Concept of the skill

What it is: Security fundamentals are the design principles and threat-modeling habits that decide whether a system can safely handle data, identity, and authority under adversarial conditions.

Mental model: Start with assets, adversaries, trust boundaries, and privileged actions. Every mitigation should be traceable to what crosses a boundary, who controls each side, what can go wrong, and how much damage remains if the boundary fails.

Why it exists: Security added after the system works is expensive and incomplete. Designing security in early makes attacks harder, slower, easier to detect, and smaller in blast radius.

What it is NOT: It is not an OWASP deep audit, LLM prompt-injection architecture, vendor webhook mechanics, scanner configuration, cryptographic primitive implementation, or legal compliance workflow.

Adjacent concepts: owasp-security owns category-specific application-security review; prompt-injection-defense owns LLM instruction-channel threats; type-safety carries validated values after boundary parsing; api-design and http-semantics shape public interfaces; webhook-integration owns vendor webhook mechanics.

One-line analogy: Security fundamentals are like structural engineering for software: load-bearing decisions must be in the design before people move in.

Common misconception: A pile of controls is not the same as a security argument; controls matter only when they are placed at the right trust boundaries and fail safely.

Security Fundamentals

Coverage

The cross-cutting design principles, threat-modeling discipline, and recurring vulnerability classes that determine whether a system can safely handle data, identity, and authority under adversarial conditions. Covers the foundational discipline upstream of any specific vulnerability or tool: Shostack's four threat-modeling questions, Saltzer and Schroeder's eight design principles (1975), trust boundaries, the authentication/authorization distinction, input validation as a boundary discipline, defense in depth, least privilege, and the OWASP Top 10 as a working enumeration of recurring failure classes. Does NOT cover the implementation of specific cryptographic primitives, the configuration of specific scanners, the regulatory artifacts of compliance regimes, the LLM-specific specialization to prompt injection, or the organizational/social side of security.

Philosophy of the skill

Security is a property of the design, not a feature added after the system works. The cost of designing security in is small; the cost of retrofitting it is order-of-magnitude larger and produces worse results. The discipline of security fundamentals is the discipline of paying these costs early — at the threat-modeling stage, at the trust-boundary stage, at the authentication-design stage — before the system has accumulated the structural debt that makes retrofitting expensive.

The discipline does not promise prevention of attacks. Any non-trivial system will be attacked; some attacks will succeed. The goal is to make attacks expensive, slow, traceable, and limited in blast radius. Every design choice is evaluated by what it costs the defender vs what it costs the attacker. Secure-by-default choices cost the defender slightly more code upfront but cost the attacker a working exploit; opt-in security features cost the defender nothing upfront but cost the attacker very little when the developer inevitably forgets. The discipline is the deliberate placement of costs on the attacker rather than the defender.

For agents writing code, the discipline is what lets the agent reason about a security-relevant change without having to read the entire system. An agent that knows the trust boundaries, the authn/authz distinction, and the input-validation discipline can look at a new endpoint and ask: 'where is this endpoint receiving data from?' 'Is the data validated at the boundary?' 'Is authentication checked?' 'Is authorization checked at the moment of the privileged action?' 'What's the blast radius if any of these fail?' These questions produce the right code without the agent having to recall every OWASP entry.

The Four Threat-Modeling Questions (Shostack)

A team that answers all four iteratively, with each system change, is doing security fundamentals. A team that answers only three, or answers them once, is doing security theater.

Saltzer & Schroeder's Eight Principles (1975)

These predate every modern technology and remain canonical because they describe properties, not implementations.

The Authn / Authz Distinction

Conflating these is the #1 issue in the OWASP Top 10. The pattern: authenticate at the entry point; authorize at every privileged action. Never short-circuit authorization on the basis of having a valid session.

Input Validation As Boundary Discipline

The pattern, in order:

1. Define expectations. Every input has an explicit shape, range, and meaning expectation. Document it as a schema (Zod, JSON Schema, OpenAPI).
2. Validate at the boundary. Validation happens at the entry point of the request, not three function calls in. The earlier validation fails, the less the system has done with bad data.
3. Parse, don't validate (Alexis King). Convert the untrusted value into a typed, validated value once; trust the typed form everywhere downstream. This composes with type-safety — the type system carries the validation forward.
4. Treat validation failure as a response-shaped outcome. Return a structured error (400 Bad Request with field-level detail) rather than throwing in the middle of business logic.
5. Log validation failures. Repeated validation failures from a specific source may be probing for vulnerabilities; logged failures feed monitoring.

Defense In Depth — Layered Controls

The property of defense in depth is the composition — no single layer is the security; the security is what remains when one layer fails.

Verification

After applying this skill, verify:

• [ ] A threat model exists for the system or feature, answering all four Shostack questions; it is dated and reviewed at meaningful intervals.
• [ ] Trust boundaries are explicitly enumerated; each boundary has a validation discipline applied.
• [ ] Authentication is required at every entry point that needs identity; authorization is checked at every privileged action — not just at session start.
• [ ] Input validation happens at the trust boundary, returns structured errors, and is logged.
• [ ] Sensitive data has been classified (public, internal, confidential, restricted, secrets); each tier has handling rules; secrets are never logged or stored unencrypted.
• [ ] Service-to-service calls use mTLS or signed tokens; no service trusts another solely on network position.
• [ ] Defaults are fail-safe (deny by default; permit by explicit exception with justification).
• [ ] Every entity (user, service, process) has minimum-necessary permissions; service accounts do not have admin keys; tokens are scoped to the minimum action set.
• [ ] Sensitive actions are logged with sufficient detail to reconstruct what happened; logs are protected from modification.
• [ ] Monitoring is in place to detect anomalies and failed validations; alerts are tested.
• [ ] Cryptographic primitives are implemented by well-reviewed libraries, not custom code; keys are managed (rotation, escrow, scoped access).
• [ ] Compliance documentation (where applicable) is downstream of the security property, not a substitute for it.

Do NOT Use When

Key Sources

• Saltzer, J. H., & Schroeder, M. D. (1975). "The Protection of Information in Computer Systems". Proceedings of the IEEE, 63(9), 1278–1308. The foundational paper on security design principles; the eight principles articulated here remain canonical across every subsequent technology shift.
• Shostack, A. (2014). Threat Modeling: Designing for Security. Wiley. The canonical modern reference on threat modeling, including the four-question framework and STRIDE (Spoofing, Tampering, Repudiation, Information Disclosure, Denial of Service, Elevation of Privilege).
• OWASP. OWASP Top 10 (2021). The current stable awareness document for recurring web application vulnerability classes. Use owasp-security for OWASP-category deep review and newer Top 10 release-candidate mapping.
• OWASP. OWASP Top 10 for Large Language Model Applications. The LLM-specific specialization, including prompt injection and excessive agency; use prompt-injection-defense for that territory.
• OWASP. Input Validation Cheat Sheet. Current practical guidance for validating untrusted input early and server-side.
• Anderson, R. (2020). Security Engineering: A Guide to Building Dependable Distributed Systems (3rd ed.). Wiley. The comprehensive treatment of security engineering across cryptography, access control, authentication, and large-system architecture; the discipline's modern textbook.
• Kerckhoffs, A. (1883). "La cryptographie militaire." Journal des sciences militaires, IX, 5–83. The foundational articulation of "security must not depend on the secrecy of the design" — the principle Saltzer & Schroeder generalized as "open design."
• NIST. Special Publication 800-63B-4: Digital Identity Guidelines — Authentication and Lifecycle Management. The current reference standard for authentication and lifecycle guidance.
• CISA et al. Shifting the Balance of Cybersecurity Risk: Principles and Approaches for Secure by Design Software. Secure-by-design and secure-by-default principles for software manufacturers.
• Lampson, B. W. (1973). "A Note on the Confinement Problem." Communications of the ACM, 16(10), 613–615. The foundational paper on confinement — the question of whether a program can be prevented from leaking information it has access to.
• King, A. (2019). "Parse, don't validate". Modern articulation of the validation discipline: convert untrusted data to typed values once at the boundary, then trust the type.
• Bell, D. E., & LaPadula, L. J. (1973). "Secure Computer Systems: Mathematical Foundations." MITRE technical report. The Bell-LaPadula model — foundational work on formal access-control models that underpin modern RBAC/ABAC systems.