SAP Testing Troubleshooting¶

Setup¶

The SAP Testing skills and commands don't appear¶

The feature is off. It switches on only when abapfs.testing.folder points at a folder that exists — run ABAP FS: Enable SAP UI Testing Features.

If it worked before and stopped, the folder was probably moved, renamed, or deleted. Point it at a folder that exists again.

VS Code asks me to approve every file Copilot writes¶

The test folder isn't open in your workspace, so every edit counts as an out-of-workspace change and needs confirmation. File → Add Folder to Workspace and pick your test folder.

You'll also get no IntelliSense or error checking in the generated test scripts until you do, because the tsconfig.json in the test folder only applies to workspace folders. Copilot warns about this when it looks the folder up.

The managed files keep coming back / my edits disappeared¶

That's intended. tsconfig.json, node_modules, playwright.config.js, and .sap-active-system are managed by ABAP FS and rewritten at startup and after every update, because they contain absolute paths into the extension's install directory. Don't edit them; if you delete them, reload VS Code and they come back.

`ERROR: No abapfs.remote entry for "..."` or `has no "client" property`¶

The connection Copilot was given doesn't exist or is incomplete. SAP Testing needs both a url and a client on the connection to build the WebGUI URL — see Connection Manager.

Models¶

"SAP testing subagent model configuration is invalid"¶

One of the models you chose is no longer offered by Copilot. Click Configure Models on the notification, or run ABAP FS: Set Models for Subagents, and pick replacements. Testing models are one all-or-nothing group when the testing folder is ready.

"No language models are currently available"¶

Copilot hadn't finished starting up when the panel loaded. Use Refresh in the panel.

Running tests¶

The test fails on a SAP logon screen¶

Auto-login didn't produce a session. Two normal causes: the connection has webGuiAutoLogin turned off, or the system doesn't issue reentrance tickets.

Check the ABAP FS output channel at Debug level and look for [sso] lines — they name the cause directly, and distinguish an authentication problem from a test problem. Never work around this by putting credentials in a test script; Copilot is instructed to refuse.

Cases come back BLOCKED instead of passed or failed¶

Blocked means the test never reached SAP because its data wasn't ready. It's a data problem, not a bug in the program.

Run phase 5 again for that system (prepare-data) and re-run. Copilot separates blocked from failed in its report for exactly this reason — six blocked cases and six failures mean very different things.

A test passed on screen but is reported as failed¶

That's the framework working. The screen said success but the database check that case declared came back wrong, which usually means a real defect: SAP reported success without persisting the right data. Investigate before dismissing it.

Microsoft Edge was not found¶

Set abapfs.testing.edgePath to your browser executable, or install Edge. Recording needs it; test runs strongly prefer it.

An ABAP FS tool returns HTTP 401, 403, or 5xx¶

ABAP FS can't reach the SAP system — almost always an expired session. Check the connection and reload VS Code to re-establish it, then retry. This is not a permissions restriction to work around, and Copilot is told not to fall back to reading tables through the browser or to invent values.

`_index.docx` couldn't be written¶

The file is open in Word. Close it and rebuild. If it can't get the lock, Copilot writes a timestamped copy instead and tells you both paths.

Working with Copilot¶

Copilot says a tool is missing¶

VS Code doesn't always surface every tool immediately, and playwright_test is the usual casualty. Starting a fresh chat normally fixes it. Copilot is instructed to tell you rather than fake a result or fall back to a terminal command — so treat the report as accurate.

Copilot redid work that looked finished¶

A reviewer agent or a gated tool rejected it. The reviewers read the actual ABAP source and challenge the analysis or the test plan, and build_test_index and playwright_test refuse to run until the required review or readiness check genuinely passed. This is the framework catching a gap — see Quality gates.

Copilot keeps asking instead of just doing it¶

By design. A control it can't drive, data it can't find, or an ambiguous screen is a stop-and-ask situation, because the alternative — a plausible guess — produces a test that looks finished and silently proves nothing. Give it the answer, a sample file, or a recording.

Copilot won't click a button¶

Anything destructive — post, delete, release, approve, or Execute on a program in update mode — needs your explicit approval first. Confirm in chat, ideally after checking the system is safe to write in.