Expand agent integration coverage and fix discovered regressions

[Kai Tao (vanzue)]

Summary

• add deterministic ACP stdio fixtures and end-to-end coverage for tool transcripts, clarification requests, session configuration, model titles, tab action directions, /new replacement, popup filtering, model lifecycle, prompt history, autofix, WSL shell context, and BYOK Credential Manager resolution
• add release checklist cases C278-C284 and document the new protocol experience suite
• fix input-dialog MouseUp handling so single-click transient selection clears while double-click word selection preserves multi-click history
• stabilize physical input, paste isolation, redock, and report-result test infrastructure discovered by the full regression run

Product regression attribution

The only confirmed product regression found by the full integration investigation was input-dialog double-click word selection. It was introduced by #624 when input-focus restoration cleared the multi-click record on the first MouseUp. This PR fixes that path and adds a focused Rust regression test.

The other investigated interaction failures were test geometry, foreground ownership, inherited navigation state, or an incomplete redock oracle rather than additional product regressions.

Scope

Hook installer behavior and hook-specific tests are intentionally unchanged from current main and are outside this PR.

Validation

• WTA after removing hook-specific changes: 1694 passed / 0 failed
• affected ACP model sync, protocol experience, and BYOK E2E suites: 9 passed / 0 failed / 0 skipped
• focused mouse suite: 4 passed / 0 failed / 2 skipped; the two physical-input cases skipped because the interactive desktop could not be acquired safely
• WSL 2 Ubuntu 24.04: Linux/bash autofix and ShellType coverage passed
• changed PowerShell files parse successfully; C278-C284 are automated in the generated release report

Full-suite investigation

The last complete 255-test run before the focused fixes reported 226 passed / 17 failed / 12 skipped. Every failure was investigated: one pre-existing product regression is fixed here, the interaction/redock test defects were corrected, legacy oracle/state cases were isolated, and three OpenCode live-event cases remain dependent on an external endpoint that returned service unavailable. Focused suites covering the changed paths pass; a newer full 255-test report has not been claimed.

[Copilot]

Pull request overview

Expands Intelligent Terminal’s agent-focused E2E coverage by adding deterministic ACP fixtures and new end-to-end protocol experience tests, while tightening existing suites (prompt history, autofix, model lifecycle/sync, BYOK) and updating documentation/release checklist coverage to reflect the new automated cases.

Changes:

• Add a new deterministic ACP “protocol experience” E2E suite plus a stdio interaction agent fixture to validate transcripts/tools, clarification requests, session config, model title confirmation, tab action direction hints, and /new session replacement behavior.
• Extend BYOK E2E to validate Credential Manager API key resolution (including authorization header verification via the OpenAI-compatible mock server).
• Reduce flakiness in existing E2E tests (prompt history uniqueness size, autofix timing, picker/title assertions) and document new automated checklist items.

Reviewed changes

Copilot reviewed 12 out of 12 changed files in this pull request and generated 2 comments.

Show a summary per file

File	Description
test/e2e/tests/Feature.PromptHistory.Tests.ps1	Shortens per-test GUID markers to reduce UI/test fragility while keeping uniqueness.
test/e2e/tests/Feature.ByokProvider.Tests.ps1	Adds Credential Manager injection via P/Invoke and asserts BYOK auth headers are passed end-to-end.
test/e2e/tests/Feature.AutofixPane.Tests.ps1	Adjusts autofix waits/dismiss loop to reduce timing-related flakes.
test/e2e/tests/Feature.AgentProtocolExperience.Tests.ps1	New end-to-end ACP protocol experience suite using a deterministic stdio fixture.
test/e2e/tests/Feature.AgentPopup.Tests.ps1	Adds substring-filtering coverage for slash command search ranking/filtering.
test/e2e/tests/Feature.AgentModelSync.Tests.ps1	Updates model picker assertion strategy for ACP model synchronization validation.
test/e2e/tests/Feature.AgentModelLifecycle.Tests.ps1	Simplifies settings-driven model restart/reconnect coverage using the model-switch fixture.
test/e2e/README.md	Documents the new protocol experience suite in the E2E suite list.
test/e2e/fixtures/Mock-OpenAIChatServer.ps1	Extends mock server to validate expected Authorization bearer token.
test/e2e/fixtures/Mock-AcpModelSwitchAgent.ps1	Enhances fixture to track and report the current model via config option updates/results.
test/e2e/fixtures/Mock-AcpInteractionAgent.ps1	New deterministic ACP interaction agent fixture (tools, input modal, config, MCP calls, session lifecycle).
doc/release-check-list.md	Adds new automated E2E/UT checklist items C278–C284 tied to the added coverage.

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

[Copilot]

Pull request overview

Copilot reviewed 12 out of 12 changed files in this pull request and generated no new comments.

Suppressed comments (2)

test/e2e/tests/Feature.AgentProtocolExperience.Tests.ps1:21

• $command embeds $script:fixture and $script:requestLog without quoting/escaping. If either path contains spaces or special characters, WTA’s process launch will fail and the test will be flaky. Prefer quoting the -File and -LogPath arguments (or using an encoded command like other fixtures do).

        $command = "pwsh -NoProfile -File $script:fixture -LogPath $script:requestLog"

test/e2e/tests/Feature.AgentModelSync.Tests.ps1:37

• Hard-coding '(?m)^┌Model' makes the test depend on an English literal that may differ under localization and deviates from existing ItE2E conventions for localized UI assertions. Use Get-WtaLocalizedTextRegex('model_picker.title') (with the existing en-US fallback) to assert the picker is open, as done elsewhere (e.g. test/e2e/tests/Feature.AgentPaneInteraction.Tests.ps1:135-137).

        Assert-AgentPaneText -App $script:app -Pattern '(?m)^┌Model' -TimeoutSec 10

[Copilot]

Pull request overview

Copilot reviewed 17 out of 17 changed files in this pull request and generated 2 comments.

Suppressed comments (1)

test/e2e/tests/Feature.AgentModelSync.Tests.ps1:37

• This assertion hard-codes an English "Model" title using box-drawing characters, which makes the test locale-dependent and inconsistent with the surrounding E2E convention of matching localized strings via Get-WtaLocalizedTextRegex (see Feature.AgentPaneInteraction.Tests.ps1).

        Assert-AgentPaneText -App $script:app -Pattern '(?m)^┌Model' -TimeoutSec 10

[Copilot]

Pull request overview

Copilot reviewed 23 out of 23 changed files in this pull request and generated 1 comment.

Suppressed comments (1)

test/e2e/tests/Feature.ByokProvider.Tests.ps1:65

• This test writes an API key into Windows Credential Manager with Persist = 2 (local-machine persistence). If the test run is aborted before cleanup, the credential can be left behind persistently. Using session persistence reduces the risk of leaving stray credentials on developer machines while still exercising CredWrite/CredRead.

Consider switching to Persist = 1 (session) for this fixture credential.

                CredentialBlob = handle.AddrOfPinnedObject(),
                Persist = 2,
                UserName = "Intelligent Terminal"

[Copilot]

Pull request overview

Copilot reviewed 23 out of 23 changed files in this pull request and generated no new comments.

Suppressed comments (2)

Previously missed (1) — in code that hasn't changed since the last review.

tools/wta/src/agent_hooks_installer.rs:2388

• preserve_copilot_registration returns false on read/parse failures without recording anything in messages, which makes locked-directory uninstall paths much harder to diagnose (and can silently prevent later restore/retry behavior). Consider pushing an explicit message for these failure cases before returning.

This issue also appears on line 2433 of the same file.

    let Ok(text) = fs::read_to_string(&config_path) else {
        return false;
    };
    let Ok(config) = serde_json::from_str::<Value>(&strip_jsonc_line_comments(&text)) else {
        return false;

tools/wta/src/agent_hooks_installer.rs:2437

• restore_deferred_copilot_registration bails out early on deferred-file read/parse errors without adding any diagnostic to messages, so a failed restore after a locked uninstall can be silent. Adding a message on these failure paths would make troubleshooting and telemetry much clearer.

    let Ok(deferred_text) = fs::read_to_string(&deferred_path) else {
        return false;
    };
    let Ok(entry) = serde_json::from_str::<Value>(&deferred_text) else {
        return false;

[Copilot]

Pull request overview

Copilot reviewed 23 out of 23 changed files in this pull request and generated 1 comment.

[Copilot]

Pull request overview

Copilot reviewed 23 out of 23 changed files in this pull request and generated no new comments.

Suppressed comments (1)

Previously missed (1) — in code that hasn't changed since the last review.

tools/wta/src/agent_hooks_installer.rs:2487

• restore_deferred_copilot_registration will write whatever JSON is present in the deferred snapshot into ~/.copilot/config.json without validating that it actually refers to the expected Copilot plugin (PLUGIN_NAME/MARKETPLACE_NAME). If the deferred file is stale/corrupted (or manually edited), this can inject an unrelated entry into Copilot’s config.

Validate the snapshot’s name and marketplace before applying it (and consider deleting the deferred file if it doesn’t match) to keep the restore path bounded to the intended plugin.

    let deferred_path = copilot_deferred_registration_path(home);
    let Ok(deferred_text) = fs::read_to_string(&deferred_path) else {
        return false;
    };
    let Ok(entry) = serde_json::from_str::<Value>(&deferred_text) else {

[Copilot]

Pull request overview

Copilot reviewed 19 out of 19 changed files in this pull request and generated no new comments.