AgentX

AgentX 9.2.0

AI delivery that leaves evidence.

AgentX turns coding agents into a structured software delivery team. Specialist roles plan, build, review, test, and capture what they learned while mechanical gates keep the work honest.

Install AgentXExplore the repository
26specialized agents
134production skills
15visible roles
1/2/3/5risk-based quality passes

Why AgentX

Structure before scale.

Ad hoc prompts can produce code. AgentX adds the operating system around that code: routing, durable state, role contracts, independent review, and proof that the change works on the current revision.

Repo first

Context comes from source

Agents read project instructions, specs, plans, and domain skills before generating. The repository stays the system of record.

Evidence first

Claims need a gate

Tests, coverage, security scans, browser checks, and review findings are recorded before work can move forward.

Learning first

Delivery compounds

Decisions, pitfalls, reviews, and reusable patterns survive the chat window and improve the next task.

The operating loop

From intent to reviewed change.

AgentX resolves work through six durable checkpoints. The names stay consistent across chat, CLI, issues, plans, reviews, and the VS Code sidebars.

01BrainstormFrame the problem and retrieve prior learning.
02PlanDefine scope, alternatives, risks, and evidence.
03WorkImplement a bounded slice against the plan.
04ReviewRun tests and independent severity-based review.
05CapturePromote reusable decisions and pitfalls.
06DoneClose only when delivery and evidence agree.

Specialist team

Use one orchestrator or one role.

AgentX Auto can execute end to end, or you can select a specialist when the work needs a narrower contract. Internal reviewers and operators add isolation without cluttering the main picker.

Product

Product Manager and Agile Coach

PRDs, outcomes, roadmaps, stories, and acceptance criteria.

Design

UX Designer and Architect

Accessible prototypes, ADRs, technical specifications, and tradeoff decisions.

AI and data

Data Scientist and Fabric Engineer

Agent evaluation, RAG and model contracts, Lakehouse, Warehouse, Spark, quality, and lineage.

Build

Engineer and DevOps

Production code, tests, pipelines, packaging, deployment automation, and rollback.

Business apps

Power Platform and Power BI

Portable Dataverse solutions, apps, flows, PCF, Copilot Studio, semantic models, and DAX.

Assurance

Reviewer, Tester, and Research

Independent review, certification, domain research, and durable findings.

Where it runs

One framework, multiple adapters.

Choose the interaction and model path that fits the workspace. Agent contracts and repository evidence remain stable across providers.

VS Code and GitHub Copilot

Declarative chat agents, sidebar guidance, Command Palette workflows, workspace initialization, and secret-backed adapter setup.

  • 15 selectable chat roles
  • 134 discoverable skills
  • GitHub, Azure DevOps, or local issues

Claude, OpenAI, and local models

Use Claude Subscription, Anthropic API, OpenAI API, or Claude Code routed through LiteLLM and Ollama.

  • Workspace-scoped adapters
  • Secrets stay out of repository files
  • Role contracts remain provider-neutral

CLI and editor portability

Use the PowerShell/Bash CLI, GitHub Copilot CLI pack, Claude Code commands, or Cursor rules and commands.

  • Windows, Linux, and macOS
  • GitHub and Azure Pipelines gates
  • Zero-copy VS Code runtime

Secure WhatsApp companion

Operate a local workspace from an allowlisted account with read-only defaults and confirmation-gated mutation.

  • Replay and sender checks
  • Transcript-first voice handling
  • Bounded, secret-reduced execution

Control layer

Autonomy with boundaries.

AgentX does not treat a model's confidence as authorization. Security policy, evidence freshness, provider permissions, and role-specific boundaries are enforced outside the prompt.

Command safety. Destructive patterns are blocked and unfamiliar commands require confirmation.
Evidence freshness. Reused or stale artifacts cannot close the quality loop.
Role boundaries. Power Platform generation cannot authenticate to or mutate a tenant.
Supply chain. Pinned actions, audits, SBOMs, and provenance attestations protect releases.

Start in VS Code

Install, then initialize each workspace.

The Marketplace extension carries the runtime. The workspace keeps local state and stable launchers, so projects remain reviewable without copying the full AgentX asset tree.

code --install-extension jnPiyush.agentx

First workspace

  1. Open the repository in VS Code.
  2. Run AgentX: Initialize Local Runtime.
  3. Optionally add a remote or LLM adapter.
  4. Select AgentX Auto or a specialist role.
Open MarketplaceRead the guide