Documentation
The full reference for Closedloop.ai — the spec-driven, AI-first delivery platform that closes the loop on your SDLC.
Welcome to the Closedloop.ai documentation hub.
Closedloop.ai is a platform for team-based agentic software development. It combines a browser-based control plane, a desktop execution client, and a suite of Claude Code plugins into a single system that turns scoped intent into shipped code through repeatable, artifact-bound loops.
This documentation is organized so both humans and AI agents can use it as ground truth:
- Getting started — the fastest path from install to a first completed loop.
- Mechanisms — how the control plane, desktop gateway, runtime, loops, judges, and self-learning actually work.
- Concepts — the mental models behind the product and the philosophy of spec-driven AI-first delivery.
- Workflows — the delivery paths from PRD to PR, including parallel features, code review, and cross-repo coordination.
- Resources — troubleshooting, FAQs, reference tables, comparisons, and templates.
- Glossary — precise, reusable definitions of every first-class concept.
Start here
Install and onboard
Install the desktop client, install the Claude Code plugins, and prepare your environment for your first loop.
Understand the control plane
Learn why Closedloop.ai treats coordination, artifacts, and execution as one coherent system.
Run the canonical workflow
Walk through the PRD → plan → critics → implementation → judges → PR path.
What ships with Closedloop.ai
| Surface | What it does |
|---|---|
| Web app (control plane) | Defines work, stores artifacts, coordinates teams, surfaces approvals, renders activity. |
| Desktop client | Electron app that exposes a signed localhost HTTP gateway and a cloud relay so the control plane can safely drive local git, filesystem, code review, and AI coding sessions. |
| Claude Code plugins | Six installable plugins — bootstrap, code, code-review, judges, platform, self-learning — that ship the multi-agent orchestration used by loops. |
| Runtime | The execution layer (your machine plus cloud-assisted orchestration) that runs loops inside a sandbox and returns reviewable output. |
| Artifacts | PRDs, implementation plans, critic reviews, judge reports, code maps, learnings, and code review findings — stored per-session and per-org. |
Why Closedloop.ai exists
LLMs are extraordinary at generating content. They are unreliable at being repeatably correct.
Closedloop.ai closes that gap by attaching every loop to a durable artifact, gating every phase with deterministic validation, and grading every output with LLM-as-judge evaluators that score against the original intent.
The result is a system where tickets are the work (not just tracking the work that's done elsewhere), context is captured, the progress of work across a team is immediately visible to everyone, and sprints of scope can land in a few PRs with minimal revisions.
Read this next
Spec-driven, AI-first delivery
The philosophical foundation for how Closedloop.ai works and why.
The closed loop
How phased, artifact-bound gates replace ad-hoc chat workflows.
Commands reference
Every slash command exposed by the Claude Code plugins.
Sessions
The list, detail, and trace of every AI coding run Closedloop.ai observes, plus the filters, status model, cost buckets, and PR/branch attribution behind it.
Agent components
The catalog of agentic building blocks — sub-agents, commands, skills, tools, and more — that each session uses.