From a YouTrack issue to an app deployed on staging. With no human involvement. An orchestrator plus 11 specialised AI agents, 18 pipeline steps, 7 autonomous workflows.
A YouTrack webhook starts the pipeline through a webhook receiver. The script picks up a new Open issue and hands it over to an AI agent (Claude / Gemini).
Full REST API integration — reading issues and comments, changing states, publishing to the Knowledge Base. Newer comments take precedence over the original description.
The main Claude session drives the whole workflow. It delegates work to 11 specialised agents — it never implements anything directly, not even trivial changes.
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.
Every issue runs in its own Docker Compose project with an isolated environment. Automatic health check and rollback on failure.
Reverse proxy configured automatically over the API. A wildcard SSL certificate covers the staging domain. The proxy host is never deleted.
Loading the issue from the YouTrack API — description, requirements, comments. Deciding whether the architect, designer, graphics and game graphics agents are needed, based on keywords.
Issue → In Progress. The first visible action in YouTrack. The pipeline script also posts a comment with the agent process PID.
Creating a dedicated working directory for the issue on the staging server.
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.
If CLAUDE.md is missing from the project root (or is older than 30 days), the agent analyses the codebase and generates the project documentation (tech stack, structure, conventions, build commands). Every following agent loads it automatically.
Technical design — data model, API specification, dependencies, implementation order. Runs whenever non-trivial technical decisions are involved.
UI/UX design — layout, components, visual specification. It identifies the graphic assets needed and signals that via NEEDS_ASSETS=YES.
Generates general graphic assets through the OpenRouter API. The cost is aggregated into openrouter_cost_usd in the log. Max 1 retry.
Game graphic assets (sprites, textures, character art) through Leonardo AI. The choice between 7 and 7b follows game-specific keywords. Max 1 retry.
Writes the code and tests and verifies the build. It receives the full context from the architect, designer and graphics agents as part of its prompt.
Code review — quality, security, conventions, tech debt detection. Max 2 iterations with the Coder. A non-empty ### Tech Debt section triggers step 15b.
Unit tests, build verification, Docker build (without up -d). Max 2 iterations with the Coder.
A PR is opened on GitHub describing the changes. On an incremental re-run the existing PR is updated automatically.
docker compose -p <issue-id> up -d — the service starts in an isolated container.
A docker compose ps check, crash indicators in the logs and an HTTP health check via curl. On failure → coder → redeploy (max 1 iteration).
The NPM API creates a proxy host for the staging URL of the issue (wildcard SSL) pointing at the container through the Docker bridge.
Playwright screenshots of the running UI → visual verification against the assignment. Max 1 iteration with the Coder plus a redeploy.
The final summary of the whole workflow — what was implemented, the PR link, the staging URL and how the agents performed.
If the reviewer returned a non-empty ### Tech Debt section and the workflow finished successfully, the orchestrator creates up to 2 sub-tasks in YouTrack (Type=Task, Priority=Low, State=Backlog) and links them to the parent issue.
Issue → To Verify. A comment with the summary, the PR URL and the staging URL.
A JSONL record in the central run log, including token telemetry (agents_run[].tokens, top-level totals) and openrouter_cost_usd. Agent token counts are filled in after the run by a deterministic post-processor.
Writing .pipeline_state (JSON) into the issue directory — it holds last_run_at, pr_url, commit_sha and branch. This is what makes an incremental re-run possible.
Generates CLAUDE.md in the project root — tech stack, structure, conventions. It passes that context to every following agent.
Technical design, data model, API specification. It analyses the codebase and proposes a technical plan.
SUCCESS | BLOCKEDUI/UX design, layout, components, visual specification. It identifies which graphic assets are needed.
SUCCESS | BLOCKEDGenerates general graphic assets (logos, icons, banners) through the OpenRouter API.
SUCCESS | PARTIAL | BLOCKED max 1 retryGame graphic assets (sprites, textures, character art, pixel art) through Leonardo AI.
SUCCESS | PARTIAL | BLOCKED max 1 retryThe heart of the pipeline. It writes the code, writes the tests and verifies the build. It consumes feedback from the Reviewer, the Tester, the Health Check and the Visual Tester.
SUCCESS | PARTIAL | BLOCKEDCode review focused on code quality, security, adherence to conventions and tech debt detection.
APPROVED | CHANGES_REQUESTED max 2 iterationsRuns the unit tests, verifies the build and builds the Docker image without up -d (so that no port conflict arises).
Takes screenshots of the deployed UI with Playwright and verifies the implementation visually against the assignment.
PASS | FAIL max 1 iterationCompiles the final summary of the whole workflow — what was implemented, the PR link, the staging URL and how the agents performed.
SUCCESSFor non-pipeline workflows: analyst (design documents), sprint-planner (KB → issues), workflow-improver (daily cron), kb-updater (daily KB update).
see WorkflowsThe 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.
.pipeline_state)Graphics Agent
Max 1 iteration
Game Graphics Agent
Max 1 iteration
Coder ↔ Reviewer
Max 2 iterations
Coder ↔ Tester
Max 2 iterations
Coder ↔ Health Check
Max 1 iteration
Coder ↔ Visual Tester
Max 1 iteration
The full 18-step workflow from an Open issue to an app deployed on staging. Delegation to 11 agents, feedback loops, telemetry.
A shortened workflow activated through .pipeline_state when the user sends an issue back for rework. It skips setup, clone, conditional agents and PR creation. Around 50 % time saved.
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.
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".
Decomposition of a KB document into structured YouTrack issues with dependencies and a sprint plan. Triggered by a request such as "create issues from [KB link]".
A daily cron agent that reviews the current state of the workflow (agents, CLAUDE.md, scripts) and proposes 1–2 concrete improvements as YouTrack issues.
Daily content generation for two KB articles in the aipowereddevelopment project: 2 new tips on Claude/AI development (article A-1) and 1 piece of AI news (article A-2). The agent has a minimal toolset (Read, WebSearch, WebFetch) — no API credentials at all, every write is done by the orchestrator after the output has been sanitised.
Every run is written as a single JSONL line into the central log. Token telemetry is filled in after the run by a deterministic post-processor.
The JSON file .pipeline_state in the issue directory — it enables the incremental re-run. When the pipeline script detects it, it adds INCREMENTAL_RERUN=true to the prompt.
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.
After every agent finishes, a comment is posted to the YouTrack issue with a prefix matching the agent:
When Done issues are cleaned up, the pipeline automatically: