Add AI demo script and resolve open design decisions

- Add docs/ai-demo-script.md with 15-min demo flow and backup test
- Resolve 10+ DECIDE items in PLAN.md (CI scope, AI config, screenshots,
  GitLab sync, robocop, browser default, etc.)
- Mark verified items: selectors, setup-uv version, MCP package

Author: HackXIt

PLAN.md

@@ -301,7 +301,7 @@ Note: `uv.lock` is NOT in .gitignore — it must be committed for reproducible i
 ### CURRENT ISSUES — Phase 1
 
 - [x] **DECIDED:** Repo created via `gh repo create` CLI (public, with description).
-- [ ] **DECIDE:** Should GitLab project be manually created or left for auto-sync?
+- [x] **DECIDED:** Left for auto-sync. `github-sync` CronJob auto-discovers all HackXIt repos and mirrors to `github-mirrors` group.
 - [x] **RESOLVED:** GitHub auth required `workflow` scope for pushing workflow files. Added via `gh auth refresh -s workflow`.
 
 ---
@@ -392,10 +392,10 @@ Must verify:
 
 ### CURRENT ISSUES — Phase 2
 
-- [ ] **VERIFY:** Does the `ghcr.io/va-h/devcontainers-features/uv:1` feature work reliably in Codespaces? If not, fall back to curl-based install in post-create.sh.
+- [ ] **VERIFY:** Does the `ghcr.io/va-h/devcontainers-features/uv:1` feature work reliably in Codespaces? Fallback already in post-create.sh.
 - [ ] **VERIFY:** Does `rfbrowser init chromium` complete within the Codespace timeout?
-- [ ] **VERIFY:** Are there Playwright system dependencies missing in the bookworm base image? May need `npx playwright install-deps` or `apt-get install` for missing libs.
-- [ ] **KNOWN ISSUE:** First Codespace build downloads ~250MB of Chromium binaries. On slow connections this can take several minutes. Consider: Codespace prebuild configuration to cache this.
+- [x] **MITIGATED:** Playwright system deps — added `npx playwright install-deps chromium` to post-create.sh.
+- [ ] **KNOWN ISSUE:** First Codespace build downloads ~250MB of Chromium binaries. Consider Codespace prebuild for caching.
 
 ---
 
@@ -519,7 +519,7 @@ Then look for a `chromium-*` directory inside the found path.
 
 ### CURRENT ISSUES — Phase 3
 
-- [ ] **DECIDE:** Should the script also check for `robocop` (linter) or only core runtime?
+- [x] **DECIDED:** Core runtime only. Robocop is a dev tool, not required for workshop tests to run.
 - [ ] **VERIFY:** Playwright browsers path detection works correctly in Codespaces (may use a non-standard path).
 - [ ] **VERIFY:** ANSI color codes render correctly in Codespaces terminal and Windows Terminal.
 
@@ -633,9 +633,9 @@ Example Test Case
 
 ### CURRENT ISSUES — Phase 4
 
-- [ ] **VERIFY:** SauceDemo CSS selectors are current (they occasionally change their markup).
+- [x] **VERIFIED:** SauceDemo CSS selectors are current — all 25 tests pass locally and in CI (March 2026).
 - [ ] **VERIFY:** Browser Library's `New Browser` keyword works headless in Codespaces (no display server).
-- [ ] **DECIDE:** Should example tests use `browser=chromium` explicitly or rely on default?
+- [x] **DECIDED:** Rely on default. `common.resource` uses `chromium` in `New Browser`, which is sufficient.
 - [ ] **KNOWN RISK:** SauceDemo could go down during the workshop. Prepare a fallback SUT (automationexercise.com, the-internet.herokuapp.com).
 
 ---
@@ -721,9 +721,9 @@ Deploy `results/` to GitHub Pages on main branch pushes:
 
 ### CURRENT ISSUES — Phase 5
 
-- [ ] **VERIFY:** `astral-sh/setup-uv@v7` is the latest stable version of the action.
-- [ ] **DECIDE:** Should CI run ALL tests or only student exercise tests? Running all on every PR may be noisy.
-- [ ] **DECIDE:** Should failed tests block PR merge or just report? For a workshop, reporting only is more friendly.
+- [x] **VERIFIED:** `astral-sh/setup-uv@v7` is latest major (v7.3.1 latest release, Feb 2026). No changes needed.
+- [x] **DECIDED:** CI runs ALL tests on every PR. This validates students don't break existing tests. 27 tests run in ~1 min.
+- [x] **DECIDED:** Failed tests do not block merge (no branch protection rules). CI reports status but workshop is learning-focused.
 - [ ] **VERIFY:** GitHub Actions free tier limits (2000 min/month for free accounts, 3000 min/month for Pro). Each run with Playwright install may take ~3-5 min.
 
 ---
@@ -790,9 +790,9 @@ This demonstrates the "AI-assisted test automation" angle.
 
 ### CURRENT ISSUES — Phase 6
 
-- [ ] **VERIFY:** `robotframework-mcp` package exists on PyPI and is compatible with current RF/Browser versions.
-- [ ] **DECIDE:** Should Claude Code / AI be pre-configured in the DevContainer, or shown as a demo only?
-- [ ] **DECIDE:** Should the repo have GitHub Copilot or Claude Code review PRs automatically via GitHub Actions?
+- [x] **VERIFIED:** `robotframework-mcp` exists but is SeleniumLibrary-based (not suitable). Using `rf-mcp` (0.30+, 71+ releases, mature). Added to dev deps.
+- [x] **DECIDED:** AI demo only (instructor shows Claude Code). Not pre-configured for students — adds complexity. `.claude/settings.json` exists for instructor use.
+- [x] **DECIDED:** No automatic AI PR review. Workshop is about learning, not automated gatekeeping. Instructor reviews manually.
 
 ---
 
@@ -909,8 +909,8 @@ Guide for the AI demo portion:
 
 ### CURRENT ISSUES — Phase 7
 
-- [ ] **DECIDE:** Should docs include screenshots? If yes, they need to be captured during Phase 11 pre-check.
-- [ ] **DECIDE:** How much RF theory vs hands-on in the cheatsheet? Students learn C-style, so comparison tables (C if/else → RF IF/ELSE) might be helpful.
+- [x] **DECIDED:** Screenshots are nice-to-have. Text instructions are sufficient; screenshots can be added during Phase 11 pre-check if time permits.
+- [x] **DECIDED:** C comparison tables already included in `docs/02-rf-syntax-cheatsheet.md`. Balance is good: minimal theory, heavy on examples.
 - [ ] **VERIFY:** Windows setup doc should be tested on an actual Windows machine or VM.
 
 ---
@@ -1250,8 +1250,8 @@ Phase 1 (Repo + pyproject.toml)
 - [x] Create `.claude/settings.json` with MCP server config
 - [x] Verify MCP package on PyPI — `robotframework-mcp` (SeleniumLibrary-based, not suitable); using `rf-mcp` (0.30+, mature)
 - [x] Add `rf-mcp` to `pyproject.toml` dev dependencies
-- [ ] Prepare AI demo script (natural language → RF test → run → refine)
-- [ ] Test the demo end-to-end
+- [x] Prepare AI demo script — `docs/ai-demo-script.md` with flow, prompts, backup test
+- [ ] Test the demo end-to-end (requires live Claude Code session)
 
 ### Phase 7: Documentation (Day 4-5)
 - [x] Create `README.md` (quick start, prerequisites, agenda, badges)

docs/ai-demo-script.md

@@ -0,0 +1,103 @@
+# AI-Assisted Testing Demo Script
+
+**Duration:** ~15 minutes
+**When:** 19:15 in the agenda
+
+## Setup
+
+- Have Claude Code (or Copilot) open in a Codespace terminal
+- Have the project README open for context
+
+## Demo Flow
+
+### 1. Natural Language to Test (3 min)
+
+**Say:** "Let me show you how AI can help write tests. I'll describe what I want in plain English."
+
+**Prompt to give Claude Code:**
+
+> Write a Robot Framework test that verifies products on SauceDemo can be sorted by name Z-to-A. The test should log in as standard_user, select the Z-A sort option, and verify the first product name starts with a letter later in the alphabet than the last product.
+
+**Expected output:** A complete `.robot` file with Settings, Variables, Test Cases, and Keywords sections.
+
+### 2. Review the Generated Test (3 min)
+
+**Say:** "Now let's review what the AI generated. Does it look correct?"
+
+Walk through:
+- Does it import the right resources?
+- Are the selectors correct for SauceDemo?
+- Does the sorting assertion make sense?
+- Are there any issues?
+
+**Key teaching moment:** AI gets the structure right but may use wrong selectors or flawed assertions. The tester must validate.
+
+### 3. Run the Test (2 min)
+
+```bash
+uv run robot tests/student_exercises/ai_generated_sort_test.robot
+```
+
+If it passes: "Great, but we still need to verify it's actually testing what we intended."
+
+If it fails: "This is normal! Let's debug and fix it together."
+
+### 4. Refine (3 min)
+
+If the test had issues, refine the prompt:
+
+> The sort dropdown on SauceDemo uses the CSS selector `.product_sort_container` and the Z-A option has value `za`. After sorting, verify the first `.inventory_item_name` text is "Test.allTheThings() T-Shirt (Red)" which is alphabetically last.
+
+### 5. Compare with Hand-Written (2 min)
+
+**Say:** "Now compare this AI-generated test with the hand-written product sort test in our test suite."
+
+Open `tests/03_product_tests/product_catalog.robot` and show the `Sort Products By Price Low To High` test.
+
+**Key points:**
+- The hand-written test uses resource file keywords (better abstraction)
+- The AI test might use raw selectors (less maintainable)
+- AI is great for boilerplate, humans are needed for test strategy
+
+### 6. Wrap Up (2 min)
+
+**Key takeaways:**
+- AI saves time on repetitive code
+- AI doesn't replace test design thinking
+- Always run the test — don't trust unexecuted code
+- Use AI for the "how", you decide the "what"
+
+## Backup Demo (if AI tools unavailable)
+
+If Claude Code / Copilot aren't available:
+
+1. Show a pre-generated test file
+2. Walk through how it was created
+3. Run it live
+4. Still do the comparison with hand-written tests
+
+## Pre-Generated Test (backup)
+
+Save this as `tests/student_exercises/ai_demo_sort_test.robot` before the workshop:
+
+```robotframework
+*** Settings ***
+Documentation     AI-generated test: Verify Z-A product sorting on SauceDemo.
+Library           Browser
+Resource          ../../resources/common.resource
+Resource          ../../resources/login_page.resource
+Resource          ../../resources/products_page.resource
+
+Suite Setup       Open SauceDemo
+Suite Teardown    Close SauceDemo
+
+*** Test Cases ***
+Products Should Sort By Name Z To A
+    [Documentation]    Verify that selecting Z-A sort orders products correctly.
+    [Tags]    student    exercise    ai-generated
+    Login With Credentials    ${VALID_USER}    ${VALID_PASSWORD}
+    Login Should Succeed
+    Sort Products By    Name (Z to A)
+    ${first_product}=    Get Text    css=.inventory_item_name >> nth=0
+    Should Be Equal    ${first_product}    Test.allTheThings() T-Shirt (Red)
+```