Open-source AI Security Operations Center: alert fusion, LLM-agent triage, MITRE ATT&CK investigation, and a replayable decision ledger for every agent step. Self-hostable, runs with no API keys, MIT licensed. Ships an MCP server for Claude, Cursor and Continue.
Visual regression (Phase 4.7)
This directory holds the AiSOC design-system visual regression suite. It runs against the static Storybook build, not the live console, because:
- Stories are pure UI — they never call out to the FastAPI backend, so the suite needs zero API mocking to be deterministic.
- Stories cover every primitive (
Button,Card,Badge,EmptyState, …) and every composite (KpiTiles,Tabs,FormFields, …) the console uses. A regression in those primitives ripples into every page; catching it at the story level catches it everywhere. - The static Storybook bundle is byte-identical run-to-run, so
Playwright's
toHaveScreenshot()diff stays at the noise floor.
Local workflow
# 1. Build storybook (once per shape-changing edit).
pnpm --filter @aisoc/web build-storybook
# 2. Run the visual suite — first run captures baselines, later runs
# diff against them.
pnpm --filter @aisoc/web visual
# 3. After an intentional design change, refresh the baselines.
pnpm --filter @aisoc/web visual:update
git add apps/web/e2e/visual/visual.spec.ts-snapshots
git commit -m "design: refresh visual baselines for <change>"
The first command produces apps/web/storybook-static/ and an
index.json describing every story id. The visual spec reads that
index, walks each story, and snapshots it at two viewports
(desktop=1440×900, mobile=390×844).
CI behaviour
.github/workflows/visual-regression.yml runs the suite on every PR
that touches apps/web/** or packages/ui/**. It uses a pinned
mcr.microsoft.com/playwright:v1.49.0-jammy container so font
rendering + Chromium build stay identical between baseline-capture and
diff-runs. Changing that tag invalidates every baseline; expect a
follow-up PR to regenerate them.
On failure the workflow uploads two artifacts:
playwright-report/— the HTML report with diff overlays.playwright-test-results/— the raw<name>-expected.png/<name>-actual.png/<name>-diff.pngtriples for offline review.
When the diff is real
A genuine regression looks like a 10–40% pixel diff localised to one
component. A run that flags every snapshot is almost always a
container / font / Chromium version drift — bisect by checking the
Playwright tag in .github/workflows/visual-regression.yml.
Excluding non-deterministic stories
Some stories animate continuously (e.g. shimmer loaders) or transition
between states mid-render (e.g. theme switcher). Add their story ids
to SKIP_STORY_IDS in visual.spec.ts with a one-line comment
explaining why. Keep the list small and reviewable.