Tier-0 Workflows

This document is a deep reference for the 9 Tier-0 commands that OMA provides. Tier-0 means "invoke once, run autonomously with approval only at checkpoints"—high-leverage workflows where OMA's core value concentrates.

Tier-0 Design Principles

All Tier-0 commands share these principles:

1. Single invocation — Users invoke the command once. Agents orchestrate multiple skills and tools.
2. Checkpoint-based approval — Follows the 5-stage pattern from aws-samples/sample-apex-skills: Gather Context → Pre-flight → Plan → Execute → Validate.
3. State persistence — All checkpoint results are saved in .omao/state/session-<id>/. Pause and resume are supported.
4. Natural language arguments — Command arguments are natural language strings. Agents parse intent.

Command Catalog

Command	Scope	Time	Checkpoints
/oma:autopilot	Full AIDLC loop	30min–hours	4–6
/oma:aidlc-loop	Single feature one-pass	10–30min	2–3
/oma:inception	Phase 1 only	5–15min	1–2
/oma:construction	Phase 2 only	10–30min	2–3
/oma:agenticops	Activate operations mode	immediate	1 (activation approval)
/oma:self-improving	Traces → improvement PR	5–20min	2
/oma:platform-bootstrap	EKS platform 5-stage build	30–60min	5
/oma:review	Artifact review	2–10min	0 (report only)
/oma:cancel	Tier-0 termination	immediate	0

Detailed descriptions of each command follow.

/oma:autopilot — Full AIDLC Loop Autonomous Execution

Purpose

Performs Inception → Construction → Operations end-to-end for a single feature or whole project. Human involvement limited to checkpoint approvals.

Invocation Example

> /oma:autopilot "Build a new anomaly detection feature in the payment service, from planning through operations"

Checkpoint Flow

Use Cases

• Initial feature build
• Full AIDLC consistency validation at project kickoff
• Rebuilding a feature under OMA standards during team transitions

Dependencies

• aidlc-inception + aidlc-construction + agenticops plugins active
• eks-mcp-server, cloudwatch-mcp-server, aws-iac-mcp-server MCP connections

/oma:aidlc-loop — Single Feature One-Pass

Purpose

Inception + Construction one-pass without the Operations phase. Use when operations automation is already configured and you need to quickly integrate a new feature.

Invocation Example

> /oma:aidlc-loop "Add MFA verification field to the user profile API"

Checkpoint Flow

CK1 (requirements) → CK2 (design & implementation). Two stages.

Use Cases

• Daily feature additions and fixes
• Auto-invocation within CI pipelines (approval via separate workflow)
• Skipping the operations portion of autopilot

/oma:inception — Phase 1 Only

Purpose

Generate requirements analysis, user stories, and workflow planning artifacts only. Use when design and implementation are handled manually.

Invocation Example

> /oma:inception "Gather initial requirements for the next-generation order management system"

Artifacts

.omao/plans/
├── spec.md
├── user-stories.md
└── workflow-plan.md

Use Cases

• Pre-workshop and design sprint preparation
• Product manager auto-generating requirements drafts
• Delegating Construction to external dev teams

/oma:construction — Phase 2 Only

Purpose

Generate design and implementation artifacts given an existing .omao/plans/spec.md. Use when Inception is already complete.

Invocation Example

> /oma:construction "Based on current spec.md, perform component design and TDD implementation"

Artifacts

.omao/plans/
├── design.md
├── adr-*.md
├── test-strategy.md
└── (code changes committed to feature branch)

Use Cases

• Copy-pasting spec from other tools into .omao/plans/spec.md then executing
• Re-running Construction after manually revising Inception
• Retrofitting OMA standard design and tests to legacy codebases

/oma:agenticops — Activate Operations Mode

Purpose

Activate continuous-eval, incident-response, and cost-governance skills in the background for continuous operations automation. This is a state transition command, not a one-time execution.

Invocation Example

> /oma:agenticops "Activate operations mode for the production cluster"

Post-Activation Behavior

• continuous-eval — Periodically evaluate Ragas metrics and regression samples; send rollback signal on regression detection
• incident-response — Auto-respond to PagerDuty and CloudWatch alarms, generate diagnosis and mitigation proposals
• cost-governance — Detect AWS Cost Explorer anomalies, recommend scaling on budget excess

Deactivation

> /oma:cancel

Use Cases

• Immediately after production cluster deployment
• End-of-sprint operations automation checkpoint refresh
• Reducing on-call burden during holidays and weekends

/oma:self-improving — Traces → Improvement PR

Purpose

Analyze Langfuse traces and failure logs to auto-generate PRs for skill and prompt improvements. Core of the feedback loop.

Invocation Example

> /oma:self-improving "Analyze failures from the last 7 days and propose improvement PRs"

Checkpoint Flow

CK1: Approve improvement candidates → CK2: Confirm regression tests pass, create PR.

Artifacts

• GitHub PR (auto-labeled agenticops/auto-improvement)
• .omao/plans/improvement-<date>.md summary report

Use Cases

• Auto-run before weekly operations meetings
• Manual invocation when specific skill failure rate exceeds threshold
• Prompt optimization after new model version release

/oma:platform-bootstrap — EKS Platform 5-Stage Build

Purpose

Build an Agentic AI Platform on EKS with 5-stage checkpoints. Covers full stack: vLLM inference, Inference Gateway, Langfuse observability, Kagent orchestration, and GPU resource management.

Invocation Example

> /oma:platform-bootstrap "Build Agentic AI Platform at 8-node GPU scale"

5-Stage Checkpoints

1. Cluster Prep — Validate EKS version, VPC, and Karpenter configuration
2. GPU & Model Serving — Deploy NVIDIA GPU Operator and vLLM
3. Inference Gateway — Deploy kgateway with routing rules
4. Observability — Connect Langfuse + Prometheus + OpenTelemetry
5. Agent Layer — Deploy Kagent + Ragas evaluation pipeline

Dependencies

• agentic-platform plugin active
• eks-mcp-server, prometheus-mcp-server, aws-iac-mcp-server connections
• Sufficient EKS permissions (minimum eks:*, ec2:*, iam:CreateRole)

Use Cases

• First-time platform build on new cluster
• Reconfiguring existing cluster to OMA standards
• One-day PoC and demo environment setup

/oma:review — Artifact Review

Purpose

Analyze AIDLC artifacts (ADR, spec, design, PR) and generate quality reports. No execution changes; returns review results only.

Invocation Example

> /oma:review "Review .omao/plans/adr-auth-refactor.md"
> /oma:review "Review current PR #123"

Review Items

• AIDLC structure compliance
• Missing ADR, test, or design documentation
• Alignment with engineering-playbook standards
• Security, cost, and compliance considerations

Use Cases

• Auto self-review before PR merge
• Quarterly quality audit
• Feedback on new team member initial artifacts

/oma:cancel — Tier-0 Termination

Purpose

Immediately terminate active Tier-0 mode. Use to stop long-running commands like autopilot and agenticops.

Invocation Example

> /oma:cancel

Behavior

• Remove current mode from .omao/state/active-mode.json
• Send termination signal to background-running skills
• Preserve partial artifacts in .omao/state/session-<id>/ (recovery possible)

Common Options

All Tier-0 commands support these common options:

Option	Effect
--dry-run	Generate plan only, do not execute
--verbose	Detailed output of intermediate artifacts per stage
--resume <session-id>	Resume a paused session

Examples:

> /oma:autopilot --dry-run "Add anomaly detection to payment service"
> /oma:autopilot --resume session-2026-04-21-a1b2 "Continue"

Advanced Checkpoint Structure

Checkpoints are stored as .omao/state/session-<id>/checkpoint-<n>.json.

{
  "checkpoint": 2,
  "phase": "construction",
  "timestamp": "2026-04-21T14:32:10Z",
  "inputs": {
    "spec_path": ".omao/plans/spec.md"
  },
  "artifacts": [
    ".omao/plans/design.md",
    ".omao/plans/adr-auth.md"
  ],
  "approval": {
    "status": "approved",
    "approver": "user",
    "comment": null
  }
}

These files are restore points for pause/resume. You can also manually edit checkpoint results and combine with --resume for flexible workflow reconstruction.

Command Selection Guide

Decision tree for choosing which command to use:

Reference Materials

Official Documentation

• aws-samples/sample-apex-skills — 5-checkpoint template source
• Langfuse Documentation — self-improving loop data source
• awslabs/mcp — MCP servers that Tier-0 depends on

OMA Internal Documentation

• Introduction — OMA overview and plugin catalog
• Philosophy — Tier-0 design background
• Keyword Triggers — Natural language input → Tier-0 auto-mapping
• Claude Code Setup — Pre-installation for Tier-0 execution