Stabilize compaction E2E tests

[stephentoub]

Compaction E2E coverage was still skipped in Node, Python, and Go even though the equivalent .NET and Rust tests had already been stabilized. This re-enables those skipped tests by using the same deterministic trigger flow across SDKs.

Summary

• Re-enable the compaction E2E tests in Node, Python, and Go.
• Register compaction event waiters before sending the prompts that trigger compaction.
• Wait for session.compaction_start and a successful session.compaction_complete, matching the .NET/Rust behavior that ignores transient failed compaction attempts.
• Verify the session can continue after compaction using the shared replay snapshot.

Validation

• npx tsc --noEmit
• npx vitest run --no-coverage test\e2e\compaction.e2e.test.ts
• pytest e2e\test_compaction_e2e.py -v
• go test .\internal\e2e\ -run TestCompactionE2E -v -timeout 300s

The required Opus 4.7 xhigh code-review/fix loop was run before this PR; it identified the need to wait for successful compaction completion events, which is included here.

[Copilot]

Pull request overview

This PR stabilizes and re-enables compaction E2E coverage for Node.js, Python, and Go, aligning them with existing .NET/Rust behavior around deterministic compaction triggering and successful completion events.

Changes:

• Removes skips from compaction E2E tests.
• Registers event waiters before triggering compaction.
• Verifies successful compaction metadata and post-compaction session continuity.

Show a summary per file

File	Description
python/e2e/test_compaction_e2e.py	Re-enables Python compaction E2E and waits for start/successful complete events.
nodejs/test/e2e/compaction.e2e.test.ts	Re-enables Node.js compaction E2E and adds successful completion assertions.
go/internal/e2e/compaction_e2e_test.go	Re-enables Go compaction E2E and adds channel-based event waiting/assertions.

Copilot's findings

Comments suppressed due to low confidence (2)

nodejs/test/e2e/compaction.e2e.test.ts:65

• The stabilized .NET and Rust compaction E2Es compare these summary tags case-insensitively (dotnet/test/E2E/CompactionE2ETests.cs:56-58, rust/tests/e2e/compaction.rs:84-87). Keeping these checks case-sensitive makes this test more brittle if a valid summary varies tag casing.

        const summary = completeEvent.data.summaryContent ?? "";
        expect(summary).toContain("<overview>");
        expect(summary).toContain("<history>");
        expect(summary).toContain("<checkpoint_title>");

go/internal/e2e/compaction_e2e_test.go:117

• The stabilized .NET and Rust compaction E2Es compare these summary tags case-insensitively (dotnet/test/E2E/CompactionE2ETests.cs:56-58, rust/tests/e2e/compaction.rs:84-87). Keeping the Go checks case-sensitive makes this test more brittle if a valid summary varies tag casing.

		if !strings.Contains(summary, "<overview>") {
			t.Errorf("Expected summary to contain <overview>, got: %q", summary)
		}
		if !strings.Contains(summary, "<history>") {
			t.Errorf("Expected summary to contain <history>, got: %q", summary)

• Files reviewed: 3/3 changed files
• Comments generated: 3

[github-actions]

Cross-SDK Consistency Review ✅

This PR re-enables the compaction E2E tests in Node.js, Python, and Go using the same deterministic pattern already established in the .NET and Rust reference implementations.

All 5 SDKs are now consistent on:

• Registering event waiters before sending the prompts that trigger compaction (no race condition)
• Filtering for success == true on compaction_complete events to tolerate transient failures
• 60-second compaction timeout constant
• Identical assertion set: conversationTokens > 0, compactionTokensUsed.inputTokens > 0, summary XML tags (<overview>, <history>, <checkpoint_title>), and the "kaedrith"/"dragon" context-preservation check

No cross-SDK consistency issues found.

Generated by SDK Consistency Review Agent for issue #1314 · ● 457.4K · ◷