ENTERPRISE SYSTEMS DESIGN REPORT • DISTRIBUTED INFERENCE

OpenSDLC Platform: Agent-Agnostic, Model-Agnostic AI SDLC Orchestration Engine

A software engineering platform that orchestrates the end-to-end software development lifecycle. Delegates structured phases—from requirements discovery and solution design to test synthesis, security auditing, and self-healing fixes—across specialized LLM provider agents under strict enterprise policy gates.

Orchestration Pipeline

7 Stages

Reqs → Arch → Build → Test → Audit → PR → Report

Self-Healing Loop

Up to 3x

Automated compiler/test failure feedback

Tool Protocol

MCP Native

Model Context Protocol for least-privilege tools

Multi-Provider

4 Adapters

Antigravity/Gemini, Claude, Ollama, OpenAI

Multi-Agent Assembly Line

Autonomous SDLC Workflow Engine

DETERMINISTIC PIPELINE

4. Self-Healing Test Loop

Autonomous Repair Loop

Captures runtime test errors and compiler traces, feeds them back into the agent, and auto-repairs code.

Assigned Agent Persona:Compiler & Test Healer

Generated Output Artifact:test_verification.log

test_verification.log

Markdown / Code Artifact

RUN: npm test -- --watchAll=false

PASS: auth.test.ts (4/4 passed)

PASS: tenant.test.ts (6/6 passed)

PASS: rateLimiter.test.ts (4/4 passed - Repaired in Loop #1)

TOTAL: 14/14 tests green (100% pass rate)

Live Loop Simulator

Autonomous Self-Healing Test Repair

Watch how OpenSDLC captures real test suite failures and repairs regressions before delivery

Phase 1 • Sandboxed Jest Suite

14/14 unit tests passed (100% green)

Phase 2 • Failure Feedback Loop

Loop idle. No unresolved failures detected.

Phase 3 • Patch Verification

Patch Applied:Switched non-atomic read to Redis EVALSHA Lua script✔ Auto-verified in sub-second sandbox

Orchestrator Execution Stream

Multi-agent event stream showing requirements handoff, security ADR creation, and sandboxed test feedback

ORCHESTRATOR LIVE

zala@workstation: ~ (interactive agent session)

$sdlc run --spec ./requirements.md --target ./workspace/crm-service

LOCAL OLLAMA (CPU)Step 1 of 4

✔ Parsing requirements: domain boundaries, endpoints, authentication invariants

✔ Spec generated: .orchestrator/01_requirements_analysis.md

→ Handing off to Stage 2: Architecture & Threat Modeling

Executing deterministic multi-agent SDLC stages with automated test verification.Use Play button to step through execution

The Enterprise Challenge

Unconstrained single-prompt code generators produce hallucinated architectures, lack architectural decision records (ADRs), bypass enterprise security standards, and fail to self-heal when automated build or unit tests fail.

The Architectural Solution

OpenSDLC orchestrates a multi-agent assembly line. Requirements are distilled using local Ollama models; architectural solution designs and threat audits are generated using Google Antigravity & Gemini; code scaffolding and self-healing test repair are delegated to Claude. A central Policy Engine enforces provider quota limits, reasoning levels (low/med/high/extreme), and deterministic human-in-the-loop gates.

Key Architectural Differentiators

Eliminates 'prompt-and-pray' coding by breaking development into formal, verifiable SDLC stages.

Self-healing execution loop captures real compiler, linter, and runtime test errors and auto-repairs code before human delivery.

Generates production-grade Architectural Decision Records (ADRs) and threat models automatically.

Production Technology Stack

Core Platform

Node.js / BunTypeScriptPython 3.12CLI Architecture

Agent Personas

Ollama (Requirements)Claude (Code & Fix)Gemini / AGY (Security & Arch)

Protocol & Tools

Model Context Protocol (MCP)Git WorkspacesStreamlit Dashboard

Governance

Quota PoliciesArchitectural Decision Records (ADR)Human Approval Gates

GitHub Repository

Next Architecture Case StudySerpZeta