feat(web): add Playwright E2E test infrastructure

[YoungY620]

Related Issue

N/A — new test infrastructure, no prior issue.

Description

Establish end-to-end testing infrastructure for the Kimi Web frontend to catch regression bugs early when frontend code changes. Tests run against a real backend using the _scripted_echo provider for deterministic, LLM-free execution — no mocks, no flaky network calls.

What's included

• web/playwright.config.ts — Chromium project, baseURL 127.0.0.1:5494, on-first-retry traces
• Shared fixtures (web/tests/e2e/helpers/fixtures.ts):
  • createSession() — creates a session via the UI dialog
  • apiTracker — intercepts /api/** calls for request counting
  • consoleErrors — captures console.error and pageerror events
  • sendMessage() / waitForResponse() — helper functions
• 6 spec files, 16 test cases:
  • session-lifecycle — create → send message → archive → delete
  • websocket-streaming — progressive response, multi-turn message order
  • session-switching — cross-session content isolation, event leak guard
  • refresh-resilience — page refresh recovery, API call storm detection
  • dynamic-bugs — console.error guard, slow network resilience, rapid send
  • navigation — URL ?session= sync on create/navigate/reload
• echo-scripts.txt — 50 rounds of text: DSL responses for _scripted_echo
• CI job (web-e2e in ci-kimi-cli.yml) — starts backend with _scripted_echo + config.toml, runs Playwright, uploads report/traces as artifacts
• npm scripts: test:e2e, test:e2e:ui, test:e2e:install

How to run locally

# 1. Build web assets into the Python package
make build-web

# 2. Start backend with _scripted_echo
mkdir -p /tmp/kimi-e2e
SCRIPTS_PATH=$(realpath web/tests/e2e/fixtures/echo-scripts.txt)
cat > /tmp/kimi-e2e/config.toml <<TOML
default_model = "scripted"
[models.scripted]
provider = "scripted_provider"
model = "scripted_echo"
max_context_size = 100000
[providers.scripted_provider]
type = "_scripted_echo"
base_url = ""
api_key = ""
[providers.scripted_provider.env]
KIMI_SCRIPTED_ECHO_SCRIPTS = "$SCRIPTS_PATH"
KIMI_SCRIPTED_ECHO_TRACE = "1"
TOML
KIMI_SHARE_DIR=/tmp/kimi-e2e uv run kimi web --no-open --dangerously-omit-auth

# 3. Run tests
cd web && npm run test:e2e

No backend changes

All tests use the existing _scripted_echo provider and text: DSL. No modifications to packages/kosong/ or the Python backend.

Checklist

• I have read the CONTRIBUTING document.
• I have linked the related issue, if any.
• I have added tests that prove my fix is effective or that my feature works.
• I have run make gen-changelog to update the changelog.
• I have run make gen-docs to update the user documentation.

---

[devin-ai-integration]

Devin Review found 2 potential issues.

View 4 additional findings in Devin Review.

[devin-ai-integration]

🟡 CI Playwright reporter "github" does not produce the playwright-report/ directory that the upload step expects

In web/playwright.config.ts:9, the reporter is set to "github" when process.env.CI is truthy. The "github" reporter only emits inline annotations to stdout—it does not generate a playwright-report/ directory. However, the CI workflow at .github/workflows/ci-kimi-cli.yml:335-338 uploads web/playwright-report/ with if: always(), clearly intending to preserve the HTML report for debugging failures. Because the directory is never created in CI, the artifact upload is effectively a no-op (the default if-no-files-found is warn), so no report will ever be available when tests fail in CI—defeating the purpose of the if: always() guard.

Suggested change

	reporter: process.env.CI ? "github" : "html",
	reporter: process.env.CI ? [["github"], ["html"]] : "html",

---

Was this helpful? React with 👍 or 👎 to provide feedback.

[devin-ai-integration]

🟡 Health-check loop exits successfully even when the backend never starts

At .github/workflows/ci-kimi-cli.yml:326, the health-check loop for i in $(seq 1 30); do curl -sf … && break; sleep 1; done silently succeeds even if all 30 curl attempts fail, because the loop's last command is sleep 1 (exit code 0). The workflow then proceeds to run Playwright tests against a server that isn't running, producing confusing "connection refused" errors instead of a clear "backend failed to start" message. Adding a final check (e.g. curl -sf http://127.0.0.1:5494/healthz after the loop) would make the step fail fast with a clear diagnosis.

Suggested change

	for i in $(seq 1 30); do curl -sf http://127.0.0.1:5494/healthz && break; sleep 1; done
	for i in $(seq 1 30); do curl -sf http://127.0.0.1:5494/healthz && break; sleep 1; done
	curl -sf http://127.0.0.1:5494/healthz || { echo "Backend failed to start after 30s"; exit 1; }

---

Was this helpful? React with 👍 or 👎 to provide feedback.