The 7 Workflow Phases in the Autonomous Pipeline Execution of Agency-Agents

The Agency-Agents repository implements a 7-phase NEXUS pipeline where each workflow phase produces concrete deliverables, passes through quality gates managed by the Agents Orchestrator, and passes deterministic hand-off packages to prevent context loss between autonomous stages.

The msitarzewski/agency-agents open-source framework orchestrates multi-agent software development through a structured autonomous pipeline execution workflow. Understanding these workflow phases is essential for teams implementing AI-driven development pipelines, as each phase defines specific agent roles, evidence requirements, and transition criteria documented in the repository's strategy playbooks.

Overview of the NEXUS Autonomous Pipeline Structure

The autonomous pipeline execution centers on the Agents Orchestrator, which manages the lifecycle of specialized AI agents across seven distinct phases. According to strategy/nexus-strategy.md, the orchestrator validates quality gates, coordinates the Evidence Collector for audit trails, and ensures deterministic hand-offs between phases to prevent context loss.

The pipeline operates as a linear progression from Phase 0 through Phase 6, with specific return paths available when the Reality Checker identifies critical defects during the hardening phase.

The 7 Workflow Phases Explained

Each phase in the NEXUS pipeline produces a concrete deliverable and requires passage through a quality gate before the Agents Orchestrator permits transition to the next stage.

Phase 0: Discovery

Core Goal: Gather market intelligence, user research, and technical feasibility signals using discovery agents.

Deliverable: An Executive Summary containing a GO decision that triggers Phase 1. This document is produced according to the guidelines in strategy/playbooks/phase-0-discovery.md.

Phase 1: Strategy & Architecture

Core Goal: Define product vision, high-level architecture, UX concepts, and compliance requirements.

Deliverable: An Architecture Package approved by the Studio Producer and Reality Checker agents. The hand-off package seeds Phase 2 scaffolding activities as detailed in strategy/playbooks/phase-1-strategy.md.

Phase 2: Foundation & Scaffolding

Core Goal: Construct the technical skeleton including repository layout, CI/CD pipeline, and infrastructure as code.

Deliverable: A Working skeleton with a fully operational DevOps pipeline and verified foundation evidence. This output is defined in strategy/playbooks/phase-2-foundation.md.

Phase 3: Build & Iterate

Core Goal: Execute short-duration sprints with continuous dev↔QA loops and performance baseline collection.

Deliverable: A Feature-complete application and Phase 4 hand-off package containing sprint deliverables, API specifications, and test results. See strategy/playbooks/phase-3-build.md for sprint criteria.

Phase 4: Quality & Hardening

Core Goal: Execute the Reality Checker automated QA, security scans, and performance audits.

Deliverable: A READY verdict (or NEEDS WORK / NOT READY) and Phase 5 hand-off package if the product meets stability criteria. Critical defects may trigger return paths to earlier phases. Documented in strategy/playbooks/phase-4-hardening.md.

Phase 5: Launch & Growth

Core Goal: Deploy to production, activate marketing pipelines, and monitor early-user metrics.

Deliverable: Stable launch status, growth KPIs, and Phase 6 hand-off package for ongoing operations. Refer to strategy/playbooks/phase-5-launch.md for launch readiness checklists.

Phase 6: Operate & Evolve

Core Goal: Continuous monitoring, incremental improvements, and periodic re-activation of the NEXUS pipeline for major features.

Deliverable: Ongoing operational dashboards, pipeline health metrics, and optional NEXUS-Sprint or NEXUS-Micro re-activations. Defined in strategy/playbooks/phase-6-operate.md.

How the Agents Orchestrator Manages Phase Transitions

The Agents Orchestrator enforces deterministic hand-offs between workflow phases to prevent context loss. According to strategy/coordination/agent-activation-prompts.md, the orchestrator validates that each quality gate produces observable evidence before permitting the next phase to commence.

Key transition mechanisms include:

• Quality Gates: Each phase requires specific deliverables (e.g., GO decision, READY verdict) before progression.
• Evidence Collection: The Evidence Collector records screenshots, pipeline logs, and test reports to create audit trails.
• Hand-off Packages: Structured context packages seed the next phase's agents with complete project state, eliminating cold-start problems.

Activating the Pipeline: Practical Examples

Teams can activate the NEXUS pipeline using the orchestration prompts defined in the repository.

Full Pipeline Activation

To launch the complete autonomous workflow from Discovery through Operation:

You are the Agents Orchestrator executing the NEXUS pipeline for **Acme-WebApp**.
Run the autonomous workflow: 
project-manager-senior → architect-ux → developer ↔ evidence-qa loop → testing-reality-checker.

This prompt initiates the full 7-phase cycle as documented in strategy/QUICKSTART.md.

Single Phase Activation

To run a specific phase, such as Phase 3 (Build & Iterate):

You are the Agents Orchestrator running **Phase 3 – Build & Iterate** for project **Acme-WebApp**.
- Spawn **frontend-developer**, **backend-architect**, **ai-engineer** agents.
- Enforce the dev↔qa loop with up-to-3 retries.
- When all tickets pass, generate the Phase 4 hand-off package.

This configuration is defined in strategy/coordination/agent-activation-prompts.md.

Evidence Collection Configuration

The Evidence Collector generates audit trails using YAML configurations:

# evidence-collector.yaml – example config used by the orchestrator

evidence:
  - type: screenshot
    path: ./evidence/phase2-ci-cd.png
  - type: pipeline-log
    url: https://github.com/msitarzewski/agency-agents/actions/runs/123456789

The orchestrator writes this file and attaches it to the hand-off package, as implemented in testing/testing-evidence-collector.md.

Summary

• The msitarzewski/agency-agents repository implements a 7-phase NEXUS pipeline (Phases 0-6) for autonomous multi-agent software development.
• Each workflow phase includes concrete deliverables, quality gates, and deterministic hand-off packages managed by the Agents Orchestrator.
• The pipeline progresses from Discovery through Operation, with the Reality Checker serving as the critical quality gate in Phase 4.
• Evidence Collection via YAML configurations creates audit trails that validate phase transitions and prevent context loss between autonomous agents.

Frequently Asked Questions

What triggers the transition between workflow phases in the Agency-Agents pipeline?

Phase transitions are triggered by quality gate approvals from the Agents Orchestrator. Each phase produces specific deliverables—such as the GO decision in Phase 0 or the READY verdict in Phase 4—that must be validated before the orchestrator releases the hand-off package to the next phase's specialist agents.

How does the Reality Checker function within the autonomous pipeline?

The Reality Checker operates during Phase 4: Quality & Hardening as an automated QA agent that executes security scans, performance audits, and integration tests. It returns a READY, NEEDS WORK, or NOT READY verdict. Critical defects can trigger return paths to earlier phases rather than permitting progression to launch.

What is the purpose of the Evidence Collector in workflow phase execution?

The Evidence Collector creates immutable audit trails for each phase transition by recording screenshots, pipeline logs, and test reports in YAML configuration files. This evidence validates that quality gates were properly satisfied and provides complete context for the next phase's agents, eliminating cold-start problems and ensuring deterministic hand-offs.

Can individual workflow phases be executed independently in the Agency-Agents pipeline?

Yes, the Agents Orchestrator supports single-phase activation through specific prompts defined in strategy/coordination/agent-activation-prompts.md. Teams can spawn specialist agents for a specific phase—such as Phase 3: Build & Iterate—while maintaining the same quality gates and evidence collection requirements as the full pipeline execution.

{
  "mcpServers": {
    "instagit": {
      "command": "npx",
      "args": ["-y", "instagit@latest"]
    }
  }
}