Excel VBA development, rebuilt for CLI-first humans and AI agents.

Official Documentation

English | 日本語

🏄‍♂️ xlflow

xlflow is an Excel VBA development framework for the AI agent era.

It turns .xlsm workbooks into a source-controlled, CLI-driven development workflow where VBA can be exported, edited, linted, imported, tested, traced, and executed from the command line.

Tip

Think of xlflow as a development harness around Excel VBA: it does not replace Excel, but it makes Excel VBA projects much easier to operate from terminals, scripts, CI-like local checks, and AI coding agents.

Demo

These samples were created by an AI agent using xlflow with only minimal natural language instructions.

Macro that summarizes world news in Excel using NewsAPI	Macro that retrieves stock prices and displays them in Excel
Macro that generates QR codes using cell colors and displays them in Excel	Macro that allows playing Tetris within Excel
Macro that allows playing Space Invader on a UserForm	Rich calendar picker

---

Why xlflow?

Traditional VBA development is still heavily tied to the Excel UI and the Visual Basic Editor. That works for small manual edits, but it becomes painful when you want repeatable development, source control, tests, diffs, or AI-agent-assisted changes.

Pain in normal VBA development	What xlflow adds
VBA code is trapped inside .xlsm files	Export/import VBA as .bas, .cls, and .frm source files
Macro entrypoints are unclear	Discover runnable Public Sub procedures with xlflow macros
Existing UserForm layout is hard to review in diffs	Persist Designer state with xlflow form snapshot
Runtime failures are hard to locate	Return structured errors, diagnostics, and trace logs
Workbook changes are hard to review	Compare values, formulas, sheets, and exported VBA source
AI agents cannot safely operate Excel through the UI	Provide stable CLI commands and JSON output

pull → edit → push → lint → test/run → trace → diff

---

What xlflow can do

Area	Capabilities
Source control	Export and import standard modules, class modules, UserForms, and document modules
Execution	Run macros from the CLI with typed arguments
Testing	Discover and run VBA test procedures
Linting	Catch Option Explicit omissions, Select/Activate, broad error handling, implicit variants, public module fields, and interactive operations
GUI safety	Detect file pickers, input boxes, modal message boxes, and other automation-hostile boundaries
Debugging	Collect trace events and return runtime diagnostics
Diffing	Compare workbook cell values, formulas, sheet structure, and exported VBA source
AI agents	Return stable JSON and install bundled Skills for Codex, Claude, Cursor, Gemini, GitHub Copilot-style agent workflows, and other agents

Important

xlflow is Windows-first. Workbook operations use Microsoft Excel + COM + PowerShell.

Note

Excel COM-backed commands report the xlflow bridge host in top-level bridge. If workbook VBA launches its own external PowerShell process, that host can still differ from xlflow's bridge host. Inspect or log the workbook-side executable when debugging powershell.exe vs pwsh.exe behavior.

---

Requirements

Requirement	Needed for
Windows	Excel COM automation
Microsoft Excel	new, init, list forms, inspect form, form snapshot, form build, form export-image, pull, push, run, export-image, edit, test, macros, trace, doctor
PowerShell	Excel automation bridge
Trust access to the VBA project object model	Reading and writing VBA projects

Note

Commands that do not require Excel COM, such as lint, parts of diff, and Go unit tests, can be verified in non-Excel environments.

Warning

In Excel, enable Trust access to the VBA project object model before using commands that read or write VBA code. Without it, pull, push, run, and related commands may fail even when Excel itself is installed.

Details

In Excel options, please enable "Trust Center" → "Macro Settings" → "Trust access to the VBA project object model".

---

Installation

winget

winget install HarumiWeb.Xlflow

Use upgrade to update an existing installation:

winget upgrade HarumiWeb.Xlflow

Note

winget availability may lag behind a GitHub Release while the manifest is submitted and accepted upstream. Use Scoop or the GitHub Releases ZIP when you need the newest release immediately.

Scoop

scoop bucket add harumiweb https://github.com/harumiWeb/scoop-bucket
scoop install xlflow

GitHub Releases

Download prebuilt Windows binaries from:

https://github.com/harumiWeb/xlflow/releases

Important

Current prebuilt distribution targets Windows only. Commands that interact with workbooks still require Microsoft Excel, Excel COM automation, and Trust access to the VBA project object model. The release binary already embeds the runtime PowerShell bridge scripts, so xlflow.exe can run workbook commands without sidecar *.ps1 files.

Verify the downloaded ZIP against the published checksums.txt file:

Get-FileHash .\xlflow_windows_x86_64.zip -Algorithm SHA256
certutil -hashfile .\xlflow_windows_x86_64.zip SHA256

The reported SHA256 must match the entry for xlflow_windows_x86_64.zip in checksums.txt.

This confirms file integrity against the published checksum file. It does not prove publisher identity and is not a substitute for Windows Authenticode signing.

Verify the GitHub Actions provenance attestation with GitHub CLI:

gh attestation verify .\xlflow_windows_x86_64.zip --repo harumiWeb/xlflow

This confirms the artifact attestation published for the release artifact. It does not mean the ZIP is Authenticode-signed by a Windows publisher certificate.

Go install

go install github.com/harumiWeb/xlflow/cmd/xlflow@latest

go install may contact the Go module mirror and checksum database configured in your Go environment. For direct source checkout development and CI, treat the Go version declared in go.mod as the supported toolchain source of truth; the repository CI and release workflows resolve Go from that file.

Verify the installation:

xlflow version
xlflow --help

For development checkout usage:

go run ./cmd/xlflow --help

With Taskfile:

task run -- --help

---

Quick start

1. Create or initialize a project

Create a new xlflow project and macro-enabled workbook:

xlflow new Book.xlsm

new automatically pushes the scaffolded VBA modules into the new workbook, so a later pull starts from the same initial source.

Or start from an existing workbook:

xlflow init Book.xlsm

init automatically pulls VBA out of the copied workbook into src/, so you can edit source files immediately without a separate bootstrap pull.

Install the AI agent Skill during project creation:

xlflow new Book.xlsm --with-skill --agent codex

Interactive xlflow new and xlflow init render a welcome banner and may check the latest GitHub Release through the GitHub Releases API. Disable that request for one invocation with --no-update-check, or set XLFLOW_NO_UPDATE_CHECK=1 to disable it for interactive scaffolding in your environment.

2. Check the Excel automation environment

xlflow doctor --json

Tip

If pull, push, run, or test fails because of Excel, COM, PowerShell, or VBIDE settings, run doctor first.

3. Export VBA into source files

xlflow pull --json

Edit the exported .bas, .cls, and .frm files under src/ with your normal editor. When folder mode is enabled, nested directories under each configured source root are mapped to Rubberduck-compatible @Folder(...) annotations during push.

4. Import edited source back into the workbook

xlflow push --json

5. Discover and run macros

xlflow macros --json
xlflow run Main.Run --json

For unattended automation, prefer headless mode:

xlflow run Main.Run --headless --json

If the macro intentionally shows file pickers, message boxes, or UserForms, use interactive mode:

xlflow run Main.Run --interactive --timeout 5m --json

6. Lint and test

xlflow lint --json
xlflow test --json

---

Common workflows

Letting an AI agent edit VBA

Installing the Skill

When having an AI agent edit VBA using xlflow, it is recommended to install the Skill provided by xlflow into the agent's environment.

xlflow skill install

You can also install it at the same time you launch a project.

xlflow new Book.xlsm --with-skill

If you wish to manage skills using a manager such as vercel-labs/skills, please install the skill as follows.

npx skills add harumiWeb/xlflow/internal/agentskill/templates --skill xlflow

Creating a project

While you can leave the creation of the project itself to an AI agent, it is recommended that the initial project setup be performed by a human.

xlflow new Book.xlsm --with-skill

Letting an AI agent edit

Using the installed skill, please provide instructions for what you want to achieve in natural language.

/xlflow Create a macro that enters "Hello, world!" into cell A1 using VBA

Human-assisted Excel sessions

Use attach when a human has Excel open and you want to validate the active workbook before working with it:

xlflow attach --active --json

Note

attach is a safety check. It confirms that the active Excel workbook matches the configured excel.path; it does not change the target used by pull, push, or run.

GUI-heavy macros

Inspect GUI boundaries before deciding whether a macro can run headlessly:

xlflow inspect-gui --json

Result	Suggested mode
No GUI boundaries	xlflow run ... --headless --json
File picker, InputBox, modal MsgBox, or UserForm detected	xlflow run ... --interactive --timeout 5m --json
GUI code wraps core logic	Refactor core logic into parameterized headless procedures

Warning

Headless automation and modal Excel UI do not mix. Use inspect-gui before unattended runs and keep GUI entrypoints thin.

Runtime-aware VBA branches

New xlflow new projects include src/modules/XlflowRuntime.bas. During xlflow run and xlflow test, xlflow injects a workbook-scoped execution mode marker before user VBA starts, so code can branch without process inspection hacks.

If XlflowRuntime.IsHeadless() Then
  Debug.Print "running unattended in " & XlflowRuntime.ModeName()
Else
  MsgBox "Running interactively"
End If

run --headless resolves to headless, run --interactive resolves to interactive, and test resolves to test. Plain run falls back to interactive unless the xlflow process environment sets XLFLOW_MODE=interactive|headless|ci|agent|test.

---

Command map

Command	Purpose	Typical usage
new	Create a new xlflow project and .xlsm workbook	xlflow new Book.xlsm
init	Initialize xlflow from an existing workbook	xlflow init Book.xlsm
doctor	Diagnose Excel, COM, PowerShell, and VBIDE access	xlflow doctor --json
attach	Validate the workbook currently active in Excel	xlflow attach --active --json
backup list	List rollback-capable workbook backups	xlflow backup list --json
pull	Export VBA components into src/	xlflow pull --json
push	Import VBA source back into the workbook	xlflow push --json
rollback	Restore the workbook from a saved backup	xlflow rollback --latest --json
session	Keep the configured workbook open for fast loops	xlflow session start
save	Save the workbook held by a session	xlflow save --session --json
runner	Manage the persistent xlflow runner marker module	xlflow runner install --json
macros	Discover runnable macro entrypoints	xlflow macros --json
list forms	Discover workbook UserForms and expected source paths	xlflow list forms --json
form snapshot	Persist strict Designer UserForm state as JSON or YAML spec	xlflow form snapshot UserForm1 --out src/forms/specs/UserForm1.yaml --json
form build	Create a Designer-backed UserForm from a saved spec	xlflow form build src/forms/specs/UserForm1.yaml --json
form export-image	Export a runtime UserForm to a PNG image	xlflow form export-image UserForm1 --out artifacts/UserForm1.png --json
run	Execute a macro from the CLI	xlflow run Main.Run --json
export-image	Export a worksheet range to a PNG image	xlflow export-image --sheet QR --range A1:AE31 --json
edit	Mutate a live session workbook for setup and tuning	xlflow edit cell --sheet Input --cell B2 --value ABC123 --session --json
trace	Enable, collect, and clean VBA trace logs	xlflow trace enable --json
test	Run VBA tests	xlflow test --json
diff	Compare workbook content and optional VBA source	xlflow diff before.xlsm after.xlsm --json
inspect	Inspect saved workbook snapshots or explicit live session state	xlflow inspect range --sheet Result --address A1:F20 --session --json
lint	Lint VBA source	xlflow lint --json
analyze	Analyze runtime-risk patterns without opening Excel	xlflow analyze --json
check	Run lint, analyze, and doctor as a preflight	xlflow check --keepalive --json
inspect-gui	Detect GUI interaction boundaries	xlflow inspect-gui --json
skill install	Install the bundled xlflow Skill for AI agents	xlflow skill install --agent codex
version	Show the installed xlflow build metadata	xlflow version

---

Commands in detail

Detailed command behavior, options, JSON payloads, and troubleshooting notes now live in the documentation site.

• Command reference
• JSON output
• Configuration
• Troubleshooting

Use the README as a quick overview and the documentation site as the source for command-level details.

---

Configuration

xlflow reads xlflow.toml from the project root.

[project]
name = "sample"
entry = "Main.Run"

[excel]
path = "build/Book.xlsm"
visible = false
display_alerts = false

[src]
modules = "src/modules"
classes = "src/classes"
forms = "src/forms"
workbook = "src/workbook"

[vba]
folders = true
folder_annotation = "update"
default_component_folders = true

[userform]
code_source = "sidecar"

[lint]
require_option_explicit = true
forbid_select = true
forbid_activate = true
forbid_on_error_resume_next = true
detect_implicit_variant = true
forbid_public_module_fields = true
forbid_interactive_input = true

project.entry is used when xlflow run is invoked without a macro name.

Set forbid_interactive_input = false when the project intentionally uses dialogs or UserForms and you want to suppress VB007 warnings. This only affects lint output; xlflow run --headless still blocks GUI boundaries.

Syntax safety lint rules for typographic quotes, C-style quote escapes, unclosed or mismatched procedures, and malformed line-continuation underscores are always enabled because they prevent VBE compile dialogs before push or run opens Excel.

---

JSON output

Every command can return AI-agent-friendly JSON by passing --json.

The basic envelope is:

{
  "status": "ok",
  "command": "lint",
  "error": null,
  "logs": []
}

On failure, status is failed, and error.code and error.message are returned.

{
  "status": "failed",
  "command": "run",
  "error": {
    "code": "macro_failed",
    "message": "Main Err 5: inputPath is required",
    "source": "Main",
    "number": 5,
    "phase": "invoke_macro"
  },
  "logs": []
}

Tip

AI agents and automation scripts should treat status, command, error.code, and command-specific top-level fields as the primary contract.

---

Exit codes

Code	Meaning
0	Success
1	Validation failure, such as lint, macro, or test failure
2	CLI argument or configuration error
3	Environment error, such as Excel, COM, VBIDE, or PowerShell failure

Note

diff returns exit code 0 even when differences are found. Inspect diff.summary.total_diffs to determine whether inputs differ.

---

License

MIT License. See LICENSE.