APEX Delivery Pipeline

A file-based VS Code extension for tracking AI-assisted delivery workflows inside the editor. It is inspired by the aidlc-extension pattern and now combines three layers: local delivery epics, spec-kit feature workspaces, and Copilot bootstrap operating assets, with optional MCP setup for Jira, Confluence, or custom workflow servers.

If your team standard is VS Code GitHub Copilot Chat rather than Claude Code, this project is the better starting point. It uses workspace-native .github/ Copilot assets and .vscode/mcp.json, and now includes a direct Run Phase with Copilot workflow from the tree view, with @apex follow-up inside Chat.

What It Does

• Adds an APEX Delivery activity-bar view.
• Scans docs/ai-delivery/epics/<KEY>/ for delivery artifacts.
• Tracks phases from phases/<phase>/status.json when present.
• Falls back to artifact completion when no status file exists.
• Creates a sample epic so teams can try the workflow quickly.
• Creates spec-kit workspaces under specs/<feature>/.
• Creates a no-overwrite Copilot bootstrap pack under .github/.
• Starts an integrated flow that creates the spec workspace and Copilot pack together.
• Opens or seeds phase artifacts from bundled templates.
• Configures per-phase Copilot run profiles from a guided UI instead of manual JSON edits.
• Proposes direct artifact updates with a diff preview and explicit apply or reject step.
• Runs the active phase directly against a Copilot language model from the tree view.
• Runs Guided Autopilot across autopilot-enabled phases by reusing the existing phase runner, verification checks, and status transitions in either agent-pause or fully-automatic mode.
• Pauses or resumes Guided Autopilot from the tree when manual intervention or verification review is needed.
• Falls back to GitHub Copilot Chat when direct model access is unavailable.
• Exposes an @apex chat participant for follow-up questions against the most recent phase context.
• Applies role-aware routing so a workspace userRole can shape preferred agent, model family, and auto-submit behavior before explicit phase overrides take precedence.
• Attaches build, test, and lint evidence to phase artifacts automatically when commands are configured or auto-detected.
• Persists a developer trace history with prompts, context files, path selection, fallback reasons, and verification results.
• Records preferred phase role, current user role, and routing mismatch notes inside the developer trace.
• Marks phases as passed and moves the next phase to in progress.
• Shows a dashboard with progress and blocked/review states.
• Auto-refreshes the tree when epic markdown files or phase status.json files change.
• Shows an activity-bar status summary that links back to the dashboard.
• Configures workspace MCP servers through .vscode/mcp.json.
• Supports Atlassian remote MCP for Jira/Confluence and custom stdio MCP servers.

Default Phases

Quick Start

1. Open this folder in VS Code.
2. Run npm install.
3. Run npm run compile.
4. Press F5 and select Run APEX Delivery Extension.
5. In the Extension Development Host, open the APEX Delivery activity-bar view.
6. Click Create Sample Epic.
7. Set apexDelivery.userRole if you want role-aware routing defaults for phases owned by Business Analyst, Developer, Reviewer, or other team roles.
8. On any phase, click Run Phase with Copilot to open GitHub Copilot Chat in agent mode with the current phase artifact already open and attached.
9. Use Run Guided Autopilot when the workflow snapshot marks the current phase as autopilot-enabled and you want the extension to execute phases automatically until verification, manual edits, or an agent handoff requires a resume step.
10. Use Pause Guided Autopilot or Resume Guided Autopilot when a human review step or unsaved artifact change interrupts the flow.
11. Use Configure Phase Profile on a phase when you want per-phase control over auto-submit, agent tag, or preferred direct-execution model family.
12. Use Propose Artifact Update when you want a direct Copilot rewrite of the current artifact with a diff preview and an explicit apply or reject step.
13. Open Developer Trace Panel to inspect run history, prompts, context files, path decisions, role-routing notes, and verification evidence.
14. If you want follow-up discussion scoped to the same phase context, continue in GitHub Copilot Chat with @apex.
15. Click Configure MCP Server if the workflow needs Jira, Confluence, or a custom MCP server.

Configuration

| Setting | Default | Purpose | |---|---|---| | apexDelivery.epicsPath | docs/ai-delivery/epics | Folder scanned for workflow epics | | apexDelivery.specsPath | specs | Folder used for spec-kit feature workspaces | | apexDelivery.ownerName | APEX Owner | Owner written into sample metadata | | apexDelivery.userRole | | Optional current user role used for role-aware routing and trace mismatch warnings | | `apexDelivery.workflowDefinitions` | `{}` | Optional custom workflow definitions snapshotted into new epics at creation time | | `apexDelivery.mcpConfigPath` | `.vscode/mcp.json` | Workspace MCP config file updated by the extension | | `apexDelivery.runPhase.autoSubmit` | `true` | Submits the Copilot Chat prompt immediately instead of opening it as a draft | | `apexDelivery.runPhase.agentTag` | | Optional best-effort routing tag added to Run Phase prompts, for example #dev-orchestrator | | apexDelivery.runPhase.preferredChatAgent | | Optional human-readable preferred GitHub Copilot Chat agent name, for example `Business Analyst` or `Code Reviewer` | | `apexDelivery.runPhase.modelFamily` | `gpt-4.1` | Preferred model family for extension-driven direct execution or direct fallback | | `apexDelivery.runPhase.phaseProfiles` | `{}` | Optional workflow-scoped phase overrides for `autoSubmit`, `agentTag`, `preferredChatAgent`, and `modelFamily` | | `apexDelivery.runPhase.rolePolicies` | `{}` | Optional role-aware defaults keyed by `apexDelivery.userRole`; explicit phase profiles still win | | `apexDelivery.autopilot.defaultRetryLimit` | `1` | Default retry limit for autopilot-enabled phases that do not define a workflow retry limit | | `apexDelivery.autopilot.executionMode` | `agent-pause` | Chooses how Guided Autopilot runs non-interactive phases: `agent-pause` launches Copilot agent mode and pauses for resume; `fully-automatic` skips chat-based paths and only continues when direct model execution completes observably | | `apexDelivery.autopilot.pauseOnManualIntervention` | `true` | Pauses Guided Autopilot when unsaved manual edits are detected in the active artifact | | `apexDelivery.verification.autoAttach` | `true` | Automatically attach build, test, and lint evidence to the current phase artifact | | `apexDelivery.verification.buildCommand` | | Optional explicit build command; otherwise the extension tries compile or build from package.json | | apexDelivery.verification.testCommand | | Optional explicit test command; otherwise the extension tries `test:smoke` or `test` from `package.json` | | `apexDelivery.verification.lintCommand` | | Optional explicit lint command; otherwise the extension tries lint from package.json | | apexDelivery.verification.maxOutputChars | 4000 | Maximum verification output length stored in the artifact and trace history |

Custom workflow example:

{
	"apexDelivery.workflowDefinitions": {
		"investigate-workflow": {
			"name": "Investigate Only",
			"phases": [
				{
					"id": "investigate",
					"name": "Investigate",
					"owner": "Business Analyst",
					"artifact": "DISCOVER.md",
					"gate": "Gate 1",
					"output": "Problem baseline captured",
					"autopilot": {
						"enabled": true,
						"retryLimit": 1,
						"pauseOnManualIntervention": true
					}
				},
				{
					"id": "triage",
					"name": "Triage",
					"owner": "Code Reviewer",
					"artifact": "REVIEW.md",
					"gate": "Gate 2",
					"output": "Risk summary captured"
				},
				{
					"id": "handoff",
					"name": "Handoff",
					"owner": "APEX Owner",
					"artifact": "LEARN.md",
					"gate": "Gate 3",
					"output": "Next action recorded"
				}
			]
		}
	}
}

Workflow-scoped phase profile example:

{
	"apexDelivery.runPhase.phaseProfiles": {
		"default": {
			"design": {
				"preferredChatAgent": "Dev Orchestrator",
				"agentTag": "#dev-orchestrator",
				"autoSubmit": true,
				"modelFamily": "gpt-4.1"
			}
		},
		"investigate-workflow": {
			"triage": {
				"preferredChatAgent": "Code Reviewer",
				"autoSubmit": false,
				"modelFamily": "gpt-4o"
			}
		}
	}
}

Role-aware routing example:

{
	"apexDelivery.userRole": "Developer",
	"apexDelivery.runPhase.rolePolicies": {
		"Developer": {
			"preferredChatAgent": "Dev Orchestrator",
			"agentTag": "#dev-orchestrator",
			"autoSubmit": false,
			"modelFamily": "gpt-4o"
		},
		"Business Analyst": {
			"preferredChatAgent": "Business Analyst",
			"autoSubmit": true,
			"modelFamily": "gpt-4.1"
		}
	}
}

You can set these values either in settings JSON or through APEX Delivery: Configure Phase Profile from the command palette or phase context menu.

When you create a new epic, the extension resolves the selected workflow and stores a snapshot in epic metadata. Later edits to workspace workflow definitions do not silently rewrite existing epics.

Guided Autopilot is phase-scoped. It only continues automatically when the snapshotted workflow marks the current phase with autopilot.enabled: true.

apexDelivery.autopilot.executionMode controls how non-interactive phases run:

• agent-pause preserves the current Copilot-agent-first behavior. The extension launches agent mode when available and pauses because the public chat API does not expose agent completion.
• fully-automatic skips agent-mode and ask-mode chat handoffs. It continues only when the extension can observe a direct model completion itself; if direct execution is unavailable, Guided Autopilot pauses with a reason instead of opening chat automatically.

In both modes, verification failure beyond the configured retry limit or unsaved manual edits still pause the flow and record the reason in workspace state plus the developer trace.

Role-aware routing is advisory. The extension can merge role defaults into the run preference resolution and show mismatches between apexDelivery.userRole and the phase owner, but explicit workflow phase profiles still win and GitHub Copilot Chat agent enforcement remains limited by the public chat API surface.

preferredChatAgent is a developer-facing preference and trace hint. The extension can show which agent a phase wants and carry that into the trace panel and prompt, but the public GitHub Copilot Chat API still does not expose the currently selected custom agent reliably enough for hard enforcement or automatic switching.

modelFamily is applied when the extension drives a direct model request itself. GitHub Copilot Chat still uses the model currently selected in the Chat UI because the public workbench.action.chat.open command does not expose model locking.

Integrated Spec Kit + Copilot Bootstrap Flow

Run APEX Delivery: Start Integrated Delivery Flow when you want the extension to create the practical bridge between aidlc-extension, spec-kit, and copilot-bootstrap.

It creates:

specs/<feature>/
	spec.md
	plan.md
	tasks.md
	research.md
	data-model.md
	contracts/README.md
	quickstart.md

.github/
	copilot-instructions.md
	prompts/apex-delivery.prompt.md
	instructions/apex-delivery.instructions.md
	agents/apex-delivery-orchestrator.agent.md
	skills/apex-delivery/SKILL.md

The Copilot pack is no-overwrite by default. Existing .github files are skipped rather than replaced, so a team can review and merge generated assets deliberately.

Use the individual commands when you want only one layer:

Use With GitHub Copilot Chat

Yes. This extension is designed to work well with GitHub Copilot Chat.

Recommended flow:

1. Create or open an epic under docs/ai-delivery/epics/<KEY>/.
2. Generate a spec workspace or Copilot pack when needed.
3. In the tree view, choose Run Phase with Copilot on the active phase.
4. The extension opens GitHub Copilot Chat in agent mode, attaches the phase files, and submits a prompt that tells Copilot to update the existing artifact.
5. Set apexDelivery.userRole and optional apexDelivery.runPhase.rolePolicies if you want role-aware defaults before explicit phase profiles take over.
6. Use Configure Phase Profile if you want different auto-submit, agent-tag, or preferred direct-model behavior per phase without editing settings JSON manually.
7. Use Run Guided Autopilot for workflow phases that are marked as autopilot-enabled when you want the extension to run and advance without manually triggering each phase. Keep agent-pause if you want Copilot agent edits plus manual resume.
8. Use Run Fully Automatic Autopilot when you want the same flow to force fully-automatic mode for that run without changing apexDelivery.autopilot.executionMode in workspace settings.
9. Each phase run updates the artifact with the latest verification evidence when the extension can detect or is given build, test, and lint commands.
10. Use Propose Artifact Update if you want a developer-controlled diff-and-apply flow instead of going straight to Chat.
11. Inspect Open Developer Trace Panel when you need to understand what prompt was used, which files were in context, which path was chosen, why a fallback happened, and whether the current user role matched the phase role.
12. If you want a scoped follow-up inside the APEX participant, continue with @apex.
13. Mark the phase passed when the artifact satisfies the gate, or let Guided Autopilot advance it automatically when policy and verification allow; resume it manually after agent-mode runs complete.

This is the main difference from aidlc-extension: aidlc is optimized around Claude Code slash-command orchestration and .claude/settings.json, while this project is optimized around VS Code-native Copilot assets and workspace MCP configuration.

MCP Setup

Run APEX Delivery: Configure MCP Server from the command palette or the view title.

Available options:

The extension is append-only for MCP servers: it does not overwrite an existing server with the same name. This keeps local or team-specific MCP configuration safe.

Example generated config:

{
	"servers": {
		"atlassian": {
			"type": "stdio",
			"command": "npx",
			"args": ["-y", "mcp-remote", "https://your-atlassian-mcp.example.com/mcp"]
		},
		"custom-workflow": {
			"type": "stdio",
			"command": "npx",
			"args": ["-y", "your-mcp-package"]
		}
	}
}

Marketplace Publishing

This project includes VS Code Marketplace packaging scripts and is configured for the vndkubi Marketplace publisher id.

Before publishing:

1. Confirm that publisher: "vndkubi" in package.json matches the Visual Studio Marketplace publisher account you will use.
2. Confirm that the vndkubi publisher id exists in the Visual Studio Marketplace publisher portal.
3. Run npm install.
4. Run npm run compile.
5. Run npm run package to create a .vsix.
6. Install the .vsix locally and smoke-test the extension.
7. Run npx vsce login vndkubi or pass a Marketplace PAT when publishing.
8. Run npm run publish.

Full SOP: docs/MARKETPLACE-PUBLISHING-SOP.md.

Copilot Bootstrap And Spec Kit Relationship

copilot-bootstrap already carries the spec-driven workflow shape: specify-feature, plan-implementation, generate-tasks, implement-feature, and prompts that expect specs/<feature>/ artifacts. That means the bootstrap layer can teach Copilot how to work with spec-kit-style artifacts.

This extension adds the missing VS Code product surface: commands that create the local specs/<feature>/ workspace, generate a minimal no-overwrite Copilot pack, and connect optional MCP context. If a repository already ran a full copilot-bootstrap, use this extension's Copilot pack command carefully; it skips existing files by default so it will not replace the richer bootstrapped assets.

Positioning Versus aidlc-extension

aidlc-extension is still the stronger choice when you specifically want Claude Code slash commands, .claude/settings.json management, and a heavier MCP-driven SDLC orchestration model.

This project fits better when you want:

• GitHub Copilot Chat as the primary AI surface inside VS Code.
• .github/ Copilot operating assets instead of Claude-specific runtime assets.
• Workspace-native .vscode/mcp.json for MCP configuration.
• A lighter cockpit that connects epics, spec-kit workspaces, review flow, and ROI-oriented delivery notes.

Compared with the earlier version of this extension, it is now closer to the aidlc-extension quality bar in the places that matter most for Copilot users: a tree action that opens full Copilot Chat agent mode against the current artifact, @apex follow-up inside Chat, auto-refreshing pipeline state, a clearer dashboard, and a less provisional product surface.

License

MIT