Xlflow Tmp Workspace E2e

xlflow tmp_workspaces E2E

Execute the repository-standard xlflow verification flow in isolated workspaces under tmp_workspaces.

Follow the workflow exactly unless the user narrows scope.

When the task is release preparation, treat this skill as a release gate rather than an optional spot check.

Prefer a session-first Excel workflow for any real-workbook verification that needs more than one Excel COM-backed command. Reopening the workbook separately for push, run, test, pull, save, or repeated inspect steps frequently makes validation much slower and can stall or look hung. Unless the check is truly one-shot, start a session once, reuse it, and stop it only after the verification slice is complete.

Preconditions

• Read tasks/lessons.md before creating the workspace.
• Work from the repository root: C:\dev\go\xlflow.
• Use xlflow from PATH, not go run, unless the user explicitly asks otherwise.
• Prefer a fresh workspace per verification attempt. Do not reuse an old tmp_workspaces project unless the user asks to inspect a prior artifact.

Workspace Rules

Create one or two fresh directories under tmp_workspaces:

• primary E2E workspace:
  • tmp_workspaces/<topic>-e2e
• optional init workspace:
  • tmp_workspaces/<topic>-init

Before using a workspace:

• remove the old directory if it exists and the user did not ask to preserve it
• recreate the directory empty
• record the exact absolute path for reporting

Standard Flow

0. Session-first rule for Excel COM verification

For blank scaffold checks such as new, doctor, first pull, and first lint, normal one-shot commands are fine.

For any workflow that chains workbook-backed execution or mutation commands, prefer this shape:

xlflow session start --json
xlflow push --fast --session --no-save --json
xlflow run <MacroName> --session --json
xlflow test --session --json
xlflow save --session --json
xlflow session stop --json

Adjust the middle commands to the scenario, but keep the same principle: reuse one live workbook session instead of reopening Excel for each step. Only skip the session when the user explicitly asks for saved-file-only validation or when the path is intentionally verifying non-session behavior.

1. Confirm tool resolution

Run Get-Command xlflow and confirm the resolved binary path.

If the installed binary must reflect current source changes, rebuild with:

go install ./cmd/xlflow

2. Blank-workbook verification

In the primary workspace, run this sequence:

xlflow new --json
xlflow doctor --json
xlflow pull --json
xlflow lint --json

Verify all commands return success.

After new, confirm the scaffold exists:

• xlflow.toml
• build/Book.xlsm or the requested workbook
• src/modules
• src/classes
• src/forms
• src/workbook
• .xlflow

After pull, inspect exported workbook modules under src/workbook.

Pay special attention to the document-module normalization rule from tasks/lessons.md:

• ThisWorkbook.bas and worksheet modules must not retain VERSION, BEGIN, MultiUse, END, or Attribute VB_ lines in source form
• blank workbook modules should still lint cleanly

3. Macro round-trip verification

Use this path when push, run, workbook save behavior, or VBA round-trip behavior matters.

Strong default: use a session for this flow. Do not run push, run, and follow-up workbook checks as separate fresh Excel opens unless the point of the test is specifically to verify non-session behavior.

Create a minimal runnable module at src/modules/Main.bas:

Attribute VB_Name = "Main"
Option Explicit

Public Sub Run()
    ThisWorkbook.Worksheets(1).Range("A1").Value = "xlflow ok"
End Sub

Then run:

xlflow lint --json
xlflow session start --json
xlflow push --fast --session --no-save --json
xlflow run Main.Run --session --json
xlflow save --session --json
xlflow session stop --json
xlflow pull --json
xlflow lint --json

Verify:

• push reports imported source files and workbook-module updates when expected
• run reports the requested macro name
• session-backed save / stop leave the workbook clean without stray save-required state
• pull exports Main.bas plus workbook modules
• the final lint still passes

4. Workbook-state verification

When macro behavior should persist to the workbook, verify it through Excel COM instead of trusting CLI success output alone.

If the behavior under test already uses a session, keep that session for the workbook mutation commands and do the direct COM state check only after save --session or session stop, unless the test explicitly needs to inspect the live unsaved workbook.

Use a PowerShell check like:

$path = 'C:\dev\go\xlflow\tmp_workspaces\<topic>-e2e\build\Book.xlsm'
$excel = New-Object -ComObject Excel.Application
$excel.Visible = $false
$excel.DisplayAlerts = $false
$wb = $excel.Workbooks.Open($path)
try {
  $wb.Worksheets.Item(1).Range('A1').Value2
} finally {
  $wb.Close($false)
  $excel.Quit()
  [void][System.Runtime.InteropServices.Marshal]::ReleaseComObject($wb)
  [void][System.Runtime.InteropServices.Marshal]::ReleaseComObject($excel)
  [gc]::Collect()
  [gc]::WaitForPendingFinalizers()
}

Adapt the cell or workbook assertions to the behavior under test.

5. init verification

Use a second fresh workspace when init must be covered.

Run:

xlflow init <absolute-path-to-existing-workbook> --json
xlflow pull --json

Verify the copied workbook exists under the second workspace build/ directory and that exported components appear under src/.

6. Release-gate coverage

Use this section when the user asks for release readiness, release preflight, or "before release" verification.

Release-gate verification must cover:

• blank workbook scaffold:
  • new
  • doctor
  • pull
  • lint
• standard module round-trip:
  • push
  • run
  • workbook-state verification through Excel COM
  • pull
  • lint
• class module round-trip
• UserForm round-trip, including .frm and .frx
• init from an existing workbook

When the release includes pack changes, also cover the pack artifact smoke described in section 7:

• produce a .xlsm with xlflow pack (file-level, no Excel)
• open the produced artifact in Excel and run a packed macro
• assert an observable effect, such as a sentinel cell value, to confirm the packed VBA compiled and ran

When the release includes session-related changes, also cover:

• session start
• push --fast --session --no-save
• run --session and/or test --session
• save --session
• session stop
• explicit confirmation of any save-required warnings or session metadata that changed

Even when the change is not primarily about session state, prefer the same session-backed execution pattern for multi-step Excel COM release checks unless a specific saved-file reopen path is itself under test.

Do not report release readiness if one of these paths was skipped without calling it out explicitly.

7. pack artifact smoke

Use this section when the release includes changes to the experimental pack command (docs/specs/pack-command.md, ADR-0012).

pack is a pure-Go, file-level path: it regenerates xl/vbaProject.bin from source without opening Excel, so it always reports pack.vbe_validation = "not_performed". This smoke exists to confirm, at the release gate, that a representative packed artifact actually compiles and runs in real Excel. It is release-gate evidence only; it does not run in PR CI and it does not change pack's permanent no-validation contract.

Keep the automated PR path Linux/pure-Go only. Do not add Excel or COM steps to PR CI; the smoke is performed here, in the release gate, on Windows with Excel.

In the primary workspace, make the standard src/modules/Main.bas macro from section 3 write a known sentinel value to A1. Keep the sentinel in a module that already exists in the template (for example Main): the experimental pack MVP updates existing modules only and rejects a brand-new module with pack_ambiguous_layout. Produce the artifact at the file level:

xlflow pack --out dist/Book.xlsm --experimental --json

Confirm the JSON envelope reports status = "ok", pack.backend = "pure-go", and pack.vbe_validation = "not_performed".

Open the produced artifact, run the packed macro, and assert the sentinel cell:

$path = 'C:\dev\go\xlflow\tmp_workspaces\<topic>-e2e\dist\Book.xlsm'
$excel = New-Object -ComObject Excel.Application
$excel.Visible = $true   # keep Excel visible: a hidden VBE compile-error dialog otherwise looks like a hang
$excel.DisplayAlerts = $false
$excel.AutomationSecurity = 1  # msoAutomationSecurityLow: allow the packed macros to run
$wb = $excel.Workbooks.Open($path)
try {
  $excel.Run('Main.Run')                      # forces a VBE compile, then runs the packed macro
  $sentinel = $wb.Worksheets.Item(1).Range('A1').Value2
  if ($sentinel -ne 'xlflow ok') {            # the value Main.Run writes in section 3
    throw "pack smoke failed: A1 was '$sentinel', expected 'xlflow ok'"
  }
  "pack smoke OK: A1 = $sentinel"             # surface the observed value for the release-gate report
} finally {
  $wb.Close($false)
  $excel.Quit()
  [void][System.Runtime.InteropServices.Marshal]::ReleaseComObject($wb)
  [void][System.Runtime.InteropServices.Marshal]::ReleaseComObject($excel)
  [gc]::Collect()
  [gc]::WaitForPendingFinalizers()
}

The smoke passes only when:

• pack exits 0 and produces the .xlsm
• the packed workbook opens in Excel without a compile error
• Run executes the packed macro and the sentinel cell holds the expected value

A compile error on open, a missing or wrong sentinel value, or a non-zero pack exit is a release blocker. If Run appears to hang, a modal VBE compile-error dialog is the usual cause — bring the Excel window to the foreground to read and dismiss it. Record the pack JSON output and the observed sentinel value in the release-gate report.

Failure Handling

Treat verification failures as product bugs unless evidence shows test setup is wrong.

When a command fails:

1. capture the exact failing command and JSON output
2. inspect generated files or workbook state
3. identify root cause before editing code
4. add or update tests when fixing the product
5. rerun the affected E2E path from a fresh workspace

Do not stop at "the command failed"; continue through fix and re-verification unless the user redirects.

Reporting Contract

Always report:

• the exact workspace path or paths used
• whether blank-workbook verification ran
• whether macro round-trip verification ran
• whether init verification ran
• whether release-gate coverage was complete, partial, or intentionally narrowed
• concrete evidence:
  • command results
  • generated artifact checks
  • Excel COM workbook-state checks when relevant
• any untested path that remains out of scope