feat: cross-browser screenshot baseline storage (RFC 0002)

.claude/commands/build-from-figma.md

@@ -278,21 +278,20 @@ Generate and run end-to-end tests appropriate to the app type.
 
 ## Phase 7: Cross-Browser Verification (Non-Blocking)
 
-Capture screenshots in Firefox and WebKit, compare against Chromium baseline.
+Diff Firefox and WebKit screenshots against their committed baselines with
+provenance verification (RFC 0002).
 
 ```bash
 # Only for web apps and PWAs (Chrome extensions are Chromium-only)
-./scripts/cross-browser-test.sh firefox http://localhost:3000
-./scripts/cross-browser-test.sh webkit http://localhost:3000
-
-node scripts/visual-diff.js --batch \
-  .claude/visual-qa/screenshots/firefox \
-  .claude/visual-qa/screenshots/chromium \
-  --output-dir .claude/visual-qa/diffs/firefox-vs-chromium \
-  --threshold 0.03
+./scripts/cross-browser-baseline.sh compare http://localhost:3000 --json
 ```
 
-Cross-browser differences are logged in the build report but do NOT block the pipeline.
+If no cross-browser baselines are committed yet the compare skips with a
+capture hint (`./scripts/cross-browser-baseline.sh capture <url>` in the
+pinned Playwright container, then commit `.claude/visual-qa/baselines/`).
+
+Cross-browser differences are logged in the build report but do NOT block the
+pipeline (`visualBaselines.blocking: false`).
 
 ## Phase 8: Quality Gate
 

.claude/pipeline.config.json

@@ -313,6 +313,41 @@
     "browsers": ["chromium"],
     "reportFile": "regression-report.md"
   },
+  "visualBaselines": {
+    "enabled": true,
+    "backend": "commit",
+    "storage": "git",
+    "baselineDir": ".claude/visual-qa/baselines",
+    "screenshotDir": ".claude/visual-qa/screenshots/cross-browser",
+    "diffDir": ".claude/visual-qa/diffs/cross-browser",
+    "browsers": ["chromium", "firefox", "webkit"],
+    "routes": ["/"],
+    "breakpoints": {
+      "mobile": 375,
+      "desktop": 1440
+    },
+    "threshold": 0.03,
+    "blocking": false,
+    "reportFile": "cross-browser-report.md",
+    "capture": {
+      "mode": "container",
+      "image": "mcr.microsoft.com/playwright:v1.61.1-noble",
+      "waitAfterLoadMs": 1500,
+      "fullPage": true
+    },
+    "provenance": {
+      "manifest": ".claude/visual-qa/baselines/manifest.json",
+      "policy": "warn"
+    },
+    "ciArtifact": {
+      "compareAgainst": "last-green-main",
+      "retentionDays": 30
+    },
+    "service": {
+      "provider": "chromatic",
+      "projectTokenEnv": "CHROMATIC_PROJECT_TOKEN"
+    }
+  },
   "deadCode": {
     "enabled": true,
     "tool": "knip",
@@ -472,9 +507,9 @@
       },
       "cross-browser": {
         "depends": ["component-build"],
-        "resources": [],
+        "resources": [{ "name": "port:dev-server", "mode": "shared" }],
         "blocking": false,
-        "description": "Firefox and WebKit screenshot comparison"
+        "description": "Firefox and WebKit baseline comparison (cross-browser-baseline.sh compare)"
       },
       "quality-gate": {
         "depends": ["component-build"],

.claude/pipeline.config.schema.json

@@ -362,6 +362,78 @@
       }
     },
 
+    "visualBaselines": {
+      "type": "object",
+      "additionalProperties": false,
+      "description": "Cross-browser (firefox/webkit) screenshot baseline storage per RFC 0002: pluggable backend, deterministic containerized capture, and per-baseline provenance. Same-browser chromium regression stays in regressionTesting.",
+      "properties": {
+        "enabled": { "type": "boolean" },
+        "backend": {
+          "type": "string",
+          "enum": ["commit", "ci-artifact", "service"],
+          "description": "Where baselines live: committed to the repo (default), fetched from the last-green-main CI artifact, or delegated to a SaaS provider."
+        },
+        "storage": {
+          "type": "string",
+          "enum": ["git", "lfs"],
+          "description": "commit backend only: plain git-tracked PNGs, or Git LFS for large baseline sets (see scripts/setup-baseline-lfs.sh)."
+        },
+        "baselineDir": { "type": "string", "description": "Baseline root, shared with regressionTesting; per-engine subdirectories." },
+        "screenshotDir": { "type": "string", "description": "Transient current-capture directory (gitignored)." },
+        "diffDir": { "type": "string", "description": "Transient diff-image directory (gitignored)." },
+        "browsers": { "$ref": "#/$defs/browserList" },
+        "routes": { "type": "array", "items": { "type": "string" } },
+        "breakpoints": { "$ref": "#/$defs/breakpointMap" },
+        "threshold": { "$ref": "#/$defs/percentageRatio", "description": "Cross-engine mismatch tolerance. Must equal e2e.crossBrowserDiffThreshold (structural check)." },
+        "blocking": { "type": "boolean", "description": "When true, compare failures exit non-zero (Phase B teeth). Keep false until determinism is proven." },
+        "reportFile": { "type": "string" },
+        "capture": {
+          "type": "object",
+          "additionalProperties": false,
+          "properties": {
+            "mode": {
+              "type": "string",
+              "enum": ["container", "local"],
+              "description": "container (required for trustworthy firefox/webkit baselines) runs capture in the pinned Playwright image; local captures are recorded with host=local and flagged by provenance."
+            },
+            "image": { "type": "string", "description": "Pinned Playwright container image (authoritative; scripts warn on drift vs the resolved @playwright/test version)." },
+            "waitAfterLoadMs": { "type": "integer", "minimum": 0 },
+            "fullPage": { "type": "boolean" }
+          }
+        },
+        "provenance": {
+          "type": "object",
+          "additionalProperties": false,
+          "properties": {
+            "manifest": { "type": "string", "description": "Committed provenance manifest path (engine, Playwright version, image, sha256, gitSha, host per baseline)." },
+            "policy": {
+              "type": "string",
+              "enum": ["warn", "enforce"],
+              "description": "warn: provenance mismatches are reported but baselines are still diffed. enforce: flagged baselines are excluded from diffing and counted as failures."
+            }
+          }
+        },
+        "ciArtifact": {
+          "type": "object",
+          "additionalProperties": false,
+          "description": "backend=ci-artifact only.",
+          "properties": {
+            "compareAgainst": { "type": "string", "enum": ["last-green-main"] },
+            "retentionDays": { "type": "integer", "minimum": 1 }
+          }
+        },
+        "service": {
+          "type": "object",
+          "additionalProperties": false,
+          "description": "backend=service only. The provider owns capture, diffing, and review; opt-in, requires a token in CI secrets.",
+          "properties": {
+            "provider": { "type": "string", "enum": ["chromatic", "percy"] },
+            "projectTokenEnv": { "type": "string", "description": "Name of the environment variable holding the provider project token." }
+          }
+        }
+      }
+    },
+
     "deadCode": {
       "type": "object",
       "additionalProperties": false,

.claude/skills/figma-to-react-workflow/SKILL.md

@@ -480,22 +480,28 @@ For each page in build-spec:
 
 ### Step 3.2.5: Cross-Browser Verification
 
-After Chromium passes the visual diff loop, verify in other browsers:
+After Chromium passes the visual diff loop, diff Firefox and WebKit against
+their committed baselines (RFC 0002):
 
 ```bash
-# Capture screenshots in Firefox and WebKit
-./scripts/cross-browser-test.sh firefox http://localhost:3000
-./scripts/cross-browser-test.sh webkit http://localhost:3000
-
-# Compare against Chromium baseline (higher threshold for browser differences)
-node scripts/visual-diff.js --batch \
-  .claude/visual-qa/screenshots/firefox \
-  .claude/visual-qa/screenshots/chromium \
-  --output-dir .claude/visual-qa/diffs/firefox-vs-chromium \
-  --threshold 0.03
+# Compare current firefox/webkit screenshots against committed baselines
+# (threshold from visualBaselines.threshold, 0.03 for cross-engine tolerance)
+./scripts/cross-browser-baseline.sh compare http://localhost:3000 --json
 ```
 
-Cross-browser results are logged in the build report but are **not blocking** by default.
+If no cross-browser baselines exist yet, the compare skips with a hint —
+capture them deterministically first and commit the result:
+
+```bash
+./scripts/cross-browser-baseline.sh capture http://localhost:3000   # pinned Playwright container
+git add .claude/visual-qa/baselines && git commit -m "test: add cross-browser baselines"
+```
+
+Each baseline's provenance (engine, Playwright version, container image,
+sha256, host) is verified from `baselines/manifest.json` before diffing, so
+stale or locally-captured baselines are flagged instead of producing false
+pixel diffs. Results are logged in the build report but are **not blocking**
+by default (`visualBaselines.blocking`).
 
 ### Step 3.3: E2E Tests
 
@@ -686,7 +692,8 @@ This skill works with:
 - **scripts/visual-diff.js** — Pixel-level screenshot comparison with region analysis
 - **scripts/verify-tokens.sh** — Token integrity enforcement in quality gate
 - **scripts/verify-test-coverage.sh** — Test existence verification (TDD gate)
-- **scripts/cross-browser-test.sh** — Multi-browser screenshot capture
+- **scripts/cross-browser-baseline.sh** — Cross-browser baseline capture + provenance-verified comparison (RFC 0002)
+- **scripts/cross-browser-test.sh** — Ad-hoc multi-browser screenshot capture (no comparison)
 - **scripts/generate-stories.sh** — AST-based Storybook story + MDX generation (Phase 4.5)
 - **.claude/pipeline.config.json** — Thresholds, iteration limits, app type config
 

.claude/skills/parallel-orchestration/SKILL.md

@@ -145,7 +145,7 @@ function hasResourceConflict(candidatePhase, runningPhases):
 | `visual-diff` | **Agent:** invoke visual-qa-verification skill with pixel-diff loop (max iterations from `iterationLoop.maxVisualIterations`) |
 | `dark-mode` | **Bash:** `./scripts/check-dark-mode.sh http://localhost:3000` |
 | `e2e-tests` | **Agent:** invoke e2e-test-generator skill with flows from build-spec.json |
-| `cross-browser` | **Bash:** `./scripts/cross-browser-test.sh firefox http://localhost:3000 && ./scripts/cross-browser-test.sh webkit http://localhost:3000` |
+| `cross-browser` | **Bash:** `./scripts/cross-browser-baseline.sh compare http://localhost:3000 --json` (diffs firefox/webkit against committed baselines with provenance checks — RFC 0002; skips with a capture hint when no baselines exist) |
 | `quality-gate` | **Parallel sub-dispatch:** run all `qualityGateSubtasks` concurrently (see below) |
 | `responsive` | **Bash:** `./scripts/check-responsive.sh http://localhost:3000` |
 | `report` | **Agent:** generate `build-report.md` in `.claude/visual-qa/` from all collected phase results, timings, and diff images |

.claude/skills/visual-qa-verification/SKILL.md

@@ -143,35 +143,28 @@ When the diff reports failing regions, use this mapping to prioritize fixes:
 
 ### Step 2.5: Cross-Browser Verification
 
-After Chromium passes the visual diff loop, verify in Firefox and WebKit.
-
-**Using Playwright MCP or cross-browser-test.sh:**
+After Chromium passes the visual diff loop, diff Firefox and WebKit against
+their committed same-engine baselines (RFC 0002):
 
 ```bash
-# Capture Firefox screenshots
-./scripts/cross-browser-test.sh firefox http://localhost:3000
-
-# Capture WebKit screenshots
-./scripts/cross-browser-test.sh webkit http://localhost:3000
-
-# Compare Firefox against Chromium baseline (not Figma)
-node scripts/visual-diff.js --batch \
-  .claude/visual-qa/screenshots/firefox \
-  .claude/visual-qa/screenshots/chromium \
-  --output-dir .claude/visual-qa/diffs/firefox-vs-chromium \
-  --threshold 0.03
-
-# Compare WebKit against Chromium baseline
-node scripts/visual-diff.js --batch \
-  .claude/visual-qa/screenshots/webkit \
-  .claude/visual-qa/screenshots/chromium \
-  --output-dir .claude/visual-qa/diffs/webkit-vs-chromium \
-  --threshold 0.03
+# Compare current firefox/webkit screenshots against committed baselines,
+# with per-baseline provenance verification from baselines/manifest.json
+./scripts/cross-browser-baseline.sh compare http://localhost:3000 --json
+
+# First time (no baselines committed yet): capture deterministically in the
+# pinned Playwright container, then commit
+./scripts/cross-browser-baseline.sh capture http://localhost:3000
+git add .claude/visual-qa/baselines
+
+# Provenance-only check (no dev server needed)
+./scripts/cross-browser-baseline.sh verify
 ```
 
-**Cross-browser threshold:** Use a slightly higher threshold (3% vs 2%) since browser rendering engines have legitimate differences in anti-aliasing, font rendering, and sub-pixel positioning.
+For ad-hoc engine captures without comparison, `./scripts/cross-browser-test.sh <browser> <url>` still works.
+
+**Cross-browser threshold:** `visualBaselines.threshold` (3% vs the 2% same-browser threshold) since browser rendering engines have legitimate differences in anti-aliasing, font rendering, and sub-pixel positioning.
 
-**Cross-browser failures are NOT blocking** by default (configurable in `pipeline.config.json` → `qualityGate.crossBrowserScreenshotsRequired`). They are reported in the build report for manual review.
+**Cross-browser failures are NOT blocking** by default (`visualBaselines.blocking`, and `qualityGate.crossBrowserScreenshotsRequired` for the gate). They are reported in the build report for manual review. Compares run outside the pinned container are marked *advisory* by the provenance check.
 
 **Common cross-browser differences to ignore:**
 - Font rendering (especially on macOS WebKit vs Windows Chromium)
@@ -349,7 +342,8 @@ This skill works with:
 - **Playwright MCP** -- Cross-browser screenshots (Firefox, WebKit) and automated interaction testing
 - **Figma MCP** -- Source design screenshots for comparison (`get_screenshot`, `get_design_context`)
 - **scripts/visual-diff.js** -- Pixel-level screenshot comparison with region analysis
-- **scripts/cross-browser-test.sh** -- Multi-browser screenshot capture
+- **scripts/cross-browser-baseline.sh** -- Cross-browser baseline capture + provenance-verified comparison (RFC 0002)
+- **scripts/cross-browser-test.sh** -- Ad-hoc multi-browser screenshot capture (no comparison)
 - **.claude/pipeline.config.json** -- Thresholds, iteration limits, breakpoint configuration
 
 ## Verification Report Template