~/dasomel
한국어

Core Concepts

Core concepts, three-tier architecture, trust models, and governance principles in OpenForge.

Core Concepts

OpenForge cleanly separates Policy, Implementation, and Evidence, linking them directly to the continuous project lifecycle.

Three-Tier Model

LayerResponsibilityArtifacts & Examples
StandardsDefines expected engineering outcomes and principlesSupply Chain Security Standard, CI/CD Resilience Standard
TemplatesProvides safe, conservative, and ready-to-use starting assetsMulti-stage Dockerfiles, GitHub Workflows, K8s Manifests
Reference ImplementationReal-world OSS adoption, trade-offs, and empirical metricsNarwhal, KubeMetal, nfs-quota-agent, Beluga Manager

8 Foundational Principles

  1. Dual-Language Documentation Policy: English is the canonical project language; Korean is a first-class translation. User-facing Markdown follows the <name>.md and <name>-ko.md pairing rule.
  2. Secure & Reproducible by Default: Projects must be reproducible, documented, testable, observable, accessible, and secure by default.
  3. Transparent Change Management & ADRs: GitHub Issues and Pull Requests serve as the primary change-management mechanism. Critical architectural decisions are recorded as ADRs.
  4. CI Quality Gating: All changes must pass build, test, lint, and security checks in CI before merging.
  5. Supply Chain Governance & Impact Analysis: Dependency compatibility alone does not justify immediate adoption of new releases. Changes require workflow-wide impact analysis.
  6. Trust Boundaries for AI Agents & Local Instructions: AI agents and repository-local instructions (AGENTS.md, CLAUDE.md) are treated as potentially untrusted execution inputs with explicit permission and sandbox boundaries.
  7. Risk-Based Governance & CI Resilience: Even single-maintainer projects maintain automated governance controls without excessive manual overhead. CI outages must never force maintainers to bypass security gates blindly.
  8. Time-Bounded Security Exceptions: Intentional deviations from the baseline must be documented with rationale, scope, and expiration dates.

Trust Boundary Model

OpenForge divides development and runtime environments into distinct trust domains:

  • Source Code & PRs: External contributions and AI-generated code are treated as untrusted inputs until validated.
  • CI Runners: Untrusted fork PR workflows run in isolated environments without access to release secrets.
  • Publishing & Release Credentials: Short-lived OIDC tokens replace static long-lived credentials.
  • Runtime Containers: Workloads execute with non-root users, read-only root filesystems, and strict NetworkPolicies.