Stabilize local agent auto model dump E2E

[wwwillchen]

Summary

• Pins code explorer off in the local-agent auto-model request dump test before sending the prompt.
• Prevents the serialized prompt/tool snapshot from flipping between code_search and explore_code based on code explorer indexing readiness.
• Records the E2E gotcha for future local-agent request-dump specs.

Verification

• npm run build
• npm run fmt
• npm run lint
• npm run lint:fix
• npm run ts
• PLAYWRIGHT_HTML_OPEN=never npm run e2e -- e2e-tests/local_agent_auto.spec.ts
• PLAYWRIGHT_RETRIES=0 PLAYWRIGHT_HTML_OPEN=never npm run e2e -- e2e-tests/local_agent_auto.spec.ts --repeat-each=5
• npm test

Failing run investigated: https://github.com/dyad-sh/dyad/actions/runs/28407790114

#skip-bugbot

---

Note

Low Risk
Test-only and documentation changes with no production code paths affected.

Overview
Stabilizes the local-agent auto model request-dump E2E by turning enableCodeExplorer off via set-user-settings and polling persisted settings before the [dump] prompt, so the serialized request snapshot does not race between code_search and explore_code when code-explorer indexing finishes at different times on CI vs locally.

Documents the same pattern in rules/e2e-testing.md for other local-agent request-dump specs that are not explicitly testing explore_code.

^{Reviewed by Cursor Bugbot for commit 600a84f. Bugbot is set up for automated code reviews on this repo. Configure here.}

[cursor]

Bugbot couldn't run - usage limit reached

Bugbot is counted against Cursor usage for this user or team, and this run hit a usage or spend limit.

A user or team admin can review and increase usage limits in the Cursor dashboard.

(requestId: serverGenReqId_4ed329e5-799a-48a8-91c8-338cf5d375a5)

[Copilot]

Pull request overview

This PR deflakes the “local-agent - auto model” E2E by removing a race where request-dump snapshots could alternate between code_search and explore_code depending on code-explorer indexing readiness.

Changes:

• Pins enableCodeExplorer: false via set-user-settings in the auto-model dump E2E and waits for the persisted setting before sending [dump].
• Documents the same stabilization pattern in rules/e2e-testing.md for other request-dump specs not explicitly testing explore_code.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated no comments.

File	Description
rules/e2e-testing.md	Adds a new E2E gotcha entry describing how to keep request-dump snapshots stable by pinning enableCodeExplorer off and polling persisted settings.
e2e-tests/local_agent_auto.spec.ts	Disables code explorer and polls po.settings.recordSettings().enableCodeExplorer before issuing the dump prompt to prevent snapshot/tool-list flakiness.

---

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

[dyad-assistant]

🔍 Dyadbot Code Review Summary

Verdict: ✅ YES - Ready to merge
Recommendation: ready

Small, well-scoped test stabilization PR. Two files changed (10 additions, 0 deletions) — all test/docs only, zero production code paths affected.

What it does: Pins enableCodeExplorer: false via IPC in the local-agent auto-model request-dump E2E test and polls until the setting persists before sending the [dump] prompt. This prevents the serialized tool snapshot from flipping between code_search and explore_code depending on whether code-explorer indexing finishes before or after the prompt is sent. Documents the pattern in rules/e2e-testing.md.

Correctness: The IPC call uses the validated set-user-settings channel (enforces UserSettingsSchema.partial()), and the expect.poll() pattern correctly waits for persistence before proceeding. The setting is pinned before po.importApp(), which is the right sequencing. The pattern matches existing specs (local_agent_code_search.spec.ts, local_agent_explore_code.spec.ts) exactly.

Security: No new IPC channels or renderer-side capabilities introduced. The test calls an existing, validated endpoint with a well-typed payload.

Code Health: Follows the established convention for similar local-agent E2E specs. The documentation bullet is clear, actionable, and placed adjacent to the related code_search coverage bullet.

UX: No user-facing changes.

✅ No issues found by persona-based review.

---

Generated by Dyadbot persona-based code review

[github-actions]

🎭 Playwright Test Results

✅ All tests passed!

OS	Passed	Flaky	Skipped
🍎 macOS	535	3	170
🪟 Windows	535	3	170

Total: 1070 tests passed (6 flaky) (340 skipped)

⚠️ Flaky Tests

🍎 macOS

• chat_image_generation.spec.ts > generate image from chat - full flow (passed after 1 retry)
• context_limit_banner.spec.ts > context limit banner shows 'running out' when near context limit (passed after 2 retries)
• setup_flow.spec.ts > Setup Flow > node.js install flow (passed after 1 retry)

🪟 Windows

• app_screenshot.spec.ts > captures an app screenshot after the first generated commit (passed after 1 retry)
• chat_input.spec.ts > send button disabled during pending proposal - reject (passed after 1 retry)
• concurrent_chat.spec.ts > concurrent chat (passed after 1 retry)

---

📊 View full report