Eleventh phase of Arc 3. Secrets as engineering, not afterthought.

Most teams handle secrets creation and distribution and ignore the rest. Production-grade platforms treat the whole lifecycle: provisioning, distribution, rotation, revocation, audit. By phase end basecamp has secrets stored in Vault (or equivalent), distributed via External Secrets Operator, rotated on schedule, audited continuously. The whole story is documented and runbook-supported.

This phase complements Phase 25 (mesh + mTLS) which handled service-to-service identity. Together they implement zero-trust-networking as a real architecture.

Prerequisites

• Phase 26 complete; ascent operational

Why this phase exists

A secret is one of: API key, database password, signing key, certificate, encryption key. Each has a lifecycle. Static secrets in .env files are how production gets compromised. The discipline is treating secrets as managed assets with rotation policies.

The pattern-first frame

Same eight steps.

1. PROBLEM

Your services need secrets. They must be: stored encrypted, distributed only to authorized consumers, rotated regularly, revoked on compromise, audited continuously. Manual handling is error-prone; production-grade systems automate every step.

2. PRINCIPLES

2.1 The secret lifecycle

Provision → Distribute → Rotate → Revoke → Audit. Each stage needs deliberate engineering.

→ Pattern: secrets-lifecycle

Investigate:

• For an API key shared with a partner: walk the full lifecycle. Where does each handoff happen?
• What does “rotation” mean for a database password used by 10 services?
• What does revocation mean operationally when the secret is already distributed?

2.2 Defense in depth at every layer

No single layer is sufficient. Network policies, mTLS, secrets, IAM, audit logs: each catches what the others miss.

→ Pattern: defense-in-depth, DEEP target this phase

Investigate:

• For a typical request to basecamp, list every security layer it traverses.
• What’s “swiss cheese model” of security?
• When does adding a layer cost more than it buys?

2.3 Zero-trust security

Trust nothing implicitly. Every interaction authenticated, authorized, encrypted, audited. Network position confers nothing.

→ Pattern: zero-trust-security

Investigate:

• How does zero-trust differ from “the perimeter is the firewall”?
• What’s a workload identity, and how does it compose with secrets?
• When is full zero-trust overkill?

2.4 Vault-as-a-service

A central secrets backend (Vault, AWS Secrets Manager, Doppler) provides storage + rotation + audit. Services authenticate to it via workload identity and pull secrets at runtime.

Investigate:

• What does HashiCorp Vault give you that AWS Secrets Manager doesn’t (and vice versa)?
• What’s dynamic secrets (Vault generates DB credentials per-request)?
• When is the operational cost of running Vault worth it?

2.5 External Secrets Operator (ESO)

A K8s-native pattern: an Operator pulls secrets from Vault (or Secrets Manager, Doppler, etc.) and creates K8s Secret objects. The cluster surfaces secrets to pods normally; the source-of-truth lives in Vault.

Investigate:

• Why is ESO the practical pattern vs Vault Agent sidecar?
• What’s the failure mode when ESO can’t reach Vault?
• How do you rotate without service restart?

2.6 Policy-as-code with Kyverno

Kyverno is a K8s-native policy engine: every policy is a CRD (ClusterPolicy, Policy), reconciled by the Kyverno controller. Policies validate, mutate, or generate K8s resources. The natural complement to secrets management: you can enforce “every Pod that mounts a Secret must have specific labels,” “no Secret can be created in this namespace,” “every Workload CRD must reference a Vault-managed secret,” etc.

→ Pattern: policy-as-code, first OUTLINE this phase

Investigate:

• Walk a Kyverno ClusterPolicy: declare → admission controller validates incoming resources → reject/mutate/generate.
• Why is Kyverno more idiomatic for the K8s-native ecosystem than OPA Gatekeeper? (Hint: Kyverno’s policies are YAML CRDs; OPA’s are Rego — separate language.)
• When does a Kyverno policy belong in basecamp’s platform layer vs in the application layer?

3. TRADE-OFFS

4. TOOLS (as of 2026-06)

• HashiCorp Vault (Open Source)
• External Secrets Operator (ESO)
• Sealed Secrets (Bitnami)
• SOPS (Mozilla)
• vault CLI
• gitleaks / trufflehog: secret scanning

Reading

• “Securing DevOps” (Vehent)
• Vault docs: Auth Methods and Dynamic Secrets sections
• External Secrets Operator docs

5. MASTERY: Secrets pipeline on basecamp

[ ] Vault deployed on basecamp (or use AWS Secrets Manager)
[ ] External Secrets Operator installed; pulls from Vault
[ ] At least 5 service secrets managed via ESO
[ ] Rotation schedule: 90 days for static secrets; 1 hour for dynamic
[ ] Sealed Secrets for at least one in-Git config (e.g., basecamp's bootstrap)
[ ] SOPS for one alternative use case (multi-environment secrets in Git)
[ ] Vault audit logs collected + shipped to Loki / equivalent
[ ] Workload identity (IRSA on EKS, Workload Identity on GKE) integrated with Vault
[ ] Secret scanning in CI: gitleaks runs on every PR
[ ] Practice rotation: rotate Postgres password without downtime

6. COMPARE: AWS Secrets Manager

Replicate one secret pipeline using AWS Secrets Manager + ESO instead of Vault. Compare.

400-word reflection.

7. OPERATE

• 4-5 runbooks: secret rotation broke deployment; Vault unsealed; compromised secret; ESO sync failing; auditing access
• 2-3 ADRs (Vault over SM; ESO over Vault sidecar; rotation cadence)
• Weekly log

8. CONTRIBUTE

• ESO providers (community-maintained)
• Vault docs
• SOPS, Sealed Secrets

What ships from this phase

• Secrets infrastructure on basecamp: Vault + ESO + Sealed Secrets + SOPS
• Rotation automation for at least one production secret
• Secrets runbooks

Validation criteria

[ ] Vault + ESO operational; 5+ service secrets managed
[ ] Rotation practiced without downtime
[ ] Sealed Secrets + SOPS used for Git-friendly cases
[ ] CI secret scanning catches a deliberate test secret
[ ] All 10 operational depth checks
[ ] Compare reflection (400 words)
[ ] 4-5 secrets runbooks
[ ] 2-3 ADRs
[ ] Pattern entries:
    - secrets-lifecycle → OUTLINE
    - defense-in-depth → DEEP
    - zero-trust-security → OUTLINE
[ ] Exit Test passed

Exit Test

Time: 2.5 hours.

Part 1: Build (75 min)

Migrate one service’s secrets from environment variables to ESO + Vault. Verify the service still works. Rotate the secret without restart.

Part 2: Diagnose (45 min)

ESO is failing to sync one secret. Possible: Vault token expired, ESO RBAC missing, secret path wrong, network policy blocking.

Part 3: Articulate (30 min)

~600 words: “Walk what happens when basecamp rotates the Postgres root password. Cover Vault rotation, ESO update, K8s Secret object change, pod restart vs hot-reload, audit log entries.”

Anti-patterns

Patterns touched this phase

• secrets-lifecycle: OUTLINE
• defense-in-depth: DEEP
• zero-trust-security: OUTLINE

→ Next: Phase 28: Observability at Platform Depth