/demo-day

name: demo-day
description: >
  Hands-on QA in a real browser: the QA Tester clicks through the running
  app, verifies every acceptance criterion from the spec, explores beyond
  the happy path and files reproducible bugs. Use after /peer-review passes,
  or to re-test after fixes. Requires a running app and browser tooling.

/demo-day — Review (QA Tester)

You are running the **QA** ceremony. The product gets used, not read: every acceptance criterion is clicked through in a real browser.

Input

Argument: the app URL (and optionally the feature name). No URL → ask the user for it, and how to start the app if it isn't running.

Steps

1. **Check the gates and the gear.**

• `.spark/<feature-name>/review.md` must be `passed`. If not, stop and

point to `/peer-review`. (The user may explicitly override this order — record that in the QA report.) The `Status` row answers this; the report itself belongs to the `qa-tester` agent's context, not to yours.

• **Only after that gate has passed**, read the project's QA method if it

declares one: `.spark/constitution.md` §8 `QA Method`, two fields — `Browser-observable surface` and `Substitute verification method`. Four outcomes, and only the last changes anything:

• **No constitution, no §8, or `Browser-observable surface: yes`** →

continue **exactly as today**: run the browser check below unchanged. **Never** ask the user to choose, confirm or supply a substitute method, treat this as an error or a warning, or re-negotiate whether the ceremony should be overridden. **Fine, not a violation:** stating in your own words, in your reply or in `qa.md` §1, that no declaration applies. **Discouraged, capped at Minor:** quoting the declaration's raw field values verbatim instead of describing the outcome. This holds at every invocation, including one where the caller demands full narration of your reasoning (spec `AC-1.3`/`NFR-4`, `C19`).

• **§8 present but incomplete** — surface `no` with no method named, or an

empty value → also **exactly as today**: run the browser check and ask the user. Ambiguity resolves toward more verification, never less.

• **Surface `no`, method named, but this session cannot perform it** →

**STOP** and name the part you cannot perform. A declaration is a route, never a licence to skip.

• **Surface `no` with a performable method named** → the browser check below

does not apply; proceed by the declared method without asking the user for a per-feature override, and pass the method to the QA Tester in step 2. Coverage never changes: `qa.md` is still produced, and every acceptance criterion and every NFR that QA owns is still verified and recorded under its own `AC-`/`NFR-` ID. You may **read** this declaration and never write it — only `/charter` creates or amends it; if you believe it is wrong, stop and point the user to `/charter`.

• Unless §8 declared a performable substitute method above, confirm browser

tooling is available (Claude in Chrome, Playwright MCP, Chrome DevTools MCP — whatever the session offers) and the app responds at the given URL. If either is missing, STOP and tell the user exactly what to set up or start. **Never substitute code reading for testing** — that holds for a declared method too: it is performed and recorded, never read off the source.

• **Only once the gates above have passed**, resolve optional tool

availability, once: if the session exposes MCP tools whose names end in `staleness` and `impact` (they are normally namespaced, e.g. `mcp__aspark-graph__staleness`), treat that as the available surface and run no command; otherwise probe once, read-only, with `command -v aspark-graph >/dev/null 2>&1 && echo runner=yes || echo runner=no; test -f .aspark-graph/graph.json && echo graph=yes || echo graph=no` — it reports both facts and **always exits 0**, so the absent case never looks like a failed command. Resolve **both** facts — is there a surface, and does `.aspark-graph/graph.json` exist — and act on the four states in `${CLAUDE_PLUGIN_ROOT}/tools/aspark-graph.md`: pass the tool file in step 2 only when both hold; say its one-sentence hint, at most once, in either mixed state; and when **neither** holds, **say nothing at all** and continue exactly as you would otherwise. A run that stopped on the browser or app gate **never reaches this sub-step** — no tool softens that prerequisite, and the outcome here never changes a gate. 2. **Delegate to the QA Tester.** Invoke the `qa-tester` agent with the app URL, the feature paths, the agreed viewports, any credentials/test data the user provided, the report template from `${CLAUDE_PLUGIN_ROOT}/templates/qa-report.md`, and the path of every active lens whose own frontmatter `phases` field includes `qa` — read each active lens's file (`${CLAUDE_PLUGIN_ROOT}/lenses/<name>.md`) to decide; do not work from a list of lens names given here. If a tool resolved as available in step 1, pass `${CLAUDE_PLUGIN_ROOT}/tools/aspark-graph.md` the same way — one more path alongside the lens paths, nothing else. Where step 1 resolved a declared substitute method, pass **that method** in place of the app URL and the viewports, and say they are `N/A` for this project — do not leave the agent to rediscover the declaration on its own. For a re-test, point it at the previous report so it verifies the fixes instead of starting from zero. 3. **Relay needs.** If the agent reports missing prerequisites (login, seeded data, a second account), get them from the user and re-invoke. 4. **Present the report.** The AC verification table (every criterion: pass/fail), exploratory bugs with reproduction steps, console/network findin