SKILL.md
DevSecOps, Container, and API Operational Security Reference
Defensive practitioner reference for the operational layer of a system — the pipeline that builds it, the container and orchestrator that run it, and the API surface it exposes at runtime. Every section is framed as defense, hardening, detection, or verification: it describes the misconfiguration, how to detect it, and how to prevent it, never how to exploit it.
This skill is split into three modules by sub-domain. The overview below gives the shared threat model and an entry point; the depth lives in the modules. The threat model is operational: a pipeline can be subverted to inject a build step, a container can be misconfigured into a host breakout, and a running API can leak one tenant's data to another. The defenses establish least privilege, isolation, detection, and runtime authorization at each layer.
Three Sub-Domains (modules)
| Sub-domain | Module | Covers |
|---|---|---|
| DevSecOps | [modules/devsecops.md](modules/devsecops.md) | CI/CD pipeline hardening, secret scanning, IaC misconfiguration detection (Terraform / CloudFormation), SAST/DAST integration |
| Container | [modules/container.md](modules/container.md) | Image scanning, Kubernetes RBAC hardening, container-escape defense (seccomp / AppArmor / read-only root / non-root), runtime threat detection |
| API operational | [modules/api-ops.md](modules/api-ops.md) | OWASP API Top 10 operational defense (BOLA / broken-auth detection in production, rate-limit enforcement, server-side authorization), WAF rule tuning, GraphQL/REST depth and complexity limiting |
Operational Trust Boundaries
The core defensive insight: each operational layer is a boundary where an attacker who has reached it can move to the next. Defense-in-depth means each boundary assumes the one before it may have failed.
| Boundary | Operational risk | Primary defense | Module |
|---|---|---|---|
| Source → pipeline | Injected build step, leaked secret, poisoned runner | Signed pipeline, secret scanning, runner isolation | DevSecOps |
| Pipeline → infra config | Misconfigured cloud resource (open bucket, permissive IAM) | IaC scanning before apply | DevSecOps |
| Image → registry | Vulnerable base image, embedded secret | Image scanning + admission control | Container |
| Container → host | Container escape to the node | seccomp, AppArmor, read-only root, non-root, drop capabilities | Container |
| Cluster identity → resources | Over-privileged ServiceAccount, broad RoleBinding | Least-privilege RBAC, PodSecurity admission | Container |
| Client → API | Broken object/function authorization, resource exhaustion | Server-side authorization, rate limiting, WAF | API operational |
Ecosystem-Neutral Tooling
This skill names tools at the category level. Where a tool is the de-facto cross-ecosystem standard for its category, it is named but framed as "the standard \<category\> tool" — the concept transfers to any equivalent tool. No single language ecosystem is privileged; pipeline examples are CLI-shape, not language-specific source.
Distinction from moai-ref-owasp-checklist (operational, not dev-time)
The API-operational module covers the runtime/operational half of API security — detecting Broken Object Level Authorization in production traffic, enforcing rate limits at the gateway, tuning a WAF, limiting query depth on a live endpoint. The dev-time half — secure-coding patterns a developer applies while writing the endpoint (parameterized queries, input validation, authentication design, security headers) — lives in moai-ref-owasp-checklist. This skill keeps its coverage operational and does not duplicate them.
Cross-References
moai-ref-owasp-checklist— dev-time web-app OWASP Top 10, authentication
patterns, input validation, HTTP security headers (the development-time surface; this skill is the operational surface).
moai-ref-supply-chain— SBOM, SLSA provenance, Sigstore signing, dependency
hygiene (the supply-chain surface that image scanning and pipeline signing here draw on for artifact provenance).
moai-ref-llm-security— AI/LLM defensive security (the LLM-specific operational
surface, distinct from the general API/container/pipeline surface here).
moai-ref-api-patterns— REST/GraphQL API design and error handling (the
API-design surface, distinct from the API-operational-defense surface here).
Defensive Severity Levels
| Level | Label | Action | Example |
|---|---|---|---|
| P0 | CRITICAL | Block release | Container runs privileged with host mounts; API endpoint has no server-side object-ownership check |
| P1 | HIGH | Fix before merge | Pipeline secret exposed in build logs; ServiceAccount bound to cluster-admin |
| P2 | MEDIUM | Fix within iteration | No image scan in the pipeline; no rate limit on an expensive endpoint |
| P3 | LOW | Track in backlog | WAF in detection-only mode where enforcement is warranted; no GraphQL depth limit on a shallow schema |
<!-- moai:evolvable-start id="rationalizations" -->
Common Rationalizations
| Rationalization | Reality |
|---|---|
| "The cluster is internal, so RBAC hardening is optional" | An internal cluster is reachable from any compromised pod. Least-privilege RBAC is the boundary that stops a single compromised workload from owning the cluster. |
| "The container runs our own trusted image, so escape defense is overkill" | A trusted image with a vulnerable dependency is still an escape risk. seccomp, non-root, and read-only root are cheap and assume the image may be compromised. |
| "Authorization is handled in the frontend, the API can trust the client" | The client is never a trust boundary. Every operational API defense — BOLA in particular — requires server-side object-ownership checks on every request. |
| "We scan images at build time, runtime detection is redundant" | Build-time scanning misses a zero-day disclosed after deploy and misses runtime behavior. Runtime detection catches what static scanning cannot. |
| "Rate limiting hurts legitimate users, we will add it if abused" | Unbounded consumption is an OWASP API Top 10 risk that is exploited before it is noticed. Rate limits and query-complexity caps are the control, not an afterthought. |
Assume-breach-by-layer: each operational layer assumes the one before it may have failed — the pipeline assumes a poisoned source, the container a vulnerable image, the API a hostile client. Defense is the control at each layer, not trust in the previous.
<!-- moai:evolvable-end -->
<!-- moai:evolvable-start id="red-flags" -->
Red Flags
- A container runs as privileged, as root, with a writable root filesystem, or with host paths mounted
- A Kubernetes ServiceAccount is bound to cluster-admin or a wildcard Role
- An API endpoint returns an object by ID with no server-side ownership check (Broken Object Level Authorization)
- The CI/CD pipeline echoes secrets into build logs or runs untrusted code on a shared, non-isolated runner
- Infrastructure-as-code is applied to production without a misconfiguration scan
- A public-facing API has no rate limit, no query-depth limit, and no request-size cap
- Runtime detection is absent, so a post-deploy compromise produces no alert
<!-- moai:evolvable-end -->
<!-- moai:evolvable-start id="verification" -->
Verification
- CI/CD pipeline runs secret scanning and IaC misconfiguration scanning before deploy; runners are isolated (see [modules/devsecops.md](modules/devsecops.md))
- Container images are scanned and gated by admission control; containers run non-root with a read-only root filesystem, dropped capabilities, and a seccomp profile (see [modules/container.md](modules/container.md))
- Kubernetes RBAC follows least privilege; no ServiceAccount is bound to cluster-admin; PodSecurity admission is enforced (see [modules/container.md](modules/container.md))
- Runtime-detection rules alert on container-escape and credential-access behavior; alerts route to a responder (see [modules/container.md](modules/container.md))
- Every API endpoint enforces server-side object-level and function-level authorization; BOLA is detectable in production traffic (see [modules/api-ops.md](modules/api-ops.md))
- Rate limits, request-size caps, and GraphQL/REST depth and complexity limits are enforced at the gateway (see [modules/api-ops.md](modules/api-ops.md))
- Operational coverage does not duplicate the dev-time patterns in
moai-ref-owasp-checklist; the two are consulted together
<!-- moai:evolvable-end -->