A developer based in Hanoi has built AIDLC, an open-source project that transforms Claude Code from a chat box into a structured software development pipeline with agents, skills, checkpoints, and token monitoring.
A developer based in Hanoi has built AIDLC, an open-source project that transforms Claude Code from a chat box into a structured software development pipeline with agents, skills, checkpoints, and token monitoring.
AIDLC is a workflow orchestration layer for Claude Code. Its tagline: "See what AI is building. Drive Claude through any pipeline you declare -- and track every run, step, and token." Instead of handing Claude one large prompt and waiting blindly for a result, you declare agents, skills, and pipelines in a single workspace.yaml file. Claude remains the execution engine; AIDLC provides the control panel on top.
The project ships as both a VS Code extension and a CLI, both powered by the same TypeScript engine and shared run state. You can edit your pipeline configuration in the editor or the terminal, and both read from the same source of truth. The architecture supports approval gates, rejection with feedback, reruns, artifact annotation, epic memory, and session insights. When a step produces a document, you can open it in a browser, click directly on a paragraph, and leave threaded feedback. Claude updates the Markdown, and revision history is preserved.
You declare your agents, skills, and pipeline stages in workspace.yaml. When you run a pipeline, AIDLC's DefaultRunner spawns Claude CLI to execute each stage. Each stage has its own agent assignment and checkpoint. If a step fails or is rejected, the pipeline pauses, you provide feedback, and the step reruns. Downstream steps that depend on a rejected step are automatically reset to pending. The monitor panel pulls session insights directly from Claude Code transcripts and token usage data, giving you a real-time view of what is happening across the pipeline.
The project also includes an annotation system (bundled in version 2.4) that lets you leave persistent, threaded comments on rendered HTML artifacts, with full revision history tracked alongside.
AIDLC is a young project with modest adoption so far. It depends tightly on Claude CLI -- if you are not already a Claude Code user, this adds little value. The project is not a multi-model orchestration platform; it is built specifically around Claude. Before using it with sensitive codebases, you should audit transcript read permissions and any plugins you install. The maintainer's GitHub profile publicly lists Hanoi as their location, which is a notable detail for the Vietnamese developer community, but it does not change the fact that this is an early-stage tool that has not yet been proven at production scale.
AIDLC is for teams already using Claude Code who want to bring structure, observability, and approval gates to their AI-assisted development process. If your workflow involves large epics where you lose track of which step the AI is on, what it has produced, and what needs review, this fills that gap. It is not for someone looking for a general-purpose AI orchestration platform or a tool that works across multiple model providers.
If you want Claude to work through a pipeline instead of a chat box, AIDLC is a project worth watching. The repository is MIT-licensed and available on GitHub and the VS Code Marketplace.