SDLC Pipeline

Track SDLC pipeline progress for software epics directly in VS Code. Visualize every phase — from planning to release — with agent ownership and real-time status.

Features

• Epic Tree View — sidebar panel showing all epics and their pipeline phases
• Pipeline Dashboard — visual dashboard with progress bars, phase dots, and hover details
• Phase Status — each phase shows: pending, in-progress, done, awaiting review, rejected, or stale
• Agent Ownership — see which agent (PO, Tech Lead, QA, Dev, RM, SRE, Archivist) owns each phase
• Pipeline Settings — enable/disable phases per epic via a settings panel
• Isolated Phase Sessions — open a dedicated editor session scoped to a single phase
• Inline Run / Re-run / Review actions — drive each phase from the tree without leaving the sidebar
• Example Project loader — bootstrap a fully-wired demo workspace (sample epic + agents + skills + schemas) in one click

How It Works

The extension scans your docs/sdlc/epics/ folder for epic artifacts (PRD, Tech Design, Test Plan, etc.) and determines pipeline status from per-phase status.json files written by the orchestrator (or from file presence for legacy epics).

Pipeline Phases

Getting Started

1. Install the extension (VS Code Marketplace or Open VSX).
2. Open a workspace folder.
3. Click the SDLC Pipeline icon in the activity bar.
4. The sidebar tree opens. From here:
   • Workspace already has docs/sdlc/epics/<KEY>/ folders → epics show up immediately.
   • Empty workspace → the tree shows a welcome card with three options:
     • Load Example Project (🚀 also pinned to the navbar) — clones a demo workspace at ~/aidlc-example, wires the MCP server, and opens it in a new window. A confirm dialog runs first so you can cancel an accidental click.
     • Open Pipeline Settings — point at an MCP package + platform (the extension never auto-installs anything).
     • Select Epics Folder — pick an existing folder elsewhere on disk.

The extension never seeds sample epics into your workspace automatically. If you want a starter epic, click Load Example Project.

Configuration

Use your own pipeline

If you have a private fork (e.g. company-specific SDLC with core-business docs), point the extension at it:

1. Open the Pipeline Settings panel (sidebar → … → Pipeline Settings, or Command Palette → SDLC: Pipeline Settings).
2. Scroll to MCP Pipeline Source at the top.
3. Enter your package spec — e.g.:
   • github:yourcompany/cf-sdlc-pipeline (GitHub repo, public or private with SSH auth)
   • @yourcompany/sdlc-pipeline (once published to npm)
4. Pick the platform that matches your project.
5. Click Apply & Reload MCP — the extension appends mcpServers.<name> to .claude/settings.json (never overwrites an existing entry) and Claude Code loads your pipeline on the next tool call.

To clear it, click Clear & Disable Auto-Configure.

How Sync Works (MCP Server)

When Claude Code boots the MCP server (triggered on the first MCP tool invocation per session), it runs syncWorkspace() against your project root. The reference server is published from hueanmy/aidlc-pipeline — pulled via npx -y github:hueanmy/aidlc-pipeline (or the npm package when available).

What gets synced on every boot

Into your project (cwd = workspace root):

<workspace>/
├── .claude/
│   ├── settings.json            # appended by extension when autoConfigureMcp=true
│   ├── skills/
│   │   ├── _gate-check.md
│   │   ├── prd/SKILL.md         # merged: generic + platform overlay
│   │   ├── tech-design/SKILL.md
│   │   ├── test-plan/SKILL.md
│   │   └── ... (19 skills total)
│   └── agents/
│       ├── po.md                 # merged: generic + platform overlay
│       ├── tech-lead.md
│       ├── qa.md
│       └── ... (7 agents total)
└── docs/
    └── sdlc/
        ├── workflow/README.md    # scaffold (only if missing)
        └── epics/                # your real epics live here

The extension itself does not create any epic folders. The MCP package may scaffold a workflow/README.md on first run; epics are yours to author (or fetched from the example project).

Layered merge (platform overrides)

Content is merged at boot using a simple overlay:

1. Generic — from the pipeline package root (skills/, agents/, templates/)
2. Platform — from platforms/<SDLC_PLATFORM>/<kind>/ (e.g., platforms/mobile/agents/po.md)
3. Project (optional) — from projects/<SDLC_PROJECT>/<kind>/

Later layers override sections of the earlier ones by matching ## headings. {{PLACEHOLDER}} tokens get substituted from the project's config.json if provided.

Timing

Claude Code boots MCP servers lazily — the server does not start until a slash command triggers an MCP tool call. Practical implications:

• First Claude Code session in a fresh workspace: slash commands may not appear immediately. Fire any / command (e.g. /dashboard) once to trigger boot → reload or reopen the prompt → commands appear.
• Subsequent sessions: sync runs again at first tool call, refreshing any out-of-date content.

If you want slash commands to work immediately without waiting for lazy boot, run once manually in the workspace:

npx -y github:hueanmy/aidlc-pipeline
# Ctrl+C after a second — sync has already run during startup.

Use with Claude Code

Once you've set cfPipeline.mcpPackage and enabled autoConfigureMcp (or written .claude/settings.json by hand), drive each phase with slash commands. Pass the Jira epic key (e.g. ABC-123) and the pipeline pulls context from Jira automatically (requires the Atlassian MCP):

Use /dashboard to see pipeline status across all epics, or /pipeline EPIC-KEY-123 to run the full SDLC sequentially.

You can also drive phases without leaving the tree view — every phase row exposes inline actions (▶ Run / 🔄 Re-run / 💬 Update Feedback / 🔔 Review Gate) that copy the right slash command to your clipboard and focus the Claude Code chat panel.

Walkthrough — start an epic from scratch

Assume your team created a new Jira epic ABC-123 with a linked Figma frame.

1. Scaffold the epic

/epic ABC-123

• Claude calls the Jira MCP to fetch the epic's title, description, children, document links, and Figma to pull design context.
• Claude creates docs/sdlc/epics/ABC-123/ with the epic template pre-filled (title, scope, acceptance criteria skeleton).
• The VS Code sidebar refreshes — ABC-123 appears with all phases in pending.

2. Draft the PRD

/prd ABC-123

• The Product Owner agent reads the Jira issue + any Figma link attached to it (via the Figma MCP) to extract screens, user flows, and design tokens.
• Writes docs/sdlc/epics/ABC-123/PRD.md with user stories, acceptance criteria, and analytics events.
• Plan phase flips to done; Design becomes in-progress.

3. Technical design

/tech-design ABC-123

• The Tech Lead agent reads the PRD + explores the current codebase to propose architecture, API contracts, file impact, and DI plan.
• Writes TECH-DESIGN.md.

4. Test plan

/test-plan ABC-123

• The QA Engineer agent maps PRD acceptance criteria to unit/UI/integration test cases.
• Writes TEST-PLAN.md.

5. Implement on a feature branch

Out of band — developer cuts feature/ABC-123-... and builds the code. Claude can assist as the Developer agent, constrained to the design + test plan.

6. Review the PR

/review <pr-url>

• Validates the PR diff against PRD + Tech Design + Test Plan and flags gaps.

7. Execute test → Release → Monitor

/execute-test ABC-123  # QA generates test script
/release               # RM prepares release notes + git tag
/deploy                # RM deploys to test/prod
/monitor               # SRE pulls crash + analytics for the first 24–48h

8. Close the loop

/doc-sync ABC-123

• The Archivist agent diffs planned docs vs shipped code and reverse-syncs the PRD/Tech Design so they reflect what actually shipped.

At any point, /dashboard shows the pipeline status for every epic, or you can run /pipeline ABC-123 to walk through phases 1 → 8 sequentially without invoking each skill by hand.

Commands

All commands are available via Cmd+Shift+P (or Ctrl+Shift+P):

Troubleshooting

Slash commands don't appear in Claude Code

Cause: the MCP server hasn't booted yet. Claude Code starts MCP servers lazily — only on the first MCP tool call.

Fix:

1. Type any / command (e.g. /list-skills) to trigger an MCP tool call. The server will boot and sync in the background.
2. Reload the Claude Code prompt / start a new chat → slash commands now discoverable.
3. Alternatively, warm up manually in your workspace root: npx -y github:hueanmy/aidlc-pipeline (kill after 1–2 seconds; the sync runs during startup).

.claude/skills/ or .claude/agents/ is empty after activation

Cause: the MCP server either hasn't run yet, or npx failed silently — or cfPipeline.autoConfigureMcp is false (the default), so no MCP entry was written at all.

Fix:

1. Set cfPipeline.mcpPackage to your package spec, then flip cfPipeline.autoConfigureMcp to true.
2. Reload the window — extension appends mcpServers.<name> to .claude/settings.json.
3. Verify the entry has command: "npx" and args: ["-y", "<your package>"].
4. If the MCP entry exists but skills/agents are still empty, run it manually:

   cd /path/to/your/project
   npx -y github:hueanmy/aidlc-pipeline
   

   If this works, .claude/skills/ and .claude/agents/ will populate. Kill with Ctrl+C.
5. If npx errors: rm -rf ~/.npm/_npx to clear the package cache, retry.

Extension didn't write .claude/settings.json on activation

Cause: by design — cfPipeline.autoConfigureMcp defaults to false. The extension only writes the file when you explicitly enable auto-config (or click Apply & Reload MCP in Pipeline Settings).

Fix:

1. Open Pipeline Settings → MCP Pipeline Source.
2. Enter your mcpPackage, pick your platform, and click Apply & Reload MCP.
3. The extension appends the MCP entry without touching any other server already in settings.json.

Slash commands use old or outdated content

Cause: npx caches the GitHub tarball; new commits to your pipeline package aren't picked up automatically.

Fix:

rm -rf ~/.npm/_npx

Then reload Claude Code. The next MCP boot will re-download the latest main from GitHub.

"MCP server sdlc already configured, leaving as-is" in output

This is expected — the extension is append-only and never overwrites an existing mcpServers.<name> entry. To replace it:

1. Open .claude/settings.json.
2. Delete the mcpServers.<name> entry.
3. Reload VS Code window — extension re-appends with current settings.

Epic folder doesn't show up in the sidebar

Cause: the extension only treats folders matching [A-Z][A-Z0-9]*-\d+ (e.g. ABC-123, EPIC-2100) as epics.

Fix: rename the folder to match the pattern. Folders like my-feature/ or abc-123/ (lowercase) are ignored.

Tree view stays empty after I add an epic folder

The file watcher updates on *.md, pipeline.json, and status.json changes. If you add only an empty folder (no files), the tree won't refresh. Either:

1. Add at least one .md, pipeline.json, or status.json to the new epic folder, or
2. Click the Refresh button in the tree's title bar.

Platform-specific agents not loading

Cause: only platforms/mobile/ has overrides shipped today in the reference pipeline. Other platforms (web, backend, desktop, generic) fall back to generic content — this is normal.

Fix: if you need platform-specific overlays, contribute them upstream to hueanmy/aidlc-pipeline under platforms/<name>/.

Extension activates but no tree view appears

Cause: a workspace folder must be open (file-only mode is not enough).

Fix: File → Open Folder → select your project root → reload.

Tree view is empty and I don't want to start from scratch

Click Load Example Project in the welcome card or the rocket icon in the tree's title bar. The extension confirms before cloning, then opens the demo at ~/aidlc-example in a new window.

Where to find logs

• Extension side: View → Output → "SDLC Pipeline" channel
• Claude Code / MCP side: Claude Code output panel, or run the MCP manually with npx -y github:hueanmy/aidlc-pipeline in a terminal — all server logs go to stderr.

Requirements

• VS Code 1.85.0+ (or compatible: VSCodium, Cursor, Windsurf)
• A workspace folder (not single-file mode)
• Claude Code extension/app for slash command support
• Network access for npx to fetch the MCP server from GitHub
• Optional: Atlassian MCP for Jira integration, Figma MCP for design context

License

MIT