Jira
Structured issue and sprint management for teams that need deep workflows, permissions, reporting, and integrations.
- Strong epics, stories, tasks, bugs, sprints
- Workflow states, reports, dashboards
- External system option for client projects
SYSTEM OPERATIONAL
LLM DRIVENAJWAIN AI
Where human intent becomes coordinated engineering delivery.
Explore how Claude Code coordinates specialist agents through MCP to plan, execute, review and record engineering work — with Ajwain as the single source of truth.
Human
Claude / agent
MCP / tool call
AJWAIN OS
External system
request path · livesolid = request · dashed = result
/map live system map
Every layer, one picture. Click anything.
Five layers sit between a person and the systems that hold real data. Hover a node to trace its connections; click it to inspect. Pick a layer to isolate it.
/work-systems project records
Jira, ClickUp, or Ajwain?
These are the work-management surfaces for the delivery system. They organize the complete record of a project: epics, stories, tasks, bugs, documents, test evidence, releases, and the people responsible for each decision.
ClickUp
A flexible workspace combining tasks, docs, goals, views, comments, and lightweight project operations in one place.
- Tasks, lists, boards, docs, goals
- Flexible views and team collaboration
- External system option for client projects
Ajwain
The project’s own Jira/ClickUp-style engineering workspace. Claude, agents, the browser portal, and MCP all write to the same governed record.
- Epics, stories, features, tasks, bugs
- Documents, files, comments, tests, releases
- Sprints, incidents, support tickets, audit history
epicstorytaskbugdocumentrelease
/playground agent simulation
Run a workflow. Inspect every event.
Pick a scenario and press run. Each step activates in order, approvals stop and wait for you, and every event is clickable. This is a demonstration of the documented flow — the browser is not calling OS, MCP or any real system.
DEMO / SIMULATED RUN
idle
event stream0 events
Run a scenario — events appear here. Click any event to inspect it.
/lifecycle the coordinator's loop
How an agent actually works
The coordinator is the only hub. It understands the request, loads context, plans, delegates with a precise brief, collects reports, verifies, and records the result. Click any stage.
/mcp connector layer
One logged phone line to every real system.
No agent touches a real system directly. Every call is loaded on demand, screened by the permission layer, sent through the MCP client to the right server, and returned as structured data.
Send a tool call
/agents agent matrix
Some hold a pen. Most hold a magnifying glass.
19 named specialists plus the coordinator. Only the write group can change files, so asking a reviewer to review can never accidentally change anything. Click a row for the full profile.
| Agent | Purpose | Access | Triggered when |
|---|
/safety safety center
The assistant can ask. The records room decides.
Security is enforced at two points: local permission rules and hooks on the machine, and independent access control on the OS server. Nothing an agent asserts locally widens what the backend allows.
Agents assist. Humans remain accountable for important decisions.
/teams team workflows
One tool. Four ways of working.
Each machine carries a work-style role that changes how Claude behaves. People own judgement, sign-off and client contact; agents own execution, review and evidence gathering.
/teams/qa DeepEval
How QA measures LLM output with DeepEval
DeepEval is an open-source, pytest-style framework for testing LLM output. QA writes test cases, attaches metrics with thresholds, and fails the run when a score drops below the gate. Drag the thresholds to see the gate respond.
Test cases→Metrics + thresholds→deepeval test run→Release gate→Test run in OS
Recording results as an OS test run (start_test_run) is the recommended wiring, matching the document's call for scheduled, recorded regression.
Gate simulatorDEMO SCORES
/jev JEV AI
JEV AI
JEV AI isn't described in the architecture document, so this page doesn't guess what it does. The slot below is wired in and ready: fill the JEV_AI data object and it renders here.
/ecosystem agents in use & in the market
What we run, and what else is out there.
Green cards are part of this setup today. The rest are market alternatives, grouped by the job they would do. The market view is a general overview — check each vendor's docs for current features and pricing.
/activity agent activity
One timeline of what every agent did.
The document recommends a central agent-action audit view. This is what it could look like, assembled from the playground scenarios. DEMO DATA
| Time | Actor | Action | Connector · tool | Status | Result |
|---|
/status implementation review
What's built. What's next.
From the document's independent review — built today, switched on but lightly used, available but not adopted, and recommended next.
| Capability | Status | Notes |
|---|
/learn learning hub
Study path for AI engineering.
Eleven skill tracks from the September 2026 job-posting review, each with free material and a portfolio project. Pick your target role, tick resources as you finish them — progress saves in this browser.
What postings ask for
0%
0 of 0 items done
| Target role | First portfolio project | Add next |
|---|
AgentOps glossary
evidence sources behind the posting figures