Autonomous Agent Organization

The Team.

Project development, research & competitive intelligence, and product management — all driven by specialized AI agents with structured handoffs and quality gates between every stage.

View ProjectsRead Blog
Agents
14
Parallel Tracks
3
Commands
21
Artifact Types
12
Architecture

Three parallel tracks, zero overlap.

Each track has its own agents, commands, and output artifacts. They share a common orchestration layer and agent identity model but run independently — you can launch a research cycle while a project pipeline is mid-execution.

11
Dev Agents
3
Researchers
3
Product Managers
21
Commands
The Three Tracks

How the workspace is organized.

01Agents 00-09, 99

Project Development

Staged pipeline from raw idea to deployed code. Work enters at three weights — a full PRD, a bounded feature (FRD), or a contained bug — over one shared engine. Rubber Ducky brainstorm, spec with adversarial review, architecture + ADRs, UI design with clickdummies, task breakdown with § 4c contracts, TDD enrichment, and slim-context implementation.

/prime/rubber-ducky/create-prd/create-frd/create-bug/architect/task-breakdown/implement/code-review/test/document/ship
02Agents 10-12

Research & CI

Vendor-neutral technology intelligence, competitive analysis, and a developer-experience industry brief across four research lenses: release tracking, security monitoring, ecosystem trends, and community voice. The track outputs strategic initiatives, not a task backlog.

/release-the-minions
03Agents PM-S, PM-J, PM-C

Product Management

Technical PM evaluation of managed Sitecore repositories — SitecoreAI Starter, JSS monorepo, and Content SDK. Each PM specialization consumes upstream research and CI briefs, then produces a register of initiatives gated by altitude and control — the bets that repo actually owns.

/release-the-minions
The Core Pipeline

From idea to production.

Seven agents in sequence: Team Lead orchestrates, Product Manager specs, Architect designs, UI Designer proposes, Lead Developer breaks down, QA enriches, Developer implements. A final /ship stage closes the loop with a release checklist. Each output becomes the next input through structured handoffs.

Orbital view — CSS 3D component
00
Team LeadOrchestration & Gates
Agent 00Orchestration & Gates
Sequences specialist agents, validates handoffs between stages, and enforces quality gates. Low ceremony — stops bad artifacts before they propagate downstream.
02
Product ManagerPRD & Spec Review
Agent 02PRD & Spec Review
Writes testable specs from brain dumps, defends MVP scope, and runs adversarial 'pessimistic roast' sessions to stress-test requirements before they reach architecture.
04
Software ArchitectArchitecture & ADRs
Agent 04Architecture & ADRs
Converts PRDs into concrete technical blueprints with explicit Architecture Decision Records. Consequence-oriented — names trade-offs and checks for conflicts with existing decisions.
05
UI DesignerDesign Proposals & Clickdummies
Agent 05Design Proposals & Clickdummies
Produces 1-3 fundamentally divergent design directions with full rationale and self-contained HTML clickdummies for browser evaluation. Research-grounded aesthetics — asks constraints before designing, not after.
06
Lead DeveloperTask Breakdown
Agent 06Task Breakdown
Breaks epics into atomic tasks with stable IDs and the mandatory § 4c execution contract — seven subsections that give the Developer everything needed without opening other docs.
07
QA SpecialistTDD Enrichment
Agent 07TDD Enrichment
Embeds test-first discipline directly into the task breakdown before implementation starts. Edits tasks in-place, reorders for RED-GREEN-REFACTOR, ensures tests are behavioral.
08
DeveloperImplementation
Agent 08Implementation
Slim-context executor who receives only prd-minimal + enriched task breakdown. Implements exactly what was planned — stops and asks rather than compensating for ambiguous specs.
On-Demand Specialists

Called when needed.

Four agents that operate outside the linear core pipeline. They are pulled in when the context calls for them — to surface hidden assumptions before planning, ground a fuzzy claim mid-conversation, extract structure from an existing site, or run a heavy documentation pass.

01
Rubber Ducky
Brainstorm & Discovery

Interrupt-first brainstorming facilitator that forces clarity through relentless questioning. Surfaces gaps, assumptions, and scope creep before they become costly.

Invoked via /rubber-ducky — standalone, pre-pipeline. Produces a concept brief that feeds into PRD creation.
03
Research Sidecar
Inline Research Delegate

On-demand research delegate invoked mid-conversation by the Rubber Ducky or Product Manager to ground a fuzzy claim. Returns a short, decision-grade synthesis from live sources — without polluting the caller's context.

Pulled in mid-conversation via /researchby the Rubber Ducky or Product Manager — returns a decision-grade synthesis from live sources without polluting the caller’s context.
09
Doc Writer
Documentation Passes

On-demand technical writer for heavy documentation — full README, architecture docs, decision logs, changelog. Invoked by /document after major milestones; /ship emits lightweight docs inline without it.

Activated via /document for heavy documentation passes — full README, architecture, and decision logs. /ship emits lightweight docs inline without it.
99
Source Analyst
Source Discovery & Parity

Analyses reference URLs in greenfield mode (lightweight) or crawls a live site for full parity rebuild. Never invents content — labels everything as extracted or inferred.

Activated during /prime when a URL is provided. Runs in greenfield (lightweight reference) or rebuild (full parity crawl) mode.
Development Track

The builders.

11 agents that take a raw idea through brainstorming, specification, architecture, design, planning, testing, implementation, and shipping. 7 form the linear core pipeline; 4 are on-demand specialists — Rubber Ducky, Research Sidecar, Doc Writer, and Source Analyst — invoked when the context calls for them.

00

Team Lead

Orchestration & Gates

Sequences specialist agents, validates handoffs between stages, and enforces quality gates. Low ceremony — stops bad artifacts before they propagate downstream.

01

Rubber Ducky

Brainstorm & Discovery

Interrupt-first brainstorming facilitator that forces clarity through relentless questioning. Surfaces gaps, assumptions, and scope creep before they become costly.

02

Product Manager

PRD & Spec Review

Writes testable specs from brain dumps, defends MVP scope, and runs adversarial 'pessimistic roast' sessions to stress-test requirements before they reach architecture.

03

Research Sidecar

Inline Research Delegate

On-demand research delegate invoked mid-conversation by the Rubber Ducky or Product Manager to ground a fuzzy claim. Returns a short, decision-grade synthesis from live sources — without polluting the caller's context.

04

Software Architect

Architecture & ADRs

Converts PRDs into concrete technical blueprints with explicit Architecture Decision Records. Consequence-oriented — names trade-offs and checks for conflicts with existing decisions.

05

UI Designer

Design Proposals & Clickdummies

Produces 1-3 fundamentally divergent design directions with full rationale and self-contained HTML clickdummies for browser evaluation. Research-grounded aesthetics — asks constraints before designing, not after.

06

Lead Developer

Task Breakdown

Breaks epics into atomic tasks with stable IDs and the mandatory § 4c execution contract — seven subsections that give the Developer everything needed without opening other docs.

07

QA Specialist

TDD Enrichment

Embeds test-first discipline directly into the task breakdown before implementation starts. Edits tasks in-place, reorders for RED-GREEN-REFACTOR, ensures tests are behavioral.

08

Developer

Implementation

Slim-context executor who receives only prd-minimal + enriched task breakdown. Implements exactly what was planned — stops and asks rather than compensating for ambiguous specs.

09

Doc Writer

Documentation Passes

On-demand technical writer for heavy documentation — full README, architecture docs, decision logs, changelog. Invoked by /document after major milestones; /ship emits lightweight docs inline without it.

99

Source Analyst

Source Discovery & Parity

Analyses reference URLs in greenfield mode (lightweight) or crawls a live site for full parity rebuild. Never invents content — labels everything as extracted or inferred.

Slim Context Chain
Later stages load only what they need. The Developer (08) receives prd-minimal + enriched task breakdown — never the full PRD, architecture, or UI specs. If the task breakdown's § 4c execution contract is incomplete, that's a planning gap to fix upstream, not a context gap for the implementer to fill.
Research & CI Track

The researchers.

3 specialists producing vendor-neutral intelligence briefs. The Research Specialist covers technology topics across the modern web stack, the CI Specialist profiles competing platforms with factual SWOT analysis, and the DevX Specialist distils a developer-experience industry trend brief. All apply four shared lenses: release tracking, security monitoring, ecosystem trends, and community voice — and the track now outputs strategic initiatives, not a task backlog.

10

Research Specialist

Technology Intelligence

Produces vendor-neutral intelligence briefs across four research lenses: release tracking, security monitoring, ecosystem trends, and community voice. One merged report per technology topic.

11

CI Specialist

Competitive Analysis

Profiles competing platforms with factual SWOT analysis, version tables, and ecosystem assessment. Supplier of downstream PM synthesis.

12

DevX Specialist

DevEx Industry Trends

Produces a single trend-only developer-experience industry brief from signal-quality evidence — the shared get-ahead driver consumed by every PM specialization alongside the technology and CI briefs.

Product Management Track

The product managers.

3 specialized PMs that evaluate managed Sitecore repositories as product surfaces. Each consumes upstream research and CI briefs, inspects the actual repo, and produces a register of initiatives — JPD-altitude bets gated by altitude and control — rather than a granular task backlog. A chosen initiative is promoted into delivery via /create-prd or /create-frd.

PM-S

SCB Product Manager

SitecoreAI Starter Evaluation

Technical PM for the xmcloud-starter-js repository. Evaluates starter layout, rendering host assumptions, template maintenance, and alignment with current Sitecore direction.

PM-J

JSS Product Manager

JSS Monorepo Assessment

Technical PM for the JSS monorepo. Assesses package topology, peer dependencies, migration pressure, release channels, and developer friction across packages.

PM-C

Content SDK Product Manager

Content SDK Lifecycle

Technical PM for the Content SDK repository. Focuses on package maturity, scaffolding quality, docs alignment, and the JSS-to-Content-SDK lifecycle transition.

Sitecore Analysis

Deep-dive into any Sitecore site.

A dedicated command that combines content model extraction, live Edge layout analysis, and component mapping into a comprehensive site profile — feeding directly into project development or research workflows.

SC
/analyze-sitecore-site
Site Intelligence Command

Extracts the content model from serialized YML items (templates, renderings, field definitions, headless variants), fetches live Experience Edge layout data for runtime placeholder trees and component keys, and produces paired analysis reports — a content model summary and content examples document that downstream agents use for component scaffolding and implementation.

Content Model
Templates, fields, renderings from serialized YML
Edge Layout
Live placeholder trees and componentName keys
Paired Reports
*-content-model.md + *-content-examples.md
Dogfood Methodology

Skills trained by shipping real apps.

The Marketplace SDK skill files in this workspace weren't written from documentation alone — they were validated by building real apps against them. Nine consecutive dogfood runs (plus a dry run) were used to surface gaps, fix incorrect assumptions, and battle-test every code pattern before it was committed as a skill.

Runs 01–03

Scaffold validation

Prove the scaffold command, SDK wiring, and base client pattern. Surface gaps in how the skill described ClientSDK.init and the XMC module registration.

Runs 04–06

Extension point coverage

Exercise Pages Context Panel (pages.context subscription), fullscreen, and sidebar variants. Patch the skill files each time a code example produced a type error or runtime mismatch.

Runs 07–09

Agent API + auth layer

Validate OAuth client-credentials flow, server-side token caching, screenshot and preview-URL endpoints. Each failed fetch corrected the auth.md and agent-api.md skills in place.

9
Dogfood runs
6
Skill files trained
4
Shipped apps (PageShot, QuickCopy, Atlas, Last-Edit Trail)
0
Assumptions left unverified

The rule is simple: a skill file is only trustworthy if an agent has shipped a real app using only that skill as its source of truth. Each dogfood run starts from a blank scaffold, loads the relevant skill files, and implements without reading the upstream Sitecore documentation directly. When the code breaks or the types don't align, the skill file — not a workaround — gets fixed. PageShot, QuickCopy, Component Atlas, and Last-Edit Trail are the shipped results of this cycle. The skills now live in the sitecore Claude Code plugin (Chris1415/sitecore-skills-pocs) — auto-discovered, invoked via Skill(skill: “sitecore:<name>”), and reflecting exactly what it takes to build and ship a Marketplace app today.

The Skills, Packaged

Where the dogfood patches live now.

The 46 skill files trained by the dogfood loop don't ship inside this repo anymore. They live in a separate, auto-discoverable Claude Code plugin and a Cursor / Codex / Cline / Gemini CLI distribution — installed once per environment, then invoked by name from any agent.

46
Skill files
5
Domain prefixes
1
Install command per consumer
0
Stale per-project copies
Claude Code
/plugin marketplace add Chris1415/sitecore-skills-pocs
/plugin install sitecore@sitecore-skills-pocs
Cursor / Codex / Cline / Gemini CLI
npx skills add Chris1415/sitecore-skills-pocs \
  --skill <name>

Once installed, every agent in this workspace invokes a skill by name — Skill(skill: “sitecore:setup-scaffold”), Skill(skill: “sitecore:marketplace-sdk-client”), Skill(skill: “sitecore:blok-setup”) — instead of reading skill files from disk. The plugin auto-discovers, the prefix sort organises (content-sdk · marketplace-sdk · blok · sitecoreai · setup), and patches accrue once instead of being copy-pasted into every project.

Chris1415 / sitecore-skills-pocs →
Core Principle: Identity / Command Separation
Agent files define persona and expertise only — no task instructions, no output formats. Commands inject the task, input/output contracts, and handoff rules at runtime. The Product Manager can write a PRD, challenge a PRD, or revise a PRD — depending on which command invokes it and in what mode. This separation means any agent can be reused for different tasks without creating new agent files.
Explore the Framework
Where this leads

Go deeper into how this gets shipped

Methodologies and case studies show the agentic pipeline in motion — the patterns the agents converge on, and what that produced when run end-to-end against real Sitecore tenants.

Read the methodologiesRead the case studies
Interested?

Build your own agentic workflow

This framework is extensible — from greenfield apps to Sitecore Marketplace modules and Content SDK head apps. The same agents adapt to any domain through command injection, not prompt rewrites.