, inherited, delegated, or created through a trust relationship. If the starting identity is not known, a list of possible escalation paths says very little.

Claude-Red’s cloud skill makes identity the first step across AWS, Azure, and GCP. It asks the practitioner to identify the cloud and the identity context—user, role, service account, or instance identity—before enumerating effective permissions. The Active Directory skill uses a related sequence: understand the directory structure, map paths toward privileged outcomes, and validate each hop. These workflows express the same principle as the earlier API example, but at a larger scale. Authorization is not a property of a resource alone. It is a relationship between a current principal, a requested action, and the resource that action would affect.

This changes how a reader should use a skill. A provider guide is not a menu of impressive capabilities. It is a way to ask which edges exist in the graph in front of them. If a role can delegate authority to a workload, that edge may matter. If a service account can alter a pipeline, the pipeline may become an identity source. If an account can modify an access-control list, the important question is what the modified relationship would permit next. The answer depends on the exact environment and permissions, not on the existence of a named technique in Claude-Red.

An identity-first workflow also controls risk. Enumeration can reveal sensitive organization structure, account names, or permission data. A controlled assessment should establish what may be observed and whether live queries are appropriate. The Active Directory skill’s preference for offline analysis where possible reflects both lower operational noise and better evidence handling. A snapshot preserves a view of relationships at a known time, while live changes can make it difficult to tell which edge existed when a conclusion was reached.

Active Directory as a graph of authority

The Active Directory material makes graph reasoning concrete. Its relevant objects include users, computers, groups, access-control entries, service accounts, certificate templates, domain controllers, trusts, and hybrid identity components. The mechanism is not the number of objects. It is the edge between them: who may read, write, delegate, enroll, impersonate, add a member, or change a policy, and what new edge that action would create.