Weave for Boomi

Read, compare, and actually run your Boomi processes — without leaving the editor.

Weave indexes your Boomi components locally, draws the process canvas VS Code can't, diffs any two versions shape by shape, and executes a process for real on a local Atom with assertions you keep in git.

Everything except the Boomi API import and AI answers works offline. Nothing is ever written back to your Boomi account — Weave reads.

What it is

Boomi's build canvas lives in a browser tab that only shows one component at a time, and the XML behind it isn't something you can usefully read or review. Weave imports a folder of component XML — a Boomi export, or a folder pulled straight from your account over the Platform API — into a local SQLite-indexed project, then renders it the way you'd want to read it: a real process canvas, typed component viewers, and full version history per component.

From there it does the three things the platform makes hardest:

• Diff two versions of a process visually, so a review is a picture rather than a wall of XML.
• Run a process on a local Atom and light up the shapes that actually executed, with assertions stored as YAML next to your components.
• Ask questions about the integration you have open, answered from your own components rather than from generic Boomi documentation.

Why install it

Requirements

• VS Code 1.99 or newer

No .NET install is needed: the Marketplace serves a platform-specific build (Windows, macOS, Linux — x64 and arm64) with a self-contained sidecar bundled. Only a manually built framework-dependent VSIX needs a .NET 10 runtime on PATH (set weave.dotnetPath if your dotnet lives somewhere unusual).

Optional, per feature:

• A Boomi API account for API import, on-demand version pulls, and Get Latest Version.
• A local Atom running the Weave agent, for live test runs.
• A chat backend for Ask AI — VS Code's native chat, or the claude CLI.

Install

From the Extensions view: ··· → Install from VSIX…, or:

code --install-extension weave-boomi-0.1.0.vsix

Open the Weave view in the activity bar. The extension starts its bundled sidecar on first use and logs to the Weave output channel; if anything fails to start, the message names what to fix.

Setup

Weave is folder-confined: the open VS Code folder IS the project. All local state lives in <folder>/.weave/ (gitignored except the committed workspace.json + manifest.json), the component XML lives in components/, and tests in weave-tests/ — one Boomi project per folder, git-friendly by construction.

1 · Get a workspace folder

Two ways in:

• Open an existing Weave workspace (e.g. a git clone) — the folder registers itself automatically; no prompt, no Boomi account needed.
• Weave: Pull Project from Boomi Account… — from the welcome view in any other folder: lists your account's folder tree and downloads a subtree over the Platform API into the open folder, with per-component progress and cancel (already-downloaded components are kept if you stop). Weave: Refresh from Boomi (Re-pull)… reconciles against the platform later.

The tree mirrors Boomi's explorer: folders from folderFullPath, components grouped under per-type nodes, re-rooted at the pulled account folder.

2 · Connect your Boomi account (optional)

Run Weave: Configure Boomi Account — it has a Test Connection probe and stores the API token in VS Code's encrypted SecretStorage rather than in settings.

3 · Set up the local test agent (optional — required to run anything)

Run Weave: Set Up Local Test Agent…, choose a folder the Atom's service account can read, and Weave copies the agent jar there and opens the exact block to append to <AtomDir>\bin\atom.vmoptions:

# Weave local test agent
-javaagent:C:\weave-agent\boomi-injector-agent.jar
-DWEAVE_LOCAL=true
-DBOOMI_ACCOUNT_ID=your-account-id
-Dboomi.agent.log=C:\weave-agent\agent-hook.log
# Optional: -DINJECTOR_DISPLAY_NAME=My Atom | -DATOM_DIR=<AtomDir>

Restart the Atom and it appears in the test view's agent picker. Verify with Weave: Test Agent Diagnostics, which round-trips a live health snapshot.

Atoms usually run as a service. The agent publishes its handshake file to %LOCALAPPDATA%\Weave\agent for normal accounts and %ProgramData%\Weave\agent for service, system, and root accounts. Weave scans both, so a service-hosted Atom is found without extra configuration — weave.test.agentDir is only for non-standard layouts.

4 · Point Ask AI at a backend (optional)

weave.ai.chatSurface decides where the Ask AI actions go: VS Code's native chat via the @weave participant, or Weave's own panel backed by the claude CLI. On auto it uses native chat when a chat extension is present and falls back to the CLI. Set weave.ai.claudePath if the CLI isn't on PATH.

5 · Write your first test

Weave: New Boomi Test File… scaffolds a *.boomitest.yaml already bound to the process you have open. The bundled JSON schema gives completions and squiggles as you type, and the file appears in the Testing pane as soon as it's saved.

Features

Process viewer

Open any process and get the shape graph with labels, branch names, and the components each shape uses rendered as links — click the operation under a start shape and you land on that operation. The version stepper walks history without leaving the tab, and the XML tab shows the source behind whatever you're looking at.

• Component tree grouped by type, per project
• Search within a component, or across every component in the project
• Zoom, pan, and a Names toggle for internal shape names
• Test View and Ask AI are one click away

The start shape names its operation and the process call names its sub-process — both links. Try and Catch branches are labelled on the wire.

Version diff

Choose a source and target version and Weave overlays the difference on the process itself. Changed shapes are outlined and badged, unchanged ones fade back, and rerouted connections are drawn along the path they now take. Click a shape for the specific change, and copy its XML if you need it.

• 🟢 Added (target) · 🔴 Removed (source) · 🟠 Config changed · 🟠 Rewired
• Toggle between the Source and Target layout
• Shape Changes for the selection, All Changes for the process
• Pure moves are deliberately not highlighted, so real edits stand out
• Also works between two selected components — visual or plain text

Six changes between v6 and v7. Init Tracking Properties is badged REWIRED, and its card names the routing edit precisely: shape15 → unset.

Test runner

Pick an agent and hit Run Test: Weave builds the deployment closure from your local snapshots, injects it into the Atom, and streams the run back. Shapes light up as batches arrive. When it finishes you get per-shape logs, documents, derived runtime properties, and coverage — and, for a YAML test case, a verdict with every assertion listed.

• Test data seeds documents into the process after its start shape — multiple documents, MIME types, and dynamic document or process properties
• Extensions overrides connection settings, operation fields, process property defaults, and cross-reference tables for the run only, then restores them
• Test cases appear in the native Test Explorer; Boomi Test Runs keeps history per case
• Step into a sub-process call and replay its own trace, no re-execution
• Export… writes the run's TestMessageSet XML; Load run results… replays one

ROUTE-03 passed in 3145 ms with nine assertions — including two checking values inside documents. The run reports 12/15 shapes, 80% shape and 50% branch coverage.

Ask AI

Ask about a project or a single component and the Weave Boomi Analyst works the question against your indexed integration — reading the process flow, following calls, fetching component detail — and shows which references it used. Because it reads the same snapshots the canvas draws, the answer is about your build, not about Boomi in general.

• Native chat via @weave, or Weave's own panel on the claude CLI
• Attach a test file or component to the question as context
• The tool trace is visible, so you can see what the answer is based on
• It can draft *.boomitest.yaml test cases for the process you're looking at

From an answer to the canvas

The interesting part of an answer is usually where do I change this. Component and shape names in a Weave answer are links: clicking one opens the right process, scrolls to the shape, selects it, and shows its configuration — so a recommendation turns into an edit without hunting for the shape it meant.

Following a reference selects that shape on the canvas and opens its card, which links on to the map component itself.

Commands

Every command is prefixed Weave: in the palette (Ctrl+Shift+P). Most are also on the context menu of the thing they act on.

Settings

Running tests in CI

The same code path the Test Explorer uses is available headlessly:

node cli/weave-test.mjs .\weave-tests --agent local-atom --junit out.xml
# exit 0 = all passed · 1 = failures · 2 = infrastructure error