ma-agents 3.5.0 → 3.5.1

package/lib/bmad-cache/gds/website/src/styles/custom.css

@@ -1,6 +1,6 @@
 /**
  * BMad Game Dev Studio Documentation - Custom Styles for Starlight
- * Electric Blue theme optimized for dark mode
+ * Dark theme matching bmadcode.com Ghost blog
  *
  * CSS Variable Mapping:
  * Docusaurus → Starlight
@@ -9,6 +9,9 @@
  * --ifm-font-color-base → --sl-color-text
  */
 
+/* Google Fonts - match Ghost blog typography */
+@import url('https://fonts.googleapis.com/css2?family=Inter:wght@400;500;600;700&family=Space+Grotesk:wght@500;600;700&family=JetBrains+Mono:wght@400;500&display=swap');
+
 /* ============================================
    COLOR PALETTE - Light Mode
    ============================================ */
@@ -19,10 +22,10 @@
   /* Full-width content - override Starlight's default 45rem/67.5rem */
   --sl-content-width: 100%;
 
-  /* Primary accent colors - purple to match BMAD */
-  --sl-color-accent-low: #e0e0ff;
-  --sl-color-accent: #8C8CFF;
-  --sl-color-accent-high: #4141FF;
+  /* Primary accent colors - blue to match Ghost blog */
+  --sl-color-accent-low: #dbeafe;
+  --sl-color-accent: #2563eb;
+  --sl-color-accent-high: #1d4ed8;
 
   /* Text colors */
   --sl-color-white: #1e293b;
@@ -35,13 +38,14 @@
   --sl-color-black: #f8fafc;
 
   /* Font settings */
-  --sl-font: system-ui, -apple-system, BlinkMacSystemFont, 'Segoe UI', Roboto, 'Helvetica Neue',
+  --sl-font: 'Inter', system-ui, -apple-system, BlinkMacSystemFont, 'Segoe UI', Roboto, 'Helvetica Neue',
     Arial, sans-serif;
+  --sl-font-mono: 'JetBrains Mono', 'Fira Code', ui-monospace, monospace;
   --sl-text-base: 1rem;
   --sl-line-height: 1.7;
 
   /* Code highlighting */
-  --sl-color-bg-inline-code: rgba(140, 140, 255, 0.1);
+  --sl-color-bg-inline-code: rgba(59, 130, 246, 0.08);
 }
 
 /* ============================================
@@ -51,35 +55,49 @@
   /* Full-width content - override Starlight's default */
   --sl-content-width: 100%;
 
-  /* Primary accent colors - purple to match BMAD */
-  --sl-color-accent-low: #2a2a5a;
-  --sl-color-accent: #8C8CFF;
-  --sl-color-accent-high: #B9B9FF;
+  /* Primary accent colors - blue to match Ghost blog */
+  --sl-color-accent-low: rgba(59, 130, 246, 0.12);
+  --sl-color-accent: #3b82f6;
+  --sl-color-accent-high: #60a5fa;
 
   /* Background colors */
-  --sl-color-bg: #1b1b1d;
-  --sl-color-bg-nav: #1b1b1d;
-  --sl-color-bg-sidebar: #1b1b1d;
-  --sl-color-hairline-light: rgba(140, 140, 255, 0.1);
-  --sl-color-hairline: rgba(140, 140, 255, 0.15);
+  --sl-color-bg: #0a0a0a;
+  --sl-color-bg-nav: #0a0a0a;
+  --sl-color-bg-sidebar: #0a0a0a;
+  --sl-color-hairline-light: rgba(255, 255, 255, 0.06);
+  --sl-color-hairline: #262626;
 
   /* Text colors */
-  --sl-color-white: #f8fafc;
+  --sl-color-white: #fafafa;
   --sl-color-gray-1: #e2e8f0;
-  --sl-color-gray-2: #cbd5e1;
+  --sl-color-gray-2: #a1a1a1;
   --sl-color-gray-3: #94a3b8;
-  --sl-color-gray-4: #64748b;
+  --sl-color-gray-4: #666666;
   --sl-color-gray-5: #475569;
-  --sl-color-gray-6: #334155;
-  --sl-color-black: #1b1b1d;
+  --sl-color-gray-6: #262626;
+  --sl-color-black: #0a0a0a;
 
   /* Code highlighting */
-  --sl-color-bg-inline-code: rgba(140, 140, 255, 0.15);
+  --sl-color-bg-inline-code: rgba(59, 130, 246, 0.15);
 }
 
 /* ============================================
    TYPOGRAPHY
    ============================================ */
+
+/* Space Grotesk for all headings - match Ghost blog */
+.sl-markdown-content h1,
+.sl-markdown-content h2,
+.sl-markdown-content h3,
+.sl-markdown-content h4,
+.sl-markdown-content h5,
+.sl-markdown-content h6,
+.site-title,
+starlight-toc h2 {
+  font-family: 'Space Grotesk', 'Inter', system-ui, sans-serif;
+  letter-spacing: -0.02em;
+}
+
 .sl-markdown-content h1 {
   margin-bottom: 1.5rem;
 }
@@ -138,14 +156,14 @@
 
 /* Active state - thin left accent bar */
 .sidebar-content a[aria-current='page'] {
-  background-color: rgba(140, 140, 255, 0.08);
+  background-color: rgba(59, 130, 246, 0.08);
   color: var(--sl-color-accent);
   border-left-color: var(--sl-color-accent);
   font-weight: 600;
 }
 
 :root[data-theme='dark'] .sidebar-content a[aria-current='page'] {
-  background-color: rgba(140, 140, 255, 0.1);
+  background-color: rgba(59, 130, 246, 0.1);
   color: var(--sl-color-accent-high);
   border-left-color: var(--sl-color-accent);
 }
@@ -216,7 +234,8 @@ header.header .header.sl-flex {
 }
 
 :root[data-theme='dark'] header.header {
-  box-shadow: 0 1px 3px rgba(0, 0, 0, 0.3);
+  box-shadow: none;
+  border-bottom: 1px solid #262626;
 }
 
 .site-title {
@@ -265,20 +284,20 @@ header.header .header.sl-flex {
 .card:hover {
   transform: translateY(-3px);
   border-color: var(--sl-color-accent);
-  box-shadow: 0 8px 24px rgba(140, 140, 255, 0.15);
+  box-shadow: 0 8px 24px rgba(59, 130, 246, 0.15);
 }
 
 :root[data-theme='dark'] .card {
-  background: linear-gradient(145deg, rgba(30, 41, 59, 0.6), rgba(15, 23, 42, 0.8));
-  border-color: rgba(140, 140, 255, 0.2);
+  background: #1a1a1a;
+  border-color: #262626;
   box-shadow: 0 2px 8px rgba(0, 0, 0, 0.3);
 }
 
 :root[data-theme='dark'] .card:hover {
-  border-color: rgba(140, 140, 255, 0.5);
+  border-color: #3b82f6;
   box-shadow:
-    0 8px 32px rgba(140, 140, 255, 0.2),
-    0 0 0 1px rgba(140, 140, 255, 0.1);
+    0 8px 32px #262626,
+    0 0 0 1px rgba(59, 130, 246, 0.1);
 }
 
 /* Starlight card grid */
@@ -297,11 +316,11 @@ header.header .header.sl-flex {
 }
 
 :root[data-theme='dark'] .sl-link-card {
-  border-color: rgba(140, 140, 255, 0.2);
+  border-color: #262626;
 }
 
 :root[data-theme='dark'] .sl-link-card:hover {
-  border-color: rgba(140, 140, 255, 0.5);
+  border-color: #3b82f6;
 }
 
 /* ============================================
@@ -341,21 +360,21 @@ table {
 }
 
 :root[data-theme='dark'] table {
-  border-color: rgba(140, 140, 255, 0.1);
+  border-color: rgba(59, 130, 246, 0.1);
 }
 
 :root[data-theme='dark'] table th {
-  background-color: rgba(140, 140, 255, 0.05);
+  background-color: rgba(59, 130, 246, 0.05);
 }
 
 :root[data-theme='dark'] table tr:nth-child(2n) {
-  background-color: rgba(140, 140, 255, 0.02);
+  background-color: rgba(255, 255, 255, 0.02);
 }
 
 /* Blockquotes */
 blockquote {
   border-left-color: var(--sl-color-accent);
-  background-color: rgba(140, 140, 255, 0.05);
+  background-color: rgba(59, 130, 246, 0.05);
   border-radius: 0 8px 8px 0;
   padding: 1rem 1.25rem;
 }
@@ -392,19 +411,19 @@ blockquote {
 
 /* Note aside */
 .starlight-aside--note {
-  background-color: rgba(140, 140, 255, 0.08);
+  background-color: rgba(59, 130, 246, 0.08);
 }
 
 .starlight-aside--note .starlight-aside__title {
-  color: #8C8CFF;
+  color: #3b82f6;
 }
 
 :root[data-theme='dark'] .starlight-aside--note {
-  background-color: rgba(140, 140, 255, 0.12);
+  background-color: rgba(59, 130, 246, 0.12);
 }
 
 :root[data-theme='dark'] .starlight-aside--note .starlight-aside__title {
-  color: #8C8CFF;
+  color: #3b82f6;
 }
 
 /* Caution aside */

package/lib/bmad-cache/tea/README.md

@@ -1,5 +1,8 @@
 # Test Architect (TEA)
 
+[![Python Version](https://img.shields.io/badge/python-%3E%3D3.10-blue?logo=python&logoColor=white)](https://www.python.org)
+[![uv](https://img.shields.io/badge/uv-package%20manager-blueviolet?logo=uv)](https://docs.astral.sh/uv/)
+
 TEA (Test Engineering Architect) is a standalone BMAD module that delivers risk-based test strategy, test automation guidance, and release gate decisions. It provides a single expert agent (Murat, Master Test Architect and Quality Advisor) and nine workflows spanning Teach Me Testing (TEA Academy), framework setup, test design, ATDD, automation, traceability, NFR assessment, CI guidance, and test review.
 
 Docs: <https://bmad-code-org.github.io/bmad-method-test-architecture-enterprise/>
@@ -379,6 +382,17 @@ After releasing:
 
 ---
 
+## Community
+
+- [Discord](https://discord.gg/gk8jAdXWmj) — Get help, share ideas, collaborate
+- [YouTube](https://youtube.com/@BMadCode) — Tutorials, master class, and more
+- [X / Twitter](https://x.com/BMadCode)
+- [Website](https://bmadcode.com)
+
+## Support BMad
+
+BMad is free for everyone and always will be. Star this repo, [buy me a coffee](https://buymeacoffee.com/bmad), or email [contact@bmadcode.com](mailto:contact@bmadcode.com) for corporate sponsorship.
+
 ## License
 
 See `LICENSE`.

package/lib/bmad-cache/tea/_git_preserved/packed-refs

@@ -1,2 +1,2 @@
 # pack-refs with: peeled fully-peeled sorted 
-bd2ac72a4def9e8b0881407827293d66d9b009d5 refs/remotes/origin/main
+05472f9f4ad0a3aeeff445252bd4dd342667f34a refs/remotes/origin/main

package/lib/bmad-cache/tea/_git_preserved/refs/heads/main

@@ -1 +1 @@
-bd2ac72a4def9e8b0881407827293d66d9b009d5
+05472f9f4ad0a3aeeff445252bd4dd342667f34a

package/lib/bmad-cache/tea/_git_preserved/shallow

@@ -1 +1 @@
-bd2ac72a4def9e8b0881407827293d66d9b009d5
+05472f9f4ad0a3aeeff445252bd4dd342667f34a

package/lib/bmad-cache/tea/docs/index.md

@@ -1,20 +1,10 @@
 ---
-title: Test Architect (TEA)
-description: Risk-based testing workflows, automation guidance, and release gates for BMad Method
-template: splash
-hero:
-  title: Test Architect (TEA)
-  tagline: Risk-based testing workflows, automation guidance, and release gates for BMad Method
-  actions:
-    - text: Tutorial
-      link: https://bmad-code-org.github.io/bmad-method-test-architecture-enterprise/tutorials/tea-lite-quickstart/
-      icon: right-arrow
-      variant: primary
-    - text: View on GitHub
-      link: https://github.com/bmad-code-org/bmad-method-test-architecture-enterprise
-      icon: external
+title: Welcome
+description: Test Architect (TEA) - Risk-based testing workflows, automation guidance, and release gates for BMad Method
 ---
 
+# Test Architect (TEA)
+
 ## What is TEA?
 
 TEA (Test Engineering Architect) is a BMAD module for testing strategy and automation. It provides nine workflows covering learning, setup, design, automation, review, and release gates.

package/lib/bmad-cache/tea/package-lock.json

@@ -1,12 +1,12 @@
 {
   "name": "bmad-method-test-architecture-enterprise",
-  "version": "1.7.3",
+  "version": "1.9.1",
   "lockfileVersion": 3,
   "requires": true,
   "packages": {
     "": {
       "name": "bmad-method-test-architecture-enterprise",
-      "version": "1.7.3",
+      "version": "1.9.1",
       "license": "MIT",
       "dependencies": {
         "@clack/prompts": "^0.11.0",

package/lib/bmad-cache/tea/package.json

@@ -1,7 +1,7 @@
 {
   "$schema": "https://json.schemastore.org/package.json",
   "name": "bmad-method-test-architecture-enterprise",
-  "version": "1.7.3",
+  "version": "1.9.1",
   "private": true,
   "description": "Master Test Architect for quality strategy, test automation, and release gates",
   "keywords": [

package/lib/bmad-cache/tea/src/agents/bmad-tea/SKILL.md

@@ -30,8 +30,8 @@ Blends data with gut instinct. "Strong opinions, weakly held" is their mantra. S
 
 ## Critical Actions
 
-- Consult `{project-root}/_bmad/tea/agents/bmad-tea/resources/tea-index.csv` to select knowledge fragments under `knowledge/` and load only the files needed for the current task
-- Load the referenced fragment(s) from `{project-root}/_bmad/tea/agents/bmad-tea/resources/knowledge/` before giving recommendations
+- Consult `./resources/tea-index.csv` to select knowledge fragments under `resources/knowledge/` and load only the files needed for the current task
+- Load the referenced fragment(s) from `./resources/knowledge/` before giving recommendations
 - Cross-check recommendations with the current official Playwright, Cypress, Pact, k6, pytest, JUnit, Go test, and CI platform documentation
 
 You must fully embody this persona so the user gets the best experience and help they need, therefore its important to remember you must not break character until the users dismisses this persona.
@@ -54,10 +54,10 @@ When you are in this persona and the user calls a skill, this persona must carry
 
 ## On Activation
 
-1. **Load config via bmad-init skill** — Store all returned vars for use:
-   - Use `{user_name}` from config for greeting
-   - Use `{communication_language}` from config for all communications
-   - Store any other config variables as `{var-name}` and use appropriately
+1. Load config from `{project-root}/_bmad/tea/config.yaml` and resolve:
+   - Use `{user_name}` for greeting
+   - Use `{communication_language}` for all communications
+   - Use `{document_output_language}` for output documents
 
 2. **Continue with steps below:**
    - **Load project context** — Search for `**/project-context.md`. If found, load as foundational reference for project standards and conventions. If not found, continue without it.

package/lib/bmad-cache/tea/src/agents/bmad-tea/resources/knowledge/playwright-cli.md

@@ -67,6 +67,8 @@ snapshot ref {role: "button", name: "Sign In"}
 
 **Evidence collection** — During `test-review`, TEA can capture screenshots, traces, and network logs as evidence without the overhead of a full MCP session.
 
+**Agent-side test debugging** — For existing failing Playwright tests, TEA should prefer Playwright's agent-facing debug loop over ad hoc manual reproduction: `npx playwright test --debug=cli` to step through the test in CLI mode (no GUI Inspector — designed for coding agents), then `npx playwright trace ...` to inspect the resulting trace artifact from the command line. The `--debug=cli` flag (Playwright 1.59+) lets agents attach, step through execution, and inspect page state without ever opening a browser window.
+
 ## How CLI Relates to Playwright Utils and API Testing
 
 CLI and playwright-utils are **complementary tools that work at different layers**:
@@ -101,11 +103,31 @@ test('GET /api/users returns user list', async ({ apiRequest }) => {
 });
 ```
 
-**For pure API testing** (no UI involved), CLI doesn't add much — there's no page to snapshot. The agent generates API tests directly from documentation, specs, or code analysis using `apiRequest` and `recurse` from playwright-utils.
+**For pure API testing** (no UI involved), `playwright-cli` browser commands (snapshot, screenshot, click) don't apply — there's no page. But **trace analysis is highly valuable**. Playwright captures full network traces for API tests (requests, responses, headers, timing), and the trace CLI lets the agent inspect them programmatically:
+
+```bash
+# API test fails in CI → open the trace artifact
+npx playwright trace open test-results/api-users/trace.zip
+
+# What HTTP call failed?
+npx playwright trace requests --failed
+# Output: #3  POST /api/users  → 422  12ms
+
+# Full request/response details (headers, body, timing)
+npx playwright trace request 3
 
-**For E2E testing**, CLI shines — it snapshots the page to get accurate selectors, observes network calls to understand the API contract, and captures auth flows via `state-save` that inform how tests use `auth-session`.
+# What assertion failed and why?
+npx playwright trace errors
 
-**Bottom line:** CLI helps the agent _write better tests_. Playwright-utils helps those tests _run reliably_.
+# Done
+npx playwright trace close
+```
+
+This gives the agent the full HTTP conversation — wrong payload, expired auth token, schema mismatch, upstream 5xx — without a human opening UI mode. The agent generates API tests directly from documentation, specs, or code analysis using `apiRequest` and `recurse` from playwright-utils, and uses trace analysis to diagnose failures.
+
+**For E2E testing**, CLI shines at both ends — browser commands (snapshot, screenshot) during test generation, and trace analysis (actions, snapshots, requests) during debugging.
+
+**Bottom line:** CLI helps the agent _write better tests_. Playwright-utils helps those tests _run reliably_. Trace analysis helps the agent _fix them when they break_.
 
 ## Session Isolation
 
@@ -125,6 +147,99 @@ For parallel safety (multiple agents on the same machine), append a unique suffi
 playwright-cli -s=tea-explore-<timestamp> open https://app.com
 ```
 
+## Autonomous Trace Investigation (Playwright 1.59+)
+
+For generated tests that already exist and are failing, Playwright 1.59 introduced CLI-native debugging and trace analysis designed specifically for AI agents. Instead of downloading traces and opening the GUI Trace Viewer, agents can now consume the entire trace context directly from the command line.
+
+### Debug a Failing Test (CLI Mode)
+
+```bash
+# Start the test in CLI debug mode — no GUI Inspector, agent-friendly output
+npx playwright test --debug=cli
+playwright-cli attach <session-id>
+playwright-cli --session <session-id> step-over
+```
+
+With `--debug=cli`, the agent can:
+
+- Step through test execution in real-time
+- Inspect the page's HTML source at each step
+- Review network calls and console logs at the moment of failure
+- Capture before/after snapshots without opening a browser
+
+### Investigate a Trace Artifact
+
+```bash
+# Open a trace from CI or local runs — this starts a session
+npx playwright trace open test-results/<run>/trace.zip
+
+# List all actions as a numbered tree (# column = 1-based ordinal)
+npx playwright trace actions
+# Output: #  Time     Action                Duration
+#         1  0:00.00  navigate(...)         120ms
+#         2  0:00.12  fill(#email, ...)     45ms
+#         ...
+#         9  0:01.50  expect(toBeVisible)   ✗ 30s
+
+# Filter to failing assertions
+npx playwright trace actions --grep="expect"
+
+# Drill into action #9 (the ordinal from the list above)
+npx playwright trace action 9
+
+# See the page snapshot after that action (valid: before | input | after)
+npx playwright trace snapshot 9 --name after
+
+# Other useful subcommands
+npx playwright trace errors                  # errors with stack traces
+npx playwright trace requests --failed       # failed network requests
+npx playwright trace console --errors-only   # console errors
+
+# Close when done (removes extracted data)
+npx playwright trace close
+```
+
+### Autonomous Diagnostic Loop
+
+When TEA encounters a failing test in healing/review mode, the recommended investigation flow is:
+
+1. **Run with `--debug=cli`** to step through the failure and identify the failing action
+2. **Get a trace artifact** — configure `trace: 'retain-on-failure'` in `playwright.config.ts` (recommended), add `--trace=retain-on-failure` to the test run, or use an existing CI trace artifact. For `playwright-cli` sessions (not `--debug=cli`), use `tracing-start` / `tracing-stop` instead.
+3. **Filter to assertions** (`trace actions --grep="expect"`) to find the failure point
+4. **Inspect the snapshot** (`trace snapshot <n> --name after`) to see exact page state at failure
+5. **Analyze network/console** to rule out backend issues or timing problems
+6. **Propose a fix** — updated locator, added wait, or flagged flake for human review
+
+This reduces Mean Time to Repair (MTTR) by giving the agent full failure context rather than just an error message.
+
+### When to Use Each Tool
+
+- `playwright-cli` session commands remain the best lightweight tool for page exploration and selector verification.
+- `npx playwright test --debug=cli` is better for stepping through an already-written failing test (agent-native, no GUI).
+- `npx playwright trace ...` is better for understanding flakes and assertion failures from saved artifacts.
+
+If your environment exposes the Playwright dashboard or bound-browser flow, it can help humans inspect what an agent is doing in the background, but TEA should treat that as optional observability rather than a hard dependency.
+
+### Binding a Browser for Agent Inspection (`browser.bind()`)
+
+Playwright 1.59 added `browser.bind()` — a programmatic API that makes a running browser instance available to `playwright-cli` and MCP clients. This is the bridge between "a test is running" and "an agent can see what the test sees."
+
+```typescript
+// In a test or fixture: bind the browser so playwright-cli can attach
+const { endpoint } = await browser.bind('my-debug-session', {
+  workspaceDir: process.cwd(),
+});
+// Now: playwright-cli attach my-debug-session
+```
+
+**When TEA uses this:**
+
+- **Debugging a complex E2E failure** — A test fixture calls `browser.bind()` before the failing scenario, then TEA runs `playwright-cli attach` to inspect live page state, network, and console without re-running the test from scratch.
+- **Bridging CLI and MCP** — A bound browser is accessible to both `playwright-cli` and `@playwright/mcp`. TEA's `auto` mode can start with lightweight CLI inspection and escalate to MCP if richer introspection is needed, all against the same browser instance.
+- **CI artifact enhancement** — A CI helper can bind the browser during test runs, letting a post-failure agent attach and investigate before the process exits.
+
+Call `await browser.unbind()` when done to release the session (async — must be awaited).
+
 ## Command Quick Reference
 
 | What you want to do       | Command                                          |
@@ -152,7 +267,7 @@ playwright-cli -s=tea-explore-<timestamp> open https://app.com
 | "Verify this selector exists"         | CLI  | Single check, minimal tokens       |
 | "Capture a screenshot for evidence"   | CLI  | Stateless capture                  |
 | "Walk through a multi-step wizard"    | MCP  | State carries across steps         |
-| "Debug why this test fails" (healing) | MCP  | Needs rich DOM introspection       |
+| "Debug why this test fails" (healing) | CLI  | `--debug=cli` + trace analysis     |
 | "Record a drag-and-drop flow"         | MCP  | Complex interaction semantics      |
 
 ## Related Fragments

package/lib/bmad-cache/tea/src/agents/bmad-tea/resources/knowledge/playwright-config.md

@@ -67,7 +67,7 @@ export const baseConfig = defineConfig({
   use: {
     actionTimeout: 15000,
     navigationTimeout: 30000,
-    trace: 'on-first-retry',
+    trace: 'retain-on-failure-and-retries',
     screenshot: 'only-on-failure',
     video: 'retain-on-failure',
   },
@@ -92,6 +92,9 @@ export default defineConfig({
   webServer: {
     command: 'npm run dev',
     url: 'http://localhost:3000',
+    wait: {
+      stdout: /ready|listening|localhost:/i,
+    },
     reuseExistingServer: !process.env.CI,
     timeout: 120000,
   },
@@ -272,8 +275,8 @@ export default defineConfig({
     // Video recording on failure + retry
     video: 'retain-on-failure',
 
-    // Trace recording on first retry (best debugging data)
-    trace: 'on-first-retry',
+    // Keep failed attempts and retries for flake analysis
+    trace: 'retain-on-failure-and-retries',
   },
 
   reporter: [
@@ -397,7 +400,8 @@ test('capture screenshot on specific error', async ({ page }) => {
 
 - `screenshot: 'only-on-failure'` saves space (not every test)
 - `video: 'retain-on-failure'` captures full flow on failures
-- `trace: 'on-first-retry'` provides deep debugging data (network, DOM, console)
+- `trace: 'retain-on-failure-and-retries'` keeps enough history to compare failing retries against passing runs
+- `webServer.wait` is better than startup sleeps when local servers print readiness to stdout/stderr
 - HTML report at `playwright-report/` (visual debugging)
 - JUnit XML at `test-results/results.xml` (CI integration)
 - CI uploads artifacts on failure with 30-day retention