Back language model tools and the Command Explorer with new LSP requests by andyleejordan · Pull Request #2298 · PowerShell/PowerShellEditorServices

Andy Jordan (andyleejordan) · 2026-06-09T22:37:03Z

Server-side support for new VS Code extension features (language model tools, a redesigned Command Explorer, and a read-only Show Help pane). The client work lives in PowerShell/vscode-powershell#5508 and depends on this PR; the two should be reviewed and merged together.

Return help text from the powerShell/showHelp request — convert showHelp from a notification to a request returning ShowHelpResult { HelpText }, capturing Get-Help -Full | Out-String and trimming both ends (it pads a leading and trailing blank line).
Enhance get_command for tools and the Command Explorer — optional Name/Module filters (wildcard, -ErrorAction Ignore), an ExcludeParameters fast path plus ModuleVersion, unconditional exclusion of editor-injected commands (the fake PSConsoleHostReadLine 0.0.0 and VS Code shell-integration helpers), and opt-in ExcludeDefaultFunctions enumerated from InitialSessionState.CreateDefault2().
Add a powerShell/getModule handler — returns a single module's metadata for the Command Explorer's hover tooltips. Covered by E2E tests verifying metadata is returned for an existing module and a null result for a missing module.
Add Copilot instructions documenting build and test.

Drafted by Copilot (Claude Opus 4.8) — please review/edit before marking ready.

Copilot

Pull request overview

This PR adds server-side LSP request support needed by upcoming VS Code extension features (language model tools, redesigned Command Explorer, and a read-only Show Help pane), aligning PowerShellEditorServices with new client capabilities in the companion VS Code PR.

Changes:

Converts powerShell/showHelp from a notification into a request that returns full help text as a string.
Enhances powerShell/getCommand with optional name/module filtering, a fast path to exclude parameter metadata, and filtering for editor-injected/default-session commands.
Adds a new powerShell/getModule request handler plus Copilot build/test instructions documentation.

Reviewed changes

Copilot reviewed 5 out of 5 changed files in this pull request and generated 4 comments.

Show a summary per file

File	Description
src/PowerShellEditorServices/Services/PowerShell/Handlers/ShowHelpHandler.cs	Switches showHelp to a request and returns trimmed help text from `Get-Help -Full \| Out-String`.
src/PowerShellEditorServices/Services/PowerShell/Handlers/GetModuleHandler.cs	Adds a new LSP request to retrieve module metadata for Command Explorer tooltips.
src/PowerShellEditorServices/Services/PowerShell/Handlers/GetCommandHandler.cs	Adds filtering/fast-path options and excludes editor-injected/default-session commands.
src/PowerShellEditorServices/Server/PsesLanguageServer.cs	Registers the new `GetModuleHandler` with the language server.
.github/copilot-instructions.md	Documents local build/test workflows and repo architecture/conventions for Copilot.

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

We had no `.github/copilot-instructions.md`, so agents working in this repo had to rediscover how to build and test, how the projects fit together, and how we label pull requests. This adds a single instructions file covering all of it: - **Build & Test** — `dotnet` directly for the fast inner loop versus `Invoke-Build` (which needs `InvokeBuild`/`platyPS`) for assembling the full module and running the complete CI suite. - **Architecture** — the project layout, key services, the LSP/DAP handler pattern, and how the server is wired up. - **Conventions** — C# style enforced by `.editorconfig`, the xUnit testing setup, and the multi-targeting story. - **Pull Request Labels** — every PR needs at least one `Area-*` label and exactly one `Issue-*` label (plus `OS-*` and `Ignore` when relevant). The `Issue-*` label is what GitHub's auto-generated release notes key off of (see `.github/release.yml`); a PR without one silently falls through to "Other Changes 🙏". The build, architecture, and conventions content previously rode along in #2298; consolidating it here keeps that PR focused on its LSP changes and gives us a single source of truth for the instructions. Drafted by Copilot (Claude Opus 4.8). Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

* Make `OnIdle` tests deterministic by polling instead of sleeping `CanRunOnIdleTask` (and its twin `CanRunOnIdleInProfileTask`) were flaky on the net462 (Windows PowerShell 5.1) CI leg — the former was just caught failing on PR #2298's Windows job. The root cause is that `PsesInternalHost.OnPowerShellIdle` calls `Events.GenerateEvent(PSEngineEvent.OnIdle, ...)`, which only *enqueues* the event. For a subscriber registered with `-Action {...}`, PowerShell doesn't run the action scriptblock inline; it becomes a pending action that the engine dispatches asynchronously on the pipeline thread, around subsequent pipeline invocations. So the action's execution was never synchronized with the test's `$handled` read, and the fixed `Thread.Sleep(2000)` was just a timing guess — sometimes too short on the slower WinPS leg, leaving `$global:handled` still `$false` at the assertion. The key realization is that each *additional* pipeline execution gives the engine another chance to drain the pending action, so re-reading the handler variable in a loop both waits for *and* drives completion. I replaced the sleep with a shared `WaitForHandledAsync` helper that polls the variable (~200ms apart, ~15s ceiling) until it reports `$true`, returning the last observed value on timeout so the assertion still fails loudly. This keeps the tests' intent intact and isn't merely a longer sleep. I validated both tests on net8.0 (green across repeated runs, ~0.4s each vs. the old fixed 2s); net462 can't run on macOS, but the mechanism is identical across targets and the 15s ceiling self-terminates on success, so it's strictly safer on the slow leg without slowing the fast one. Drafted by Copilot (Claude Opus 4.8). Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com> * Move `OnIdle` assertion into helper and drop list return Small follow-up to review feedback: both call sites only ever asserted the handler variable became `$true`, so the `IReadOnlyList<bool>` return was needless ceremony. `AssertHandledAsync` now owns the assertion — it returns once the variable reports `$true` and otherwise fails via `Assert.Fail` when the ~15s poll window elapses, which reads as "the OnIdle handler never ran." `Assert.Fail` is fine here — we're on xUnit 2.9.3 and already use it in the E2E tests. No behavior change to what's being verified; the call sites just shrink to a single `await OnIdleTestHelpers.AssertHandledAsync(...)`. Still green on net8.0. Drafted by Copilot (Claude Opus 4.8). Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com> * Assert `OnIdle` poll result with `Assert.True` Follow-up to review feedback: the helper short-circuited with a bare `return` on success and only asserted (`Assert.Fail`) on timeout, so the happy path had no explicit assertion. Restructure the loop to poll until the handler variable is `$true` or the ~15s window elapses, then assert the outcome once with `Assert.True(handled, ...)`. Same behavior, but the success and timeout paths now share a single, self-describing assertion. Still green on net8.0. Drafted by Copilot (Claude Opus 4.8). Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com> --------- Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

We had no `.github/copilot-instructions.md`, so agents working in this repo had to rediscover how to build and test, how the projects fit together, and how we label pull requests. This adds a single instructions file covering all of it: - **Build & Test** — `dotnet` directly for the fast inner loop versus `Invoke-Build` (which needs `InvokeBuild`/`platyPS`) for assembling the full module and running the complete CI suite. - **Architecture** — the project layout, key services, the LSP/DAP handler pattern, and how the server is wired up. - **Conventions** — C# style enforced by `.editorconfig`, the xUnit testing setup, and the multi-targeting story. - **Pull Request Labels** — every PR needs at least one `Area-*` label and exactly one `Issue-*` label (plus `OS-*` and `Ignore` when relevant). The `Issue-*` label is what GitHub's auto-generated release notes key off of (see `.github/release.yml`); a PR without one silently falls through to "Other Changes 🙏". The build, architecture, and conventions content previously rode along in #2298; consolidating it here keeps that PR focused on its LSP changes and gives us a single source of truth for the instructions. Drafted by Copilot (Claude Opus 4.8). Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Convert `powerShell/showHelp` from a fire-and-forget notification into a request that returns `ShowHelpResult { HelpText }`, so the client can render help in a read-only editor pane (and the language model `get_help` tool can reuse the same path) instead of printing into the integrated console. The handler captures `Get-Help -Full | Out-String` and `.Trim()`s both ends — `Out-String` pads the output with a leading and trailing blank line, which looked wrong in the help pane and in tool output. Drafted by Copilot (Claude Opus 4.8). Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

The handler previously serialized the entire command table (names, modules, and full parameter metadata) on every request, which is slow enough to hang the `get_command` language model tool and made the Command Explorer take minutes to populate. Extend `GetCommandParams` so callers can ask for only what they need: - `Name`/`Module` (both wildcard-capable) scope the `Get-Command` call so we don't materialize everything; an unmatched filter writes a non-terminating error, so we pass `-ErrorAction Ignore` and return an empty list instead. - `ExcludeParameters` takes a fast path that returns just name, module, and the new `ModuleVersion` without touching `Parameters`/`ParameterSets`, whose resolution and serialization dominate the cost. - Editor-injected commands are always skipped: the PSES host's fake `PSConsoleHostReadLine` (version 0.0.0) and VS Code's shell-integration helpers (`__VSCode-Escape-Value`, `Set-MappedKeyHandler[s]`) are plumbing, not commands a user authored or imported. - `ExcludeDefaultFunctions` (opt-in) drops PowerShell's module-less default-session functions (`cd..`, `prompt`, `TabExpansion2`, ...) and the install's `pwsh.profile.resource` script. The names come from `InitialSessionState.CreateDefault2()` so the list stays correct across PowerShell versions; module-provided commands are never affected. Drafted by Copilot (Claude Opus 4.8). Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

The Command Explorer groups commands under versioned module nodes and shows a tooltip on hover. Add a `getModule` request that returns a single module's metadata (version, description, path, author, company, project URI, required PowerShell version) so the client can populate those tooltips lazily, and register the handler in `PsesLanguageServer`. Drafted by Copilot (Claude Opus 4.8). Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Merging #2328 un-skipped the Windows PowerShell E2E suite, which surfaced a pre-existing host gap in `CanSendGetModuleRequestAsync`: it probed `Microsoft.PowerShell.Management`, but Windows PowerShell's in-box core modules are snap-in based and are not returned by `Get-Module -ListAvailable`. The handler correctly resolved nothing, so `Assert.NotNull` failed on the Windows PowerShell leg while passing on PowerShell 7 (where the core modules ship as files). Probe `Microsoft.PowerShell.Archive` instead: it is a file-based module present in the module path on both Windows PowerShell and PowerShell 7+, so the test exercises the handler on every host without skipping or loosening any assertion. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

The `TestE2EPowerShell` build task launches the server with `$env:PSModulePath = ''` (the vscode-powershell#3886 workaround for inheriting `powershell.exe`'s module path). PSES's `UpdatePSModulePath()` then appends the bundled module directory, so the runspace ends up with `;<bundled>` — a non-empty path. That distinction matters: Windows PowerShell only auto-injects its default module locations (`$PSHOME\Modules`, etc.) when `PSModulePath` is empty. The moment it's non-empty, no defaults are added, so `Get-Module -ListAvailable` finds nothing and `powerShell/getModule` returns null for every module — not just `Microsoft.PowerShell.Archive`. I confirmed this by probing `powershell.exe` directly: an empty path resolves the module, while any non-empty path (including the leading-`;` shape PSES produces) does not. pwsh 7 isn't affected because `TestE2EPwsh` never clears the path, and real VS Code sessions always have a populated `PSModulePath`, so the feature works in practice. This is a test-host artifact, not a product bug, so skip both getModule tests on the in-box Windows PowerShell leg: - `CanSendGetModuleRequestAsync` was failing outright (null module). - `CanSendGetModuleRequestForMissingModuleAsync` was passing vacuously — it asserts null, which every module now returns — so it validated nothing on this leg. Promote it from `[Fact]` to `[SkippableFact]` and skip it too. Both still run on Linux, macOS, and the PowerShell 7 Windows legs. This mirrors the earlier `73293931e` skip that was reverted when we tried the "probe Archive without skipping" approach, which the harness's emptied `PSModulePath` makes unworkable. Drafted by Copilot (Claude Opus 4.8). Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Andy Jordan (andyleejordan) mentioned this pull request Jun 9, 2026
Add PowerShell language model tools and redesign the Command Explorer PowerShell/vscode-powershell#5508
Merged

Andy Jordan (andyleejordan) requested review from Justin Grote (JustinGrote) and Patrick Meinecke (SeeminglyScience)June 9, 2026 22:58

Andy Jordan (andyleejordan) marked this pull request as ready for review June 10, 2026 00:48

Andy Jordan (andyleejordan) requested a review from a team as a code owner June 10, 2026 00:48

CopilotAI review requested due to automatic review settings June 10, 2026 00:48

Copilotstarted reviewing on behalf of Andy Jordan (andyleejordan)June 10, 2026 00:48 View session

CopilotAI reviewed Jun 10, 2026
View reviewed changes

Copilotstarted work on behalf of Andy Jordan (andyleejordan)June 10, 2026 01:02 View session

Copilotfinished work on behalf of Andy Jordan (andyleejordan)June 10, 2026 01:07

Copilotstarted work on behalf of Andy Jordan (andyleejordan)June 10, 2026 01:08 View session

Copilotfinished work on behalf of Andy Jordan (andyleejordan)June 10, 2026 01:11

Andy Jordan (andyleejordan)force-pushed the andyleejordan/lm-tools-command-explorer branch from a8b603f to 8dfaff7Compare June 10, 2026 17:10

Andy Jordan (andyleejordan) commented Jun 10, 2026
View reviewed changes

Comment threadsrc/PowerShellEditorServices/Services/PowerShell/Handlers/GetModuleHandler.cs

Andy Jordan (andyleejordan) commented Jun 10, 2026
View reviewed changes

Comment threadsrc/PowerShellEditorServices/Services/PowerShell/Handlers/ShowHelpHandler.cs

Andy Jordan (andyleejordan)force-pushed the andyleejordan/lm-tools-command-explorer branch 3 times, most recently from 7d3efc5 to 25305c7Compare June 10, 2026 18:19

Andy Jordan (andyleejordan) enabled auto-merge (rebase) June 15, 2026 15:54

Andy Jordan (andyleejordan) added Area-Language Server Area-API Issue-Enhancement A feature request (enhancement). labels Jun 16, 2026

Andy Jordan (andyleejordan)force-pushed the andyleejordan/lm-tools-command-explorer branch from 25305c7 to ac66fc4Compare June 16, 2026 15:56

Andy Jordan (andyleejordan) mentioned this pull request Jun 16, 2026
Fix flaky CanRunOnIdleTask by polling instead of sleeping #2314
Merged

Andy Jordan (andyleejordan)force-pushed the andyleejordan/lm-tools-command-explorer branch from ac66fc4 to 3ab4c88Compare June 23, 2026 21:20

Andy Jordan (andyleejordan)and others added 2 commits June 24, 2026 15:15

Andy Jordan (andyleejordan)force-pushed the andyleejordan/lm-tools-command-explorer branch from 452909f to 44cfcc0Compare June 24, 2026 22:16

Justin Chung (jshigetomi) approved these changes Jun 25, 2026
View reviewed changes

Andy Jordan (andyleejordan) merged commit 0190500 into mainJun 25, 2026
10 checks passed

Andy Jordan (andyleejordan) deleted the andyleejordan/lm-tools-command-explorer branch June 25, 2026 17:53

Navigation Menu

Uh oh!

Back language model tools and the Command Explorer with new LSP requests - #2298

Back language model tools and the Command Explorer with new LSP requests#2298
Andy Jordan (andyleejordan) merged 5 commits into
mainfrom
andyleejordan/lm-tools-command-explorer

Andy Jordan (andyleejordan) commented Jun 9, 2026•
edited by Copilot AI
Loading

Uh oh!

CopilotAI left a comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

Uh oh!

Conversation

Andy Jordan (andyleejordan) commented Jun 9, 2026• edited by Copilot AILoading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

CopilotAI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

Andy Jordan (andyleejordan) commented Jun 9, 2026•
edited by Copilot AI
Loading