Skip to main content

oh-my-aidlcops

oh-my-aidlcops (OMA) is a Claude Code · Kiro plugin marketplace that turns the two reliability axes of the AIDLC methodology — Ontology Engineering (correctness) and Harness Engineering (safety) — into installable plugins on AWS. AWS's official AIDLC Workflows serve as the process spine, and AgenticOps closes the feedback loop back into the ontology. OMA is a sister project of oh-my-claudecode (OMC), specializing its orchestration philosophy in making the AIDLC loop reliable enough to run with agents.

One-Paragraph Summary​

Agentic AIDLC fails on reliability, not capability — hallucination/drift, runaway execution, and self-grading. The methodology answers with a reliability dual-axis: Ontology Engineering guarantees the correctness of what agents produce (WHAT/WHEN), and Harness Engineering enforces the safety of how they execute (HOW). OMA is the installable implementation of that dual-axis: a one-install easy button that activates a typed ontology, a harness DSL, and AWS Hosted MCP wiring without hand-rolling schemas, policies, or hooks. Users approve at checkpoints; agents handle diagnosis, proposal, and execution.

Where this is going — an enterprise operations open toolset​

OMA is being built into an open toolset for enterprise operations automation:

  1. Today — Ontology + Harness Engineering as installable plugins, AWS Hosted MCP (awslabs/mcp) as the default runtime data layer, AgenticOps closing the Outer Loop.
  2. Next — deeper AWS Hosted MCP coverage plus first-class DevOps agent and Security agent integrations, so deploy, observability, and security review run as governed agents inside the same Tier-0 approval model.
  3. The promise — install a few plugins and get enterprise-grade operations automation that is auditable, policy-gated, and harness-constrained by default — a drop-in open toolset, not a bespoke platform you assemble yourself.

Plugin Catalog​

PluginRoleExample Skills
ai-infraBuild and operate Agentic AI Platform on EKSagentic-eks-bootstrap, vllm-serving-setup, inference-gateway-routing, langfuse-observability, gpu-resource-management, ai-gateway-guardrails
agenticopsAgent-driven operations automationself-improving-loop, autopilot-deploy, incident-response, continuous-eval, cost-governance, audit-trail
aidlcAIDLC Phase 1 (Inception) + Phase 2 (Construction) extensions (opt-in)requirements-analysis, user-stories, workflow-planning, component-design, code-generation, test-strategy, risk-discovery, quality-gates
modernizationLegacy workload modernization to AWS (6R strategy)workload-assessment, modernization-strategy, to-be-architecture, containerization, cutover-planning

Detailed plugin definitions are in the repository root at .claude-plugin/marketplace.json.

Tier-0 Workflows​

Tier-0 workflows are high-leverage operations that start an entire flow with a single invocation and require user approval only at checkpoints.

CommandPurpose
/oma:autopilotAutonomous AIDLC full-loop execution (Inception → Construction → Operations)
/oma:aidlc-loopSingle feature AIDLC one-pass
/oma:inceptionPhase 1 only
/oma:constructionPhase 2 only
/oma:agenticopsOperations mode (continuous-eval + incident-response + cost-governance running in parallel)
/oma:self-improvingTraces (via your opt-in trace MCP) → skill/prompt improvement PR feedback loop
/oma:platform-bootstrapAgentic AI Platform 5-checkpoint bootstrap on EKS
/oma:reviewAIDLC artifact review (ADR, spec, design, PR)
/oma:cancelTerminate active Tier-0 mode

See Tier-0 Workflows for detailed invocation of each command.

AIDLC × AgenticOps Fusion Diagram​

The diagram above shows how the AIDLC 3-phase structure closes via an agent-driven feedback loop. Observability data from the Operations phase (Langfuse traces, Prometheus metrics, CloudWatch logs) flows back through self-improving-loop to automatically improve Construction-phase skills and prompts. Note that trace-based feedback requires an external Langfuse instance plus a trace MCP server configured in the profile (observability.trace_mcp).

Supported Harnesses (Dual Harness)​

OMA operates identically across two agent harnesses.

  • Claude Code — Install via native /plugin marketplace add or bash scripts/install/claude.sh (or oma setup). Integrates into .claude/plugins/, .claude/commands/oma/, and .claude/settings.json.
  • Kiro — Install via bash scripts/install/kiro.sh (or select harness=kiro in oma setup). Symlinks SKILL.md to .kiro/skills/ and steering to .kiro/steering/.
  • Shared state — The .omao/ directory at project root is harness-agnostic; both harnesses read and write the same files.
  • Recommended path — Single oma setup command handles profile + seed ontology + plugin installation all at once. See Easy Button for details.

Installation and configuration for each harness are covered in Claude Code Setup and Kiro Setup.

Target Users​

  • Platform engineers building and operating Agentic AI platforms on AWS EKS
  • LLM and agent operations teams seeking to cover planning through operations with AIDLC
  • Claude Code or Kiro users who prefer validated drop-in marketplace plugins over building custom skills

Reused Assets​

OMA follows the principle of reuse over reinvention. Full attribution is documented in NOTICE.

SourceLicenseHow Used
awslabs/agent-pluginsApache-2.0Plugin, Skill, MCP, and Marketplace JSON schemas adopted
awslabs/aidlc-workflowsMIT-0Used as AIDLC core; contributions are *.opt-in.md extensions only
awslabs/mcpApache-2.011 hosted MCP servers referenced
aws-samples/sample-apex-skillsMIT-05-checkpoint workflow template
oh-my-claudecode—Tier-0 orchestration philosophy and .omc/ state management inherited

Next Steps​

  1. Getting Started — 5-minute Quickstart to experience your first Tier-0 execution.
  2. Philosophy — Understand the AIDLC × AgenticOps design premise.
  3. Claude Code Setup or Kiro Setup — Proceed with actual installation.

Reference Materials​

Official Documentation​

OMA Internal Documentation​