AIECP Jira Delivery Orchestrator for Visual Studio Code

AIECP Jira Delivery Orchestrator runs as a self-contained Visual Studio Code extension. Its default embedded mode starts the control plane on a random loopback port, applies SQLite migrations automatically, and stores data in the extension's VS Code global storage directory. No Docker container or separately installed server is required.

The bundled workflow engine remains the single source of truth for execution state. The UI observes and controls that engine rather than simulating delivery steps.

Install the VSIX

To build from source:

corepack pnpm install --frozen-lockfile
corepack pnpm package:vscode
code --install-extension apps/vscode-companion/dist/aiecp-standalone.vsix --force

Alternatively, select Extensions → … → Install from VSIX…, choose aiecp-standalone.vsix, and reload VS Code.

Open the source repository you want the agent to modify, then select AI Delivery in the Activity Bar. The extension starts its embedded control plane, disables mock agents, and uses a real local CLI agent to inspect and change files in the open workspace. If the executable is unavailable or unauthenticated, execution is rejected instead of being reported as successful.

Connect a real coding agent

1. Install and authenticate a CLI agent that supports non-interactive execution. For Codex CLI, confirm that codex --version works and complete codex login if required.
2. Run AI Delivery: Configure Local Workspace Agent from the Command Palette.
3. Select the executable and protocol. The default codex-jsonl protocol runs codex exec with the workspace-write sandbox. Read-only steps are automatically restricted to the read-only sandbox.
4. Select a model (CLI default, gpt-5.6-sol, gpt-5.6-terra, gpt-5.6-luna, or a custom model name) and reasoning effort (default, minimal, low, medium, high, or xhigh). Availability depends on the authenticated Codex account and installed CLI version.
5. Continue only after VS Code displays Real workspace agent connected.

The model/effort button in the cockpit footer opens the same picker as the AI Delivery: Select Model and Reasoning command.

After you select a Jira issue, Requirement Analysis and Planning execute without modifying source files. The workflow pauses at Human Plan Approval. After approval, Code Implementation creates a feature branch when the workspace is a Git repository, changes real files, performs code review, and runs detected build, unit, integration, and API-test commands for Maven, Gradle, pnpm, Yarn, npm, or Make.

Code Implementation is blocked when the Git workspace already contains uncommitted changes. This prevents delivery changes from being mixed with unrelated user work.

Operating modes

• aiecp.mode = embedded — default, self-contained mode with no external process.
• aiecp.mode = external — connects to a separately deployed control plane.
• aiecp.serverUrl — used only in external mode; remote endpoints must use HTTPS.
• aiecp.installPath — optional legacy setting used by the Start Local Server command in external mode.

Reload the VS Code window after changing aiecp.mode.

Connect Jira Cloud and select a task

1. Create an API token in your Atlassian account. Do not use your Jira password. The account must have Browse Projects permission for the issues it needs to read.
2. Run AI Delivery: Configure Jira Cloud from the Command Palette.
3. Enter the Jira site URL, account email, project key, JQL, and API token. The token is stored in VS Code SecretStorage.
4. Select Run Jira task in the top card, Browse Jira & Run in the footer, or run AI Delivery: Select Jira Work Item and Run. Both UI buttons open the same dynamic Jira picker; neither uses internal mock data.
5. The extension loads up to 100 issues with Jira enhanced JQL search. Type in the picker to filter the results.
6. Select a published workflow version. The Delivery Run is pinned to that exact immutable version, so later graph edits cannot alter an active run.

Default JQL:

project = "PROJECT_KEY" AND issuetype in (Epic, Story, Task, Bug) AND resolution = unresolved ORDER BY updated DESC

Set aiecp.jiraJql to filter by the current assignee, sprint, label, or status. See Atlassian API-token authentication and the enhanced JQL search API.

Developer Cockpit

The sidebar lets you:

• search, select, synchronize, and execute real Jira Cloud issues;
• select a local Work Item and start a Delivery Run;
• inspect the workflow version, node progress, agent, changed files, evidence, and coverage;
• approve, request changes, or reject an approval gate;
• open changed files, raw artifacts, and execution logs;
• inspect the latest persisted step output while live CLI, build, and test output streams to the AI Delivery Orchestrator Output channel;
• cancel an active run;
• create a Work Item from selected editor content; and
• receive WebSocket events with automatic reconnect and polling fallback.

The primary action displays Running… during execution and Awaiting approval at a human gate. It is disabled in both states to prevent accidental duplicate Jira runs.

Build a custom workflow graph

1. Select Workflow graph in the Developer Cockpit or run AI Delivery: Open Workflow Designer.
2. The designer opens in VS Code's Simple Browser and connects to the extension's embedded control plane. No Docker container or external server is required.
3. Drag nodes to position them. Drag from a node handle to create an edge. Select an edge and press Delete to disconnect it. Use the node library to add AGENT_TASK, TOOL_TASK, approval, condition, parallel, join, integration, or evidence nodes.
4. Select a node to configure its capability, instructions, context, permissions, tool, approval role, retry policy, timeout, and failure route.
5. Select Validate, Save version, and then Publish. Invalid or unpublished graphs do not appear in the workflow picker for Jira runs.

The selected graph defines the execution order. Nodes run sequentially by default; only branches directly following a PARALLEL node execute concurrently. CODE_GENERATION and CODE_FIX fail when an agent claims success without changing the workspace. Build and test steps complete only when their real commands return exit code 0.

Data and security

Embedded mode binds only to 127.0.0.1 on an operating-system-assigned port. API access requires a random bearer token held only in extension memory. The SQLite database is stored in the extension's private global storage and persists across VS Code restarts. No fixed port is exposed.

The agent executable and argument vector are stored in VS Code settings and passed directly to the child process without a shell. The Jira API token remains in SecretStorage. Codex CLI uses its own local authenticated session; the extension does not copy those credentials into its database.

External mode supports AI Delivery: Configure API Token. That token is also stored in SecretStorage, not settings.json. Do not create Work Items from code selections that contain secrets, credentials, customer data, or personally identifiable information.

Extension Development Host

1. Open this repository in VS Code.
2. Run corepack pnpm install --frozen-lockfile.
3. Select Run AIECP Companion under Run and Debug, then press F5.

You do not need to run pnpm start when aiecp.mode is embedded.

Troubleshooting

• The workspace agent is offline: open a local source folder, confirm the configured executable works, authenticate the CLI, and run AI Delivery: Configure Local Workspace Agent again.
• No Jira issues are listed: verify the Jira project key, JQL, API token, and the account's Browse Projects permission.
• No workflow is available: open Workflow graph and ensure the version has passed Validate → Save version → Publish.
• The workflow is waiting: inspect the approval gate and select Approve, Request changes, or Reject.
• A step failed: select Open logs or Open artifact to inspect the persisted agent, build, or test output.
• Pull request creation failed: confirm that the workspace is a Git repository with a remote and that GitHub CLI gh is installed and authenticated.

Current limitations

• Desktop and remote Node extension hosts are supported; vscode.dev is not supported.
• The extension is currently distributed as a VSIX until its Marketplace publication is complete.
• Creating a real pull request requires a Git repository with a remote and an authenticated GitHub CLI. The workflow fails at the PR step when these prerequisites are missing instead of producing fake evidence.
• Use external mode for a shared multi-user control plane.
• Embedded data belongs to the current VS Code profile and machine and is not synchronized automatically.