AIDLC-VIEWER

A companion tool for AI-DLC, the AWS AI-Driven Development Lifecycle workflow.

AI-DLC has your agent work through a lifecycle, producing Markdown artifacts as it goes. This extension does two things for you while that happens.

It sets AI-DLC up in one click. Starting AI-DLC otherwise means cloning a repository and copying rule files into the right place for your editor. Press one button, answer which generation you want, and it is set up.

It shows you where the workflow actually is. The furthest stage reached, which file needs you next, and each artifact opened in a readable split view, so you are not digging through folders to find out what your agent has been doing.

Works in Kiro and in VS Code and its forks. This extension's v2 setup targets Kiro IDE or Kiro CLI, which are the two it has verified against a real run; upstream also ships definitions for other agents, which you can install yourself. v1 works with Kiro, Claude Code and your editor's own convention.

Getting started

1. Install the extension and click the AIDLC icon in the activity bar.
2. Choose the project folder to watch. Everything below acts on that folder, so this comes first.
3. Press Set up AI-DLC Rules and pick a generation. AI-DLC comes in two: v1 is rule files your agent reads, v2 is an engine that drives the workflow itself. If your project already has AI-DLC, the extension detects which one and you can skip this step.
4. Start the workflow. On v1, tell your agent "Using AI-DLC, build ...". On v2, type /aidlc — it installs a conductor agent plus a set of role agents, so /aidlc starts or resumes a workflow and /aidlc-mvp and the rest are there when you want a specific one.

New artifacts appear in the tree as they are created.

Two things to know about v2 before you start. Its engine is a program installed on your machine, so the first v2 setup downloads about 110MB and takes a minute; every project after that is seconds. And Kiro reads /aidlc from the workspace root, so if you set up a folder that is not the root of your window, open that folder and /aidlc will be there. AI-DLC v2: Check setup in the Command Palette tells you which of these is missing.

Setup does not overwrite anything. Existing files are detected and left alone.

What it shows you

Where the workflow is

The top of the tree names the furthest stage reached and how many stages are done. Opening any artifact adds a strip across the top of the viewer, so your position stays visible while you read. Click it to see the whole stage list grouped by phase.

There is no percentage, on purpose. AI-DLC adapts to the work: most stages are conditional, and Construction repeats once per unit of work with a unit count that does not exist until units have been generated. Any denominator would be guesswork, so reached stages are counted instead, and stages that repeat show their own progress like 2/3.

A stage the workflow moved past without producing anything is shown as skipped rather than left blank, and one your scope never included says so. Both are normal.

Which file needs you

New artifacts are marked NEW until you open them, then READ, with counts rolled up to their folders. Folders containing new files open themselves, so you are not hunting for what just appeared. Changes are picked up whether they came from your agent, from git, or from anything else.

The artifact itself

Each .md file opens with the source on the left and a styled preview on the right. Mermaid diagrams render inline. You can edit and save back to the file, resize or reset the split, sync the scrolling, and zoom the preview.

Supervision, on AI-DLC v2

In v2 your agent reads the artifacts and you approve a summary, and the workflow continues whether or not you read anything yourself. A summary is useful precisely because it leaves things out, so this panel shows you what it left out:

• which stage is waiting on you, because a stage at its approval gate is not slow, it is stopped, and the chat said so once and scrolled away
• what the engine's own checks flagged, grouped per check, with a link to read each finding
• the agent's record beside the engine's check of it, per stage, never reconciled — the two answer slightly different questions, so the judgement stays yours
• the units of work and what depends on what

The panel appears only on v2 projects. Nothing about v1 changes.

Toolbar

Two more commands are in the Command Palette for v2 projects. AI-DLC v2: Check setup reports what is present and what is missing when /aidlc does not appear. AI-DLC v2: Remove takes back only what the extension installed, and refuses if it has no record of installing anything.

Notes

AI-DLC v2's engine is installed on your machine, and this extension installs it for you. It runs the installer AWS publishes with each release, which verifies what it downloads and puts everything under your home directory, so no administrator rights are needed. If that cannot run where you are, setup hands you the same command to run yourself rather than leaving you a project whose gates never fire. v1 needs nothing and works offline.

v1 is installed from a snapshot inside the extension. v2 comes from the latest published release of awslabs/aidlc-workflows, and the version you have is recorded in the project so you can always tell which engine it is.

Nothing is read out of your Markdown. Progress comes from the project's structure, so an artifact in an unexpected format cannot put your position wrong, and anything the extension does not recognise is reported rather than guessed at.

What's New

0.0.25

• v2 setup works again, and now installs the engine. AI-DLC v2 changed shape: its engine used to be files copied into your project and is now a program installed on your machine, which the project's hooks call by name. Copying files alone would have produced a project that looked complete and never advanced. Setup downloads the engine from the official release, checks every file against the release's published checksums, and lets the engine set your project up itself — one button still does the whole thing. The same setup runs on macOS, Windows and Linux.
• bun is no longer required. The engine reports that itself now. Setup used to stop without bun, which would have blocked people over something v2 does not use.
• Your files are the engine's to keep, not ours to guess at. Setting up over a project the workflow has already grown goes through the engine's own configuration, which reports what it left alone. Verified against a real project: accumulated rules, generated artifacts and hand-written steering all survive, and so do v1's rules where the two generations share a folder.
• Removal works on projects the engine installed, by reading the engine's own record of what it wrote rather than requiring one of ours.
• v2 is recognised in more projects. Detection knew only the older name for the agent file, so a project carrying only the newer one could be missed. Both count now, and Check setup no longer reports a correctly installed project as missing a file.
• The number on the activity-bar icon matches the tree again. Files the engine keeps for itself were being counted as new artifacts you had to read, so the badge said more than the folder showed and nothing you could open would clear it.

0.0.23

• v2 setup works again. Upstream renamed its v2 branch to main, and every v2 install failed with a message about a missing commit, which was not what had happened. The extension follows the default branch now, and a 404 says whether a branch or a commit is missing.
• Kiro CLI setup works again. Upstream also renamed that harness's directory, and the old path extracted nothing, which read as a broken download rather than a wrong path.
• Verified against the current engine end to end: install, reinstall over a project the workflow has grown, and removal. Your accumulated rules and the workflow's own files are still left alone.

0.0.21

• The panel tells you when a stage is waiting on your approval. In v2 a stage at its gate is stopped, not slow, and the chat says so once and scrolls away. It says whether the engine recorded the wait itself or the extension concluded it from the stage's own state, so the two are not presented as equally certain.
• v2 installs from the current head of the upstream branch, and records the commit it resolved to, so a project can always say which engine it has.
• Three states instead of one word. A stage your scope never included reads not in scope, a conditional stage that may decline to run reads if needed, and one simply awaiting its turn reads not started. All three used to read skipped, which claimed the run had passed stages it had not reached.
• Setup no longer asks twice. Choosing a generation and a harness were the decisions; the confirmation after them was a dialog with one button. It now reports what it installed instead.
• The generation picker describes what each generation is, rather than where its files come from.
• Fixes: changes to v2's own data now refresh the view, the progress notification no longer lingers after setup finishes, and the journey strip appears on v2 projects.

0.0.20

AI-DLC v2 support. The extension detects which generation your project uses and shows the matching view. Nothing about v1 changes.

• Supervision panel for v2 — which stage is waiting on you, what the engine's checks flagged, the units of work, and the agent's own record placed beside the engine's check of it.
• The journey strip works on v2, including per-unit progress and the marker for the file you have open. Same strip, nothing new to learn.
• One setup button that asks which generation, with v1 listed first.
• v2 setup tells you it needs bun instead of producing a project whose gates silently never fire.
• Removal takes back only what was installed, never the workflow's own files, and preserves v1's rules where the two generations share a directory.
• Guide updated for both generations, in English and Korean.

0.0.19

• A journey strip in the viewer — the furthest stage reached and a rail with one segment per stage, across the top of the editor and preview, so your position stays visible while you read.
• The progress percentage is gone. It could not be computed honestly and reported 0% on real projects full of artifacts. Reached-stage counts replaced it.
• Skipped stages are shown as skipped rather than as gaps, which is normal on greenfield projects and whenever the workflow decides a design stage adds nothing.
• Progress comes from the artifact tree, not from file contents, so a state file in an unexpected format can no longer put your position wrong.
• Unrecognised artifacts are listed rather than guessed into a stage.

0.0.18

• Persona Proxy: position-bias mitigation — persona recommendations were skewing toward the first option. A persona must now evaluate every option on its content and say why at least one it did not choose was rejected, and a recommendation must hold regardless of the order options were presented in.

0.0.17

• Persona Proxy: Deliberation Points — each recommendation now includes 2–4 points to weigh before deciding, independent of whether you follow it, and at least one must say when a different option would be the better choice.
• Unambiguous combined recommendations — a persona recommending a combination must say when each part applies.

0.0.16

• Persona Proxy — when your team is missing a professional role, an AI persona adds an advisory recommendation to question files in that role's domain. Every [Answer]: is still left for a human, so approval gates are never bypassed.
• 7 personas selectable at opt-in: Product Owner, Software Architect, Backend, Frontend, DevOps, Security and QA Engineer. Each comments only on stages in its own domain.
• Answers are logged as persona-aligned or human-override, so recommendation quality can be reviewed later.

0.0.15

• State-file-based progress — when the workflow keeps its own state file, progress is read from it and the in-progress step is named.
• Auto-expand new artifacts, so freshly generated files are visible without digging.
• Workspace auto-detect — if the open workspace already contains AI-DLC artifacts, the extension offers to watch it.
• Open in Text Editor from the right-click menu, for a regular editor instead of the split viewer.
• Unread badge on the activity bar icon.
• Viewer guide rewritten, in English and Korean, and it opens automatically the first time.
• Webview hardening — a content security policy, sanitised link and image URLs, and Mermaid restricted to its safe mode.

0.0.14

• One-click AI-DLC rule setup — install the workflow rules for your agent from the toolbar, skipping the manual download and copy steps.
• Rules are bundled with the extension, so no network access is needed.
• Generates the real per-platform files for Kiro, Claude Code, and your current editor's own convention.
• Idempotent — existing files are detected and never overwritten.