Skip to content

Nexus SDLC

Nexus SDLC is a human-in-the-middle orchestration framework that coordinates a specialized swarm of autonomous agents to automate the end-to-end software development lifecycle.


Abstract

Nexus SDLC transforms the traditional, manual Software Development Lifecycle into a dynamic, multi-agent collaboration. By placing a Human-in-the-Middle (HITM) at the architectural "Nexus," the system ensures that AI-driven velocity never diverges from human-defined quality, security, and architectural standards.

The framework operates as a synchronized collective of specialized agents that reason, execute, and self-correct. The human remains the central node of the process — providing high-level intent, resolving ambiguity, and validating critical pivots before code enters production.


Core Architecture: Managed Autonomy

The framework is built on the principle of Managed Autonomy. Rather than a static set of tools, Nexus utilizes a modular agentic swarm where roles are defined by the specific needs of the project phase.

  • The Nexus (Human): The strategic control center. Provides high-level goals, resolves logic paradoxes, and performs final validation.
  • The Swarm (Agents): Specialized autonomous units capable of task decomposition, implementation, and rigorous validation.
  • Dynamic Orchestration: A coordination layer that manages state, handles agent hand-offs, and maintains a unified context across the entire lifecycle.
flowchart TD
    classDef nexus    fill:#c9b8e8,stroke:#6b3fa0,color:#1a0a2e,font-weight:bold
    classDef orch     fill:#e8d4b8,stroke:#9e6b2d,color:#2e1a0a,font-weight:bold
    classDef swarm    fill:#b8d4e8,stroke:#2d6b9e,color:#0a1a2e,font-weight:bold
    classDef artifact fill:#b8e8c9,stroke:#2d9e5a,color:#0a1e0a,font-weight:bold

    N["👤 The Nexus<br/>─<br/>Sets goals · Approves plans<br/>Resolves ambiguity · Signs off"]:::nexus

    subgraph OL["Dynamic Orchestration Layer"]
        direction LR
        ME["Methodologist<br/>─<br/>Configures the swarm"]:::orch
        OR["Orchestrator<br/>─<br/>Routes work · Manages state"]:::orch
    end

    subgraph SW["The Swarm"]
        direction LR
        AN["Analyst"]:::swarm
        AU["Auditor"]:::swarm
        AR["Architect"]:::swarm
        DE["Designer ◦"]:::swarm
        SC["Scaffolder ◦"]:::swarm
        PL["Planner"]:::swarm
        BU["Builder"]:::swarm
        VE["Verifier"]:::swarm
        SE["Sentinel"]:::swarm
        DO["DevOps"]:::swarm
        SR["Scribe"]:::swarm
    end

    N      -->|"High-level goal"| ME
    ME     -->|"Methodology Manifest"| OR
    OR    <-->|"Task routing · Escalation"| SW
    OR     -->|"Gate briefings · Demo briefing"| N
    N      -->|"Approval · Feedback · Changes"| OR

◦ optional — invoked when the project profile and delivery channel require it


Getting Started

  1. InstallINSTALL.md — clone the repo, run the install script, start Claude Code (5 minutes)
  2. UseUSAGE.md — invoke the Methodologist with your project description and let the swarm take it from there
  3. UnderstandRATIONALE.md — the design thinking behind the framework (optional, read when curious)

How It Works

  1. Define — The Nexus states the high-level goal and constraints.
  2. Ingestion — Analyst elicits requirements; Auditor validates them in a loop with the Nexus resolving ambiguities. Output: the Brief and Requirements List. The Nexus approves at the Requirements Gate.
  3. Decomposition — Architect produces the architectural approach; Auditor audits it; Planner decomposes into atomic tasks; Designer and Scaffolder are invoked when the project requires them. The Nexus approves at the Architecture Gate and Plan Gate.
  4. Execution Cycle — Builder implements one task at a time under strict TDD.
  5. Verification Cycle — Verifier tests against acceptance criteria and Sentinel runs security audits concurrently. Failures feed back into an autonomous iterate loop.
  6. Demo Sign-off — The Nexus explores the running software. On approval, the cycle is signed off and the swarm is ready for the next cycle or release.
  7. Go-Live — DevOps deploys to production; Scribe publishes documentation and release notes.
flowchart TD
    classDef nexus fill:#c9b8e8,stroke:#6b3fa0,color:#1a0a2e,font-weight:bold
    classDef phase fill:#b8d4e8,stroke:#2d6b9e,color:#0a1a2e,font-weight:bold
    classDef gate  fill:#e8d4b8,stroke:#9e6b2d,color:#2e1a0a,font-weight:bold
    classDef loop  fill:#b8e8c9,stroke:#2d9e5a,color:#0a1e0a,font-weight:bold

    DEFINE["👤 DEFINE<br/>─<br/>Nexus states the goal"]:::nexus
    INGEST["INGESTION<br/>─<br/>Analyst + Auditor loop<br/>Nexus resolves issues"]:::phase
    RG["⬡ REQUIREMENTS GATE<br/>─<br/>Nexus approves requirements"]:::gate
    DECOMP["DECOMPOSITION<br/>─<br/>Architect · Auditor · Planner<br/>optional Designer · Scaffolder"]:::phase
    AG["⬡ ARCHITECTURE GATE<br/>─<br/>Nexus approves architectural approach"]:::gate
    PG["⬡ PLAN GATE<br/>─<br/>Nexus approves task plan"]:::gate
    EXEC["EXECUTION CYCLE<br/>─<br/>Builder implements atomic tasks"]:::phase
    VERIFY["VERIFICATION CYCLE<br/>─<br/>Verifier · Sentinel run concurrently"]:::phase
    ITERATE["↺ ITERATE<br/>─<br/>Autonomous correction loop<br/>within cycle"]:::loop
    DS["⬡ DEMO SIGN-OFF<br/>─<br/>Nexus explores running software<br/>Approves cycle · Triggers retrospective"]:::gate
    GL["⬡ GO-LIVE<br/>─<br/>Release decision<br/>Automatic · On sign-off · Business decision"]:::gate

    DEFINE  --> INGEST
    INGEST  --> RG
    RG      -->|"Amendments"| INGEST
    RG      -->|"Approved"| DECOMP
    DECOMP  --> AG
    AG      -->|"Amendments"| DECOMP
    AG      -->|"Approved"| PG
    PG      -->|"Amendments"| DECOMP
    PG      -->|"Approved"| EXEC
    EXEC    --> VERIFY
    VERIFY  -->|"Failures"| ITERATE
    ITERATE --> VERIFY
    VERIFY  -->|"All pass"| DS
    DS      -->|"New requirements"| INGEST
    DS      -->|"Next cycle"| EXEC
    DS      -.->|"CD philosophy"| GL

Key Objectives

  • Reduced Cognitive Load: Focus on intent and validation rather than syntax and repetitive boilerplate.
  • Autonomous Iteration: Agents self-correct based on technical feedback loops without constant human prompting.
  • Traceable Reasoning: Every decision made by the agentic collective is logged, providing a transparent audit trail of the development process.
  • Safety by Design: Critical checkpoints ensure that AI agents cannot execute high-risk operations without Nexus approval.

Documentation

Document Description
RATIONALE.md Design rationale — the why, the reasoning, and the process model behind an agentic SDLC
REFERENCES.md Full bibliographic reference library: SDLC methodologies, agentic AI research, and foundational theory
process/INDEX.md Architecture decisions (DEC) and open questions (OQ) — the living design record
guidelines/diagram-guidelines.md Mermaid diagram standards for all agents that produce visual output

Test Projects

End-to-end runs of the Nexus SDLC framework on real projects.

Project Profile Report Transcript Retrospective
NexusScan Casual — Python CLI tests/NexusScan.md tests/NexusScan-TRANSCRIPT.md tests/NexusScan-RETROSPECTIVE.md

Agent Definitions

Loadable agent files in /agents/:

Agent Plane Role
methodologist.md Configuration Configures the swarm; runs retrospectives; versions the Methodology Manifest
orchestrator.md Control Routes work; manages lifecycle state; prepares Nexus gate briefings
analyst.md Analysis & Planning Produces the Brief and Requirements List
auditor.md Analysis & Planning Validates requirements and architectural decisions; enforces traceability
architect.md Analysis & Planning Trade-off analysis, architectural characteristics, ADRs, fitness functions
designer.md Design & Structure ◦ UX/IxD for projects with a UI delivery channel
scaffolder.md Design & Structure ◦ Translates component decisions into code structure before Builder work begins
planner.md Analysis & Planning Decomposes requirements into atomic tasks with acceptance criteria
builder.md Execution Implements one task at a time under strict TDD
verifier.md Verification & Security Tests against acceptance criteria; produces Verification Reports and Demo Scripts
sentinel.md Verification & Security Dependency security audit and live OWASP testing against staging
devops.md Delivery CI/CD pipeline, environment provisioning, deployment
scribe.md Delivery Documentation transformation at release time; produces release notes and changelog

◦ optional — invoked when the project profile and delivery channel require it


License

This project is licensed under the Apache License 2.0. See the LICENSE file for details.