AIDLC Native

Drive Claude through the six stages of the AI-Native SDLC Playbook — Plan → Design → Build → Test → Deploy → Maintain — from VS Code or the terminal, and see every step, artifact and token on the way.

A fork. AIDLC Native is built from delete101020/ai-native-sdlc, a fork of aidlc-io/aidlc by hueanmy (hueanmy.aidlc). It is not affiliated with or endorsed by the upstream author. Its commands and settings live under aidlcNative.*, so the two extensions can be installed side by side.

You declare agents, skills and pipelines once in .aidlc/workspace.yaml. Each piece of work is an epic that owns its pipeline: Claude runs a step, writes the step's artifact, and the pipeline waits at the review gates you set. The extension and the aidlc CLI read and write the same files, so you can start an epic in one and finish it in the other.

The six stages

The AI-Native SDLC workflow follows the AI-Native SDLC Playbook. Each stage is run by its own agent and leaves a file under docs/epics/<epic>/artifacts/:

Verify and review are run by fresh-context agents, not by the session that wrote the code. Stage 6 closes the loop: a production signal becomes an incident epic, and the fix it needs opens as a new epic at stage 1 with its intent.md already written.

Not every change needs all six. Start Epic reads a one-line brief and picks a recipe, which you can override:

Getting started

1. Make sure the Claude Code CLI is installed and signed in (claude --version).
2. Open a project folder and click the AIDLC icon in the activity bar.
3. Apply the AI-Native SDLC workflow from the sidebar's workflow list, or run aidlc preset apply ai-native in a terminal. This writes the six agents, their skills, the ai-native-full pipeline and the native-* recipes into .aidlc/workspace.yaml.
4. Start Epic, describe the work in a line, and accept or change the suggested recipe.
5. Work each step: run its slash command in Claude (for example /ai-native-full-intent EPIC-1), read the artifact, then Approve, Reject with feedback, or Rerun. Or use Run to completion to execute the remaining steps back to back, pausing at the gates you keep.

The same epic from a terminal:

npm install -g @delete101020/aidlc
aidlc preset apply ai-native
aidlc epic start EPIC-1 --brief "fix the login redirect loop"   # → recipe native-fix
aidlc run exec EPIC-1                                            # runs steps, stops at review gates

Just exploring? AIDLC Native: Load Demo Project drops a finished example (the classic workflow and six sample epics) into the open folder.

Features

• Epics that own their pipeline — an epic's assembled pipeline lives in docs/epics/<id>/pipeline.yaml, its depth (strict_mode) in its own state.json, and its step list and gates stay editable while it runs.
• Review gates — human_review pauses for you; auto_review runs a validator headlessly. Reject sends your feedback back to the step that produced the artifact and resets what came after.
• Run to completion — one button, the same engine as aidlc run exec. Cancelling takes effect at the next step boundary, so a half-written artifact never passes a gate.
• Incident loop — Report a Signal turns a production signal into an incident epic; aidlc maintain --signal does the same from a webhook or a cron job.
• Artifact review in the browser — open any artifact in the bundled annotron, point and click your feedback, and Claude applies it back to the Markdown with an attributed revision history.
• Epic memory — a short per-epic digest of decisions and constraints that any agent can load cheaply.
• Code graph for Claude — an ast-graph (default) or CodeGraph MCP server, registered for the project so Claude answers structural questions without grepping. CodeGraph suits large repos.
• AIDLC Monitor — token usage and cost, session insights from Claude Code transcripts, and live agent sessions through agents-observe.
• One window, one Claude account — pin a Claude config dir per workspace (aidlcNative.claude.configDir) or switch with AIDLC Native: Switch Claude Account.
• Workspace Builder — a visual editor for agents, skills, pipelines and recipes.
• Requirements import — pull requirements from Jira, GitHub Issues, Linear, Redmine or a file into requirements.md (Builder → Analyze, or aidlc analyze).

The extension also ships the upstream workflows: the classic AIDLC SDLC pipeline (Plan → (Design ∥ Test Plan) → Implement ∥ Generate Test Cases → Execute Test), Spec Kit, and the single-agent code-review and release-notes presets.

Network and privacy

AIDLC Native sends no telemetry of its own. Two things reach the network:

• Claude — every agent run shells out to your installed claude CLI, under your account.
• The code-graph binary — on first use the extension downloads the ast-graph or CodeGraph release for your platform from GitHub, checks it against a pinned SHA-256, and keeps it in the extension's storage. CodeGraph's own telemetry stays off unless you enable aidlcNative.astGraph.codegraphTelemetry. Set aidlcNative.astGraph.enabled to false to skip the download altogether.

Commands

Everything is in the Command Palette under AIDLC Native. The ones you will use most:

Requirements

• VS Code 1.85+, or an editor that installs from Open VSX (Antigravity, Cursor, VSCodium, Windsurf)
• The Claude Code CLI on PATH, signed in
• A folder open (single-file mode is not supported)
• For the CLI: Node.js 18+

Upgrading from a build before 4.0? Settings under aidlc.* are copied to aidlcNative.* the first time the extension starts. Keybindings are yours to update: change aidlc. to aidlcNative. in keybindings.json.

Credit

Built on aidlc-io/aidlc by hueanmy. If it saves you time, sponsor the original author ❤️.

License

MIT — the original copyright line is kept; this fork's changes are added under the same terms.