Fully autonomous

AI-Powered
Development Pipeline

From a YouTrack issue to an app deployed on staging. With no human involvement. An orchestrator plus 18 specialised AI agents, 18 pipeline steps, 12 autonomous workflows.

YouTrack Claude Code OpenAI Codex Gemini GitHub Docker Nginx Proxy Manager Playwright OpenRouter Leonardo AI ElevenLabs
This presentation was built by the very AI-powered pipeline it describes Built by the AI-powered pipeline
// Architecture

System overview

🔗

Webhook Trigger

A YouTrack webhook starts the pipeline. It picks up a new Open issue and hands it over to the assigned AI agent — Claude Code, OpenAI Codex or Gemini.

📋

YouTrack Integration

Full REST API integration — reading issues and comments, changing states, publishing to the Knowledge Base. Newer comments take precedence over the original description.

🤖

AI Orchestrator

The main Claude session drives the whole workflow. It first recognises the type of task, then delegates work to 18 specialised agents — it never implements anything directly, not even trivial changes. Tasks for Codex go through the same coder → reviewer → tester cycle.

🐙

GitHub CI/CD

Automatic clone, a branch named after the issue ID, commit, push and Pull Request creation. On an incremental re-run the existing PR is updated instead.

🐳

Docker Deployment

Every issue runs in its own Docker Compose project with an isolated environment. Automatic health check, with a fix and redeploy on failure.

🌐

Nginx Proxy Manager

Reverse proxy configured automatically over the API. A wildcard SSL certificate covers the staging domain. The proxy host is never deleted.

// Workflow

18 pipeline steps

01

Research direct

Loading the issue from YouTrack — description, requirements, comments. Choosing the workflow type and deciding whether the architect, designer, graphics, game graphics or voice-over agent is needed.

02

State change direct

Issue → In Progress. The first visible action in YouTrack.

Complexity classification direct

The task is classed as trivial, standard or complex. A small text change skips the design steps; a large change makes the architect mandatory.

03

Setup direct

Creating a dedicated working directory for the issue on the staging server.

04

Clone & Branch direct

Git clone of the repository and creation of a feature branch. The source branch from YouTrack is respected, otherwise the pipeline falls back to main/master.

4b

📄 Project Doc Agent conditional

If the project lacks up-to-date project documentation, the agent analyses the codebase and writes it (tech stack, structure, conventions, build). Every following agent loads it automatically.

05

🏗 Architect Agent conditional

Technical design — data model, API specification, dependencies, implementation order. Runs whenever non-trivial technical decisions are involved.

06

🎨 Designer Agent conditional

UI/UX design — layout, components, visual specification and design direction (palette, typefaces). It decides which graphic assets need to be generated.

07

🖼 Graphics Agent conditional

Generates general graphic assets (logos, icons, illustrations) through OpenRouter. The cost is recorded in the run log.

7b

🎮 Game Graphics Agent conditional

Game graphic assets (sprites, textures, character art) through Leonardo AI. Used instead of, or alongside, general graphics when the project is a game.

🎙 Voice-over Agent conditional

Character lines are voiced through ElevenLabs and levelled to a consistent loudness. If a line fails, the pipeline carries on with subtitles.

08

🔧 Coder Agent agent

Writes the code and tests and verifies the build. It receives the full context from the architect, designer, graphics and voice-over agents. For UI tasks it checks its own result on screenshots.

09

📋 Reviewer Agent agent

Code review — quality, security, conventions. Small issues in new code are fixed straight away, larger debt becomes separate tasks (step 15b).

10

🧪 Tester Agent agent

Unit tests, build verification and a Docker image build — without starting any containers, so it never clashes with the real deployment.

11

Pull Request direct

A PR is opened on GitHub describing the changes. On an incremental re-run the existing PR is updated automatically.

12

Deployment direct

The service starts in the issue's own isolated Docker Compose project.

12b

Health Check direct

Are all containers running, are the logs free of crashes and does the service answer over HTTP? On failure → coder → redeploy.

13

Proxy configuration direct

Nginx Proxy Manager publishes the service on the project's staging URL with a wildcard SSL certificate.

14

👁 Visual Tester Agent temporarily off

Screenshots of the running UI and a visual check against the assignment. The step is switched off for now — the coder checks the look itself and the E2E tester runs click-through checks on request.

15

📝 Final summary direct

What was implemented, the PR, the staging URL and how the agents performed. The orchestrator writes the summary itself for now; the Summarizer agent is temporarily off.

15b

🔧 Tech Debt sub-tasks conditional

If the reviewer found technical debt and the workflow finished successfully, the orchestrator creates follow-up sub-tasks in YouTrack, tagged by the type of debt.

16

Finalisation direct

Issue → To Verify. A comment with the summary, the PR URL and the staging URL.

17

Workflow Execution Log direct

A record of the run in the central log — agents, results, iterations, token usage and cost. The record is validated before it is written.

18

Pipeline State File direct

Saving the state of the run (time, PR, staging, last commit, branch). This is what makes an incremental re-run possible.

// Specialisation

18 AI agents

Conditional
📄

Project Doc Agent

Writes the project documentation — tech stack, structure, conventions. It passes that context to every following agent.

Claude Sonnet SUCCESS | PARTIAL | BLOCKED
Conditional
🏗

Architect Agent

Technical design, data model, API specification. It analyses the codebase and hands the coder a plan and a map of the affected code.

Claude Fable SUCCESS | BLOCKED
Conditional
🎨

Designer Agent

UI/UX design, layout, components, visual specification and design direction. It decides which graphic assets are needed.

Claude Opus SUCCESS | BLOCKED
Conditional
🖼

Graphics Agent

Generates general graphic assets (logos, icons, banners, illustrations).

OpenRouter SUCCESS | PARTIAL | BLOCKED
Conditional
🎮

Game Graphics Agent

Game graphic assets (sprites, textures, character art, pixel art).

Leonardo AI SUCCESS | PARTIAL | BLOCKED
Conditional
🎙

Voice-over Agent

Voices character lines, picks the best take and levels the loudness of the recordings.

ElevenLabs SUCCESS | PARTIAL | BLOCKED
Mandatory
🔧

Coder Agent

The heart of the pipeline. It writes the code and tests, verifies the build and checks UI work on screenshots. It consumes feedback from the Reviewer, the Tester and the Health Check.

Claude Opus SUCCESS | PARTIAL | BLOCKED
Mandatory
📋

Reviewer Agent

Code review focused on code quality, security and conventions. Small issues get fixed straight away, technical debt becomes separate tasks.

Claude Sonnet APPROVED | CHANGES_REQUESTED
Mandatory
🧪

Tester Agent

Runs the unit tests, verifies the build and builds the Docker image — without starting any containers.

Claude Sonnet PASS | FAIL
On request
🕹

E2E Tester Agent

Clicks through the deployed app like a user — navigates, fills in forms and checks the resulting behaviour.

Claude Sonnet PASS | FAIL | BLOCKED
Temporarily off
👁

Visual Tester Agent

Takes screenshots of the deployed UI and checks the implementation visually against the assignment, contrast included.

Claude Haiku PASS | FAIL
Temporarily off
📝

Summarizer Agent

The final workflow summary. Until further notice the orchestrator writes it itself.

Claude Haiku SUCCESS
Workflow
📊

Analyst Agent

Writes the application design — purpose, requirements, user stories, architecture, data model and UI concept.

Claude Opus SUCCESS | BLOCKED
Workflow
🔍

Problem Analyst Agent

Investigates a reported problem, assesses its severity and decides whether it is a real bug. If it is, it drafts the fix as a task.

Claude Fable PROBLEM | NOT_A_PROBLEM | NEEDS_MORE_INFO
Workflow
🖌

UI Variants Agent

Produces three visually distinct variants of a single screen as clickable HTML pages with previews.

Claude Opus SUCCESS | BLOCKED
Workflow
🗂

Sprint Planner Agent

Breaks a design document down into issues with acceptance criteria, dependencies and user stories.

Claude Opus SUCCESS | BLOCKED
Workflow
⚙

Workflow Improver Agent

Once a day it reviews the whole workflow and proposes 1–2 concrete improvements as new tasks.

Claude Opus SUCCESS
Workflow
📰

KB Updater Agent

Every day it finds new tips on working with AI and news from the AI world, each with a verified source.

Claude Sonnet SUCCESS | BLOCKED

🚫 The key rule

The orchestrator NEVER writes code, reviews, tests, designs architecture or UI directly — not even for a trivial one-line change. It always delegates to the proper agent. Implementing anything itself is a workflow violation.

// Flow diagram

Orchestration and feedback loops

▸ Orchestrator (the main Claude session)
├── Step 0: workflow type detection (UI variants, design, analysis, deployment, implementation)
├── Steps 1–2: research, state
├── Complexity classification (trivial / standard / complex)
├── Steps 3–4: setup, clone
├── Step 4b: → project-doc (project documentation) [CONDITIONAL]
├── Step 5: → architect (technical design) [CONDITIONAL]
├── Step 6: → designer (UI/UX design) [CONDITIONAL]
├── Step 7: → grafika (OpenRouter) [CONDITIONAL]
├── Step 7b: → herni-grafika (Leonardo AI) [CONDITIONAL]
├── → dabing (ElevenLabs) [CONDITIONAL]
├── Step 8: → coder (implementation)
├── Step 9: → reviewer (code review)
│      └── CHANGES_REQUESTED / small fixes → coder → reviewer (max 2×)
├── Step 10: → tester (tests)
│      └── FAIL → coder → tester (max 2×)
├── Steps 11–12: PR, deploy
├── Step 12b: health check
│      └── FAIL → coder → redeploy → health-check (max 1×)
├── Step 13: proxy configuration
├── Step 14: visual-tester [TEMPORARILY OFF]
├── Step 15: final summary + YouTrack comment (by the orchestrator, summarizer temporarily off)
├── Step 15b: Tech Debt sub-tasks [CONDITIONAL]
├── Step 16: finalisation → To Verify
├── Step 17: Workflow Execution Log (+ token/cost telemetry)
└── Step 18: Pipeline State File (for the incremental re-run)

⟳ Graphics retry

Graphics & Game Graphics
Max 1 iteration

⟳ Review loop

Coder ↔ Reviewer
Max 2 iterations

⟳ Test loop

Coder ↔ Tester
Max 2 iterations

⟳ Health check loop

Coder ↔ Health Check
Max 1 iteration

⟳ E2E loop

Coder ↔ E2E Tester
Max 1 iteration

// More workflows

12 autonomous workflows

🚀

Issue Handling (the main pipeline)

The full 18-step workflow from an Open issue to an app deployed on staging. Delegation to specialised agents, feedback loops, telemetry.

Open → 18 steps → To Verify
🔁

Incremental re-run

A shortened workflow for when the user sends an issue back for rework. It skips setup, clone, conditional agents and PR creation — the new comments are the main brief.

New comments → coder → reviewer → tester → redeploy → summary
🔗

Task series and User Stories

A larger feature is split into consecutive tasks that the pipeline works through one after another. Each task builds on the previous one; tests, the Pull Request and deployment happen once at the end.

User Story → Open → task 1 → task 2 → … → one PR
🖌

UI variants

Three visually distinct variants of a single screen as HTML pages. A person picks one in a comment and the chosen design is then implemented as a binding specification.

ui-navrhar → variants A / B / C → choice → implementation
🔍

Problem analysis

Investigating a reported problem without touching the code. If it turns out to be a real bug, a new task with a proposed fix is created.

problem-analyst → verdict → new task / comment
🚢

Deployment and operations

Deploying an existing branch, changing a domain or diagnosing staging without any code change. As soon as a file in the repository has to change, the run switches to a full implementation.

build → deployment → health check → proxy
🕹

E2E test (on request)

A click-through test of the running app, acting as a real user. It runs only on request; on failure the coder fixes it and the test is repeated.

e2e-tester → click-through scenarios → PASS / FAIL → coder → retest
✅

AutoVerify

Automatic merge of verified issues into the auto-verify branch and chaining on to the next issue. Per-project locking lets different projects run in parallel.

To Verify + autoVerify=true → Merge into auto-verify → State → Waiting → Next issue activated → Manual merge into main
📄

Design Document

Design documents written by the analyst, the architect and the designer, then published to the YouTrack Knowledge Base. Triggered by keywords such as "design", "specification" or "application plan".

analyst → architect → designer [CONDITIONAL] → Compilation → KB publication → To Verify
📅

Sprint Planning (KB → issues)

Decomposition of a KB document into structured YouTrack issues with dependencies and a sprint plan. Larger capabilities are grouped into User Stories, which then run as a series.

KB document → sprint-planner → Issues and User Stories in YouTrack → Dependencies linked
⚙

Workflow Improver

A daily agent that reviews the current state of the workflow (agents, rules, tools) and proposes 1–2 concrete improvements as YouTrack issues.

Analysis of agents, rules, tools → 1–2 proposals → Issues created in YouTrack
Daily
📰

KB Updater

A daily addition to the Knowledge Base: 2 tips on Claude/AI development and 1 piece of AI news. The agent may only read and search the web; the orchestrator writes the content after a duplicate check.

kb-updater (verified sources) → Duplicate check → Write to the KB → Tracking issue (To Verify)
Every morning
// Infrastructure

Deployment & conventions

🌐 Staging domain

URL a dedicated staging URL for every issue
SSL wildcard certificate for the staging domain
1 level → covered by the wildcard certificate ✓
2+ levels → not covered by the certificate ✗
// A wildcard cert covers a single subdomain level only
// a multi-level subdomain → SSL error
certificate = wildcard("staging domain").covers("1 level")

🐳 Docker

Project a separate Docker Compose project per issue
Working dir a separate directory per issue
Forward host Docker bridge gateway
NPM API internal, not reachable from outside
# Starting the issue's project
docker compose up <issue>

# Proxy: an existing host is updated, never deleted
proxy.upsert("staging domain")

📊 Workflow Execution Log (step 17)

Every run is written as a single record into the central log. Token usage is filled in after the run by a deterministic post-processor; the monitoring dashboard reads from the log.

issue, workflow type, complexity, duration
agents: role, result, iterations,
  tokens and cost
external services: graphics, voice-over
outcome: PR, staging, final state
tech debt: follow-up tasks created

💾 Pipeline State File (step 18)

The state of the last run, stored with the issue — it enables the incremental re-run. When the pipeline finds it on the next start and it belongs to the same issue, the shortened workflow is chosen.

{
  "last run": // time in UTC
  "pull request": // the existing PR
  "staging": // staging URL of the issue
  "commit": // the last commit
  "state": "To verify",
  "branch": // branch name
}
// Lifecycle

YouTrack issue states

Open
→
In Progress
→
To Verify
→
Done
The alternative path when a blocker appears:
In Progress
→
Waiting

🛑 Falling back to Waiting

This happens when the review/test loop runs out of iterations without success, when the health check fails, on a critical blocker with no way forward, or when input from the user is missing. The issue receives a comment explaining the reason and the information needed.

📝 YouTrack comments

After every agent finishes, a comment is posted to the YouTrack issue with a prefix matching the agent:

📄 Project Doc: <output>
🏗 Architect: <output>
🎨 Designer: <output>
🖼 Grafika: <output>
🎮 Herni Grafika: <output>
🎙 Dabing: <output>
🔧 Coder: <output>
📋 Reviewer: <output>
🧪 Tester: <output>
🕹 E2E Tester: <output>
📊 Analyst: <output>
🔍 Problem Analyst: <output>
🖌 UI Navrhar: <output>
🗂 Sprint Planner: <output>
📝 Summary: <run outcome>

🧹 Cleanup (the Done state)

When Done issues are cleaned up, the pipeline automatically:

# 1. Stops the containers of the issue
docker compose down

# 2. Removes the working directory
rm -rf <working directory>

# 3. The proxy host is NEVER deleted!
# (because of the SSL certificate)