test(server): run fake provider CLIs through a Node stub on every host

[juliusmarminge]

Seven suites faked a provider CLI with a #!/bin/sh script made executable
by chmod, which Windows cannot run: the shebang is ignored and there is no
extension for PATHEXT to match, so the spawn either failed or fell through
to a real CLI on PATH. CodexTextGeneration's fake was a 120-line sh
argument parser.

Add apps/server/src/testUtils/fakeCli.ts. writeFakeCli writes the CLI's
behaviour as a Node stub and a launcher shaped for the host: a shebang
script on posix, a .cmd shim on Windows, which resolveSpawnCommand already
routes through a shell. execScriptSource covers the common case of handing
over to the mock ACP agent after an argv check. ClaudeTextGeneration, which
had its own .cmd branch, moves onto the shared helper too.

The tests that assert the agent logged SIGTERM on stop are skipped on
Windows, where the child is terminated rather than signalled.

Part of the Windows test-suite stack rooted at #9564; the manual Windows lane comes from #9538.

Model: Claude Fable 5 in Claude Code.

🤖 Generated with Claude Code

Note

Run fake provider CLI fixtures through a shared Node stub on Windows and POSIX hosts

• Adds a shared writeFakeCli utility in fakeCli.ts that emits a Node stub module plus a platform-appropriate launcher (POSIX shell or Windows .cmd), replacing per-test shell-script generation across provider and text-generation test suites
• Refactors CursorAdapter, CursorProvider, GrokAdapter, GrokProvider, CodexTextGeneration, CursorTextGeneration, GrokTextGeneration, and ClaudeTextGeneration tests to build their fake agent binaries through the shared writer
• Adds a Windows .cmd wrapper for the Codex collaboration mock peer and selects it on win32 in the integration test; adds .cmd/.bat CRLF checkout rules in .gitattributes
• Guards SIGTERM-dependent child-process exit assertions with a win32 skip condition in CursorAdapter, CursorProvider, GrokAdapter, CursorTextGeneration, and CodexCollabRuntime tests, since Windows terminates processes without delivering SIGTERM
• Risk: the new writeFakeCli writer is the single path for fake CLI creation; any mismatch between the generated Node stub contract and an individual test's expectations (e.g. argument validation in execScriptSource) will surface as fixture failures across all refactored suites

^{Macroscope summarized b4e082c.}

---

Note

Low Risk
Changes are limited to test fixtures, git attributes, and conditional test skips; no production provider or spawn logic is modified.

Overview
Provider and text-generation tests no longer build fake CLIs with hand-written #!/bin/sh wrappers (and chmod), which fail or hit the real binary on Windows. They now go through writeFakeCli in apps/server/src/testUtils/fakeCli.ts, which writes a Node stub plus a host-specific launcher (POSIX shell script or .cmd shim), with env applied via a JSON sidecar. execScriptSource covers the common “validate argv, then import the mock ACP agent” pattern; Codex’s large shell-based fake is reimplemented in Node the same way.

Codex collab integration picks codexCollabMockPeer.cmd on win32 (new file); .gitattributes forces CRLF for *.cmd / *.bat.

Tests that assert SIGTERM in an exit log are skipIf on Windows, where children are killed without that signal. Production spawn paths are unchanged—this is test/fixture plumbing for the Windows test suite.

^{Reviewed by Cursor Bugbot for commit b4e082c. Bugbot is set up for automated code reviews on this repo. Configure here.}

[github-actions]

Thread transfer impact

✅ Thread transfer remains within every enforced ceiling.

ℹ️ No successful main baseline artifact is available yet. This run establishes the initial measurement.

Provider	Metric	Main baseline	This PR	Impact	PR ceiling
Codex	Total thread wire	—	13.8 KiB	—	15.1 KiB	✅
Codex	Thread snapshot wire	—	7.0 KiB	—	7.3 KiB	✅
Codex	Live turn WebSocket wire	—	6.7 KiB	—	7.8 KiB	✅
Codex	Live turn WebSocket decoded	—	58.5 KiB	—	66.4 KiB	✅
Codex	Live turn messages	—	10	—	21	✅
Claude	Total thread wire	—	13.8 KiB	—	15.1 KiB	✅
Claude	Thread snapshot wire	—	7.0 KiB	—	7.3 KiB	✅
Claude	Live turn WebSocket wire	—	6.7 KiB	—	7.8 KiB	✅
Claude	Live turn WebSocket decoded	—	59.3 KiB	—	66.4 KiB	✅
Claude	Live turn messages	—	10	—	21	✅

Baseline: unavailable · PR result: b4e082c · Source CI: success

Scenario and decoded snapshot size

10 historical turns, 5 command tools per turn, 878.9 KiB retained MCP result per historical turn, and a 1.05 MiB retained result in the measured turn.

• Codex decoded thread snapshot: 113.8 KiB
• Claude decoded thread snapshot: 114.5 KiB

Updated in place by a trusted workflow. PR artifacts are strictly validated and never executed.

[macroscopeapp]

Approvability

Verdict: Not approved

Macroscope's review found this PR not approvable — The changes are confined to cross-platform test fixtures and test utilities, with no product-runtime or default-setting impact. Human review is required because the PR adds file-level and line-level directives that suppress static-analysis diagnostics.

No code changes detected at b4e082c. Prior analysis still applies.

^{You can add or adjust custom eligibility rules. Learn more.}