npm - delimit-cli - Versions diffs - 4.3.4 → 4.5.0 - Mend

delimit-cli 4.3.4 → 4.5.0

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.

Files changed (46) hide show

package/CHANGELOG.md +96 -0
package/README.md +25 -18
package/adapters/codex-security.js +64 -0
package/adapters/codex-skill.js +78 -0
package/adapters/cursor-rules.js +73 -0
package/bin/delimit-setup.js +23 -0
package/gateway/ai/backends/governance_bridge.py +168 -2
package/gateway/ai/backends/memory_bridge.py +218 -3
package/gateway/ai/backends/tools_design.py +563 -83
package/gateway/ai/backends/tools_infra.py +21 -7
package/gateway/ai/backends/tools_real.py +3 -1
package/gateway/ai/content_grounding/__init__.py +98 -0
package/gateway/ai/content_grounding/build.py +350 -0
package/gateway/ai/content_grounding/consume.py +280 -0
package/gateway/ai/content_grounding/features.py +218 -0
package/gateway/ai/content_grounding/fixtures/fail/01_missing_evidence.json +9 -0
package/gateway/ai/content_grounding/fixtures/fail/02_unknown_evidence_prefix.json +9 -0
package/gateway/ai/content_grounding/fixtures/fail/03_banned_comparative.json +17 -0
package/gateway/ai/content_grounding/fixtures/fail/04_banned_adoption.json +17 -0
package/gateway/ai/content_grounding/fixtures/fail/05_aggregate_no_numeric.json +17 -0
package/gateway/ai/content_grounding/fixtures/fail/06_unversioned_inference_rule.json +18 -0
package/gateway/ai/content_grounding/fixtures/pass/01_feature_shipped.json +18 -0
package/gateway/ai/content_grounding/fixtures/pass/02_aggregate_claim.json +23 -0
package/gateway/ai/content_grounding/fixtures/pass/03_attestation.json +16 -0
package/gateway/ai/content_grounding/schemas/claim.schema.json +40 -0
package/gateway/ai/content_grounding/schemas/event.schema.json +23 -0
package/gateway/ai/content_grounding/schemas.py +276 -0
package/gateway/ai/content_grounding/telemetry.py +221 -0
package/gateway/ai/governance.py +89 -0
package/gateway/ai/hot_reload.py +148 -7
package/gateway/ai/inbox_drafts/__init__.py +61 -0
package/gateway/ai/inbox_drafts/registry.py +412 -0
package/gateway/ai/inbox_drafts/schema.py +374 -0
package/gateway/ai/inbox_executor.py +565 -0
package/gateway/ai/ledger_manager.py +1483 -25
package/gateway/ai/license_core.py +3 -1
package/gateway/ai/mcp_bridge.py +1 -1
package/gateway/ai/reddit_proxy.py +8 -6
package/gateway/ai/server.py +451 -9
package/gateway/ai/supabase_sync.py +47 -7
package/gateway/ai/swarm.py +1 -1
package/gateway/ai/workers/executor.py +1 -1
package/gateway/core/diff_engine_v2.py +45 -10
package/gateway/core/zero_spec/express_extractor.py +1 -1
package/lib/delimit-template.js +5 -0
package/package.json +1 -1

package/gateway/ai/backends/tools_design.py CHANGED Viewed

@@ -70,30 +70,111 @@ def _read_text(path: Path, limit: int = 200_000) -> str:
 _CSS_VAR_RE = re.compile(r"--([a-zA-Z0-9_-]+)\s*:\s*([^;]+);")
 _MEDIA_QUERY_RE = re.compile(r"@media[^{]*\(\s*(?:min|max)-width\s*:\s*([^)]+)\)")
+# LED-1010: selector-aware extraction. Captures the selector (e.g. `:root`,
+# `.dark`, `[data-theme="dark"]`) that owns each custom property block so
+# light/dark variants of the same token name don't silently dedupe into one.
+_CSS_BLOCK_RE = re.compile(
+    r"(?P<selector>[^{}@\n][^{}]{0,200})\{(?P<body>[^{}]*)\}",
+    re.DOTALL,
+)
+def _mode_from_selector(selector: str) -> str:
+    """Best-effort mode guess from a CSS selector: 'dark' / 'light' / 'base'."""
+    s = selector.strip().lower()
+    if any(t in s for t in (".dark", "[data-theme=\"dark\"]", "[data-mode=\"dark\"]", "prefers-color-scheme: dark")):
+        return "dark"
+    if any(t in s for t in (".light", "[data-theme=\"light\"]", "[data-mode=\"light\"]", "prefers-color-scheme: light")):
+        return "light"
+    if s in (":root", "html", "body", "*"):
+        return "base"
+    return "scoped"
+# LED-1010: common domain/semantic token prefixes. Anything starting with these
+# reads as application-meaning rather than theme-primitive, and belongs in the
+# `semantic` bucket, not `other`. Downstream generators need this split to know
+# which tokens are safe to remap vs which carry app meaning.
+_SEMANTIC_PREFIXES = (
+    "score-", "status-", "blur-", "price-", "rank-", "tier-", "risk-",
+    "badge-", "level-", "alert-",
+)
-def _extract_css_variables(text: str) -> Dict[str, List[Dict[str, str]]]:
-    """Extract CSS custom properties grouped by category."""
+def _token_taxonomy(name: str) -> str:
+    """Classify a token name as primitive | semantic | other."""
+    n = name.lower().lstrip("-")
+    if any(n.startswith(p) for p in _SEMANTIC_PREFIXES):
+        return "semantic"
+    # Core theme primitives
+    if any(n.startswith(p) for p in (
+        "color-", "bg-", "text-", "border-", "fill-", "stroke-",
+        "accent", "primary", "secondary", "muted", "foreground",
+        "background", "surface", "ring-", "input-",
+        "space-", "spacing-", "gap-", "size-", "radius-",
+        "font-", "leading-", "tracking-",
+    )):
+        return "primitive"
+    return "other"
+def _extract_css_variables(text: str, source: str = "") -> Dict[str, List[Dict[str, str]]]:
+    """Extract CSS custom properties grouped by category.
+    LED-1010: now selector-aware. Each returned entry carries `selector`
+    (the CSS rule it was declared in), `mode` (dark/light/base/scoped), and
+    `taxonomy` (primitive/semantic/other) so consumers can tell
+    `--bg-base` in `:root` from `--bg-base` in `.dark`.
+    """
     colors: List[Dict[str, str]] = []
     spacing: List[Dict[str, str]] = []
     typography: List[Dict[str, str]] = []
     other: List[Dict[str, str]] = []
+    semantic: List[Dict[str, str]] = []
+    # Walk each CSS rule block so we know which selector owns each declaration.
+    for m in _CSS_BLOCK_RE.finditer(text):
+        selector = m.group("selector").strip()
+        mode = _mode_from_selector(selector)
+        body = m.group("body")
+        for name, value in _CSS_VAR_RE.findall(body):
+            value = value.strip()
+            taxonomy = _token_taxonomy(name)
+            entry = {
+                "name": f"--{name}",
+                "value": value,
+                "selector": selector,
+                "mode": mode,
+                "taxonomy": taxonomy,
+            }
+            if source:
+                entry["source"] = source
+            lower = name.lower()
+            # Semantic tokens bypass the keyword bucket: they carry app meaning
+            # and should be routed to the semantic bucket regardless of value type.
+            if taxonomy == "semantic":
+                semantic.append(entry)
+                continue
-    for name, value in _CSS_VAR_RE.findall(text):
-        value = value.strip()
-        entry = {"name": f"--{name}", "value": value}
-        lower = name.lower()
-        if any(k in lower for k in ("color", "bg", "text", "border", "fill", "stroke", "accent", "primary", "secondary")):
-            colors.append(entry)
-        elif any(k in lower for k in ("space", "gap", "margin", "padding", "size", "width", "height", "radius")):
-            spacing.append(entry)
-        elif any(k in lower for k in ("font", "line", "letter", "text", "heading")):
-            typography.append(entry)
-        elif _is_color_value(value):
-            colors.append(entry)
-        else:
-            other.append(entry)
+            if any(k in lower for k in ("color", "bg", "text", "border", "fill", "stroke", "accent", "primary", "secondary", "surface", "foreground", "background", "ring")):
+                colors.append(entry)
+            elif any(k in lower for k in ("space", "gap", "margin", "padding", "size", "width", "height", "radius")):
+                spacing.append(entry)
+            elif any(k in lower for k in ("font", "line", "letter", "text", "heading", "leading", "tracking")):
+                typography.append(entry)
+            elif _is_color_value(value):
+                colors.append(entry)
+            else:
+                other.append(entry)
-    return {"colors": colors, "spacing": spacing, "typography": typography, "other": other}
+    return {
+        "colors": colors,
+        "spacing": spacing,
+        "typography": typography,
+        "other": other,
+        "semantic": semantic,
+    }
 def _is_color_value(v: str) -> bool:
@@ -128,6 +209,127 @@ def _parse_tailwind_config(text: str) -> Dict[str, Any]:
     return {"colors_count": colors_count, "spacing_count": spacing_count, "breakpoints": breakpoints}
+# ---------------------------------------------------------------------------
+# LED-1010: Tailwind-awareness helpers
+# ---------------------------------------------------------------------------
+# Tailwind ships an opinionated default scale. When a repo has tailwind.config.js
+# and doesn't override these, the framework defaults are the design tokens in
+# use — `extract_tokens` previously reported 0 for each because it only scanned
+# CSS `--*` variables.
+TAILWIND_DEFAULT_BREAKPOINTS = ["sm=640px", "md=768px", "lg=1024px", "xl=1280px", "2xl=1536px"]
+TAILWIND_DEFAULT_SPACING = [
+    "0=0px", "px=1px", "0.5=0.125rem", "1=0.25rem", "1.5=0.375rem", "2=0.5rem",
+    "2.5=0.625rem", "3=0.75rem", "3.5=0.875rem", "4=1rem", "5=1.25rem", "6=1.5rem",
+    "7=1.75rem", "8=2rem", "9=2.25rem", "10=2.5rem", "11=2.75rem", "12=3rem",
+    "14=3.5rem", "16=4rem", "20=5rem", "24=6rem", "28=7rem", "32=8rem", "36=9rem",
+    "40=10rem", "44=11rem", "48=12rem", "52=13rem", "56=14rem", "60=15rem",
+    "64=16rem", "72=18rem", "80=20rem", "96=24rem",
+]
+TAILWIND_DEFAULT_FONT_SIZES = [
+    "text-xs=0.75rem", "text-sm=0.875rem", "text-base=1rem", "text-lg=1.125rem",
+    "text-xl=1.25rem", "text-2xl=1.5rem", "text-3xl=1.875rem", "text-4xl=2.25rem",
+    "text-5xl=3rem", "text-6xl=3.75rem", "text-7xl=4.5rem", "text-8xl=6rem",
+    "text-9xl=8rem",
+]
+# Responsive prefix regex. Matches `sm:`, `md:`, `lg:`, `xl:`, `2xl:` in a
+# JSX className/string context. Tailwind is mobile-first by default: these
+# prefixes apply at and above the named breakpoint.
+_TAILWIND_RESPONSIVE_PREFIX_RE = re.compile(r"(?<![\w-])(sm|md|lg|xl|2xl):[\w\[\]-]+", re.IGNORECASE)
+# Dark-mode class usage in JSX: `dark:bg-black` etc.
+_TAILWIND_DARK_PREFIX_RE = re.compile(r"(?<![\w-])dark:[\w\[\]-]+")
+# Tailwind spacing/layout utility class usage — rough counter for the
+# responsive_units_count signal.
+_TAILWIND_UTILITY_RE = re.compile(
+    r"(?<![\w-])(?:w|h|m|p|mx|my|mt|mb|ml|mr|px|py|pt|pb|pl|pr|gap|space|"
+    r"max-w|min-w|max-h|min-h|inset|top|bottom|left|right|z|grid-cols|col-span)"
+    r"-[\w./\[\]%-]+"
+)
+def _has_tailwind_config(root: Path) -> Optional[Path]:
+    """Return the first found tailwind.config.{js,ts,mjs,cjs} or None."""
+    for tw_name in ("tailwind.config.js", "tailwind.config.ts", "tailwind.config.mjs", "tailwind.config.cjs"):
+        tw_path = root / tw_name
+        if tw_path.exists():
+            return tw_path
+    # Tailwind v4 uses @import "tailwindcss" in CSS with no separate config
+    return None
+def _detect_tailwind_v4(root: Path) -> bool:
+    """Detect Tailwind v4 (no config file, uses @import "tailwindcss" in CSS)."""
+    for cf in list(root.rglob("*.css"))[:20]:
+        try:
+            text = cf.read_text(errors="replace")[:5000]
+            if '@import "tailwindcss"' in text or "@import 'tailwindcss'" in text:
+                return True
+        except Exception:
+            continue
+    return False
+def _scan_tailwind_utilities(root: Path) -> Dict[str, Any]:
+    """Scan JSX/TSX/Vue/Svelte files for Tailwind utility class usage.
+    Returns counts of responsive-prefixed classes, dark-prefix classes,
+    and general utility classes — what the old validate_responsive missed
+    entirely when it only looked at raw CSS.
+    """
+    component_files = _find_files(root, [".tsx", ".jsx", ".ts", ".js", ".vue", ".svelte", ".html"])
+    responsive_hits = 0
+    dark_hits = 0
+    utility_hits = 0
+    breakpoints_seen: set = set()
+    files_with_utilities = 0
+    # Cap to keep runtime bounded on huge monorepos
+    for cf in component_files[:500]:
+        text = _read_text(cf, limit=200_000)
+        if not text:
+            continue
+        # Require a Tailwind signal before counting (otherwise every JS file
+        # matches via collisions like `py-pi`)
+        if "className" not in text and "class=" not in text and "tw`" not in text:
+            continue
+        file_had_utility = False
+        for m in _TAILWIND_RESPONSIVE_PREFIX_RE.finditer(text):
+            responsive_hits += 1
+            breakpoints_seen.add(m.group(1).lower())
+            file_had_utility = True
+        dark_hits += len(_TAILWIND_DARK_PREFIX_RE.findall(text))
+        for m in _TAILWIND_UTILITY_RE.finditer(text):
+            utility_hits += 1
+            file_had_utility = True
+        if file_had_utility:
+            files_with_utilities += 1
+    return {
+        "responsive_prefix_count": responsive_hits,
+        "dark_prefix_count": dark_hits,
+        "utility_count": utility_hits,
+        "files_with_utilities": files_with_utilities,
+        "files_scanned": len(component_files),
+        "breakpoints_seen": sorted(breakpoints_seen, key=lambda b: ["sm", "md", "lg", "xl", "2xl"].index(b) if b in ["sm", "md", "lg", "xl", "2xl"] else 99),
+    }
+# ---------------------------------------------------------------------------
+# LED-1010: status taxonomy
+# ---------------------------------------------------------------------------
+# Consumers branch on `status` to decide whether to gate CI. A tool that
+# returns `ok` while producing partial/wrong results is a silent failure —
+# the biggest gap identified by the 2026-04-24 pilot run. These constants
+# define the explicit vocabulary.
+STATUS_OK = "ok"                          # all checks ran, no gaps
+STATUS_DEGRADED = "degraded"              # ran, but known gaps (e.g. only CSS scanned, not JSX)
+STATUS_PARTIAL_COVERAGE = "partial_coverage"  # ran a subset of the requested standard
+STATUS_TOOLCHAIN_MISSING = "toolchain_missing"  # required external tool absent
+STATUS_ERROR = "error"                    # unrecoverable
 # ---------------------------------------------------------------------------
 # Component scanning helpers
 # ---------------------------------------------------------------------------
@@ -209,61 +411,139 @@ def design_extract_tokens(
     if not root.is_dir():
         return {"tool": "design.extract_tokens", "error": f"Directory not found: {root}"}
-    all_tokens: Dict[str, List] = {"colors": [], "spacing": [], "typography": [], "breakpoints": [], "other": []}
+    all_tokens: Dict[str, List] = {
+        "colors": [], "spacing": [], "typography": [], "breakpoints": [],
+        "other": [], "semantic": [],
+    }
     source_files: List[str] = []
+    coverage: Dict[str, Any] = {
+        "css_scanned": False,
+        "tailwind_config_found": False,
+        "tailwind_v4_detected": False,
+        "tailwind_defaults_emitted": False,
+        "selector_aware": True,
+        "semantic_taxonomy": True,
+    }
+    gaps: List[str] = []
+    # 1. Tailwind config (LED-1010: emit framework defaults when present)
+    tw_path = _has_tailwind_config(root)
+    tw_v4 = False
+    tw_config_text = ""
+    if tw_path:
+        coverage["tailwind_config_found"] = True
+        tw_config_text = _read_text(tw_path)
+        parsed = _parse_tailwind_config(tw_config_text)
+        source_files.append(str(tw_path))
+        # User-defined breakpoints override defaults
+        if parsed["breakpoints"]:
+            all_tokens["breakpoints"].extend(
+                [{"name": bp, "source": str(tw_path), "origin": "tailwind_config"} for bp in parsed["breakpoints"]]
+            )
+    else:
+        tw_v4 = _detect_tailwind_v4(root)
+        coverage["tailwind_v4_detected"] = tw_v4
+    # If Tailwind is in play at all, surface the framework's default token
+    # scales so downstream consumers can see what utility classes reference.
+    # Previously these returned zero and the caller couldn't tell if the
+    # design system was truly empty or just invisible to us.
+    if tw_path or tw_v4:
+        coverage["tailwind_defaults_emitted"] = True
+        framework_source = str(tw_path) if tw_path else "tailwind-v4 (@import)"
+        # Emit defaults only when user config didn't already cover them
+        if not all_tokens["breakpoints"]:
+            for bp_def in TAILWIND_DEFAULT_BREAKPOINTS:
+                name, value = bp_def.split("=", 1)
+                all_tokens["breakpoints"].append({
+                    "name": name, "value": value, "source": framework_source,
+                    "origin": "tailwind_default",
+                })
-    # 1. Tailwind config
-    for tw_name in ("tailwind.config.js", "tailwind.config.ts", "tailwind.config.mjs", "tailwind.config.cjs"):
-        tw_path = root / tw_name
-        if tw_path.exists():
-            text = _read_text(tw_path)
-            parsed = _parse_tailwind_config(text)
-            source_files.append(str(tw_path))
-            if parsed["breakpoints"]:
-                all_tokens["breakpoints"].extend(
-                    [{"name": bp, "source": str(tw_path)} for bp in parsed["breakpoints"]]
-                )
-            break
+        for sp_def in TAILWIND_DEFAULT_SPACING:
+            name, value = sp_def.split("=", 1)
+            all_tokens["spacing"].append({
+                "name": f"spacing.{name}", "value": value, "source": framework_source,
+                "origin": "tailwind_default", "taxonomy": "primitive",
+            })
-    # 2. CSS / SCSS files
+        for fs_def in TAILWIND_DEFAULT_FONT_SIZES:
+            name, value = fs_def.split("=", 1)
+            all_tokens["typography"].append({
+                "name": name, "value": value, "source": framework_source,
+                "origin": "tailwind_default", "taxonomy": "primitive",
+            })
+    # 2. CSS / SCSS files (now selector-aware via _extract_css_variables)
     css_files = _find_files(root, [".css", ".scss", ".sass"])
     for cf in css_files:
         text = _read_text(cf)
         if "--" not in text and "@media" not in text:
             continue
         source_files.append(str(cf))
-        vars_found = _extract_css_variables(text)
-        for cat in ("colors", "spacing", "typography", "other"):
-            for entry in vars_found[cat]:
-                entry["source"] = str(cf)
-                all_tokens[cat].append(entry)
+        coverage["css_scanned"] = True
+        vars_found = _extract_css_variables(text, source=str(cf))
+        for cat in ("colors", "spacing", "typography", "other", "semantic"):
+            all_tokens[cat].extend(vars_found.get(cat, []))
-        # breakpoints from media queries
+        # breakpoints from media queries (user-authored, origin=css_media)
         for bp_val in _MEDIA_QUERY_RE.findall(text):
-            all_tokens["breakpoints"].append({"value": bp_val.strip(), "source": str(cf)})
+            all_tokens["breakpoints"].append({
+                "value": bp_val.strip(), "source": str(cf),
+                "origin": "css_media",
+            })
     # 3. Filter by token_types if specified
     if token_types:
         all_tokens = {k: v for k, v in all_tokens.items() if k in token_types}
-    # Deduplicate breakpoints
-    seen_bp = set()
-    unique_bp = []
+    # LED-1010 dark-mode dedup: collapse (name, mode) duplicates rather than
+    # eliding mode altogether. A token with the same name in `:root` and
+    # `.dark` is one logical token with two values — surface both.
+    for cat in ("colors", "spacing", "typography", "other", "semantic"):
+        if cat not in all_tokens:
+            continue
+        seen: Dict[tuple, Dict] = {}
+        for entry in all_tokens[cat]:
+            key = (entry.get("name", ""), entry.get("mode", ""))
+            if key in seen:
+                continue
+            seen[key] = entry
+        all_tokens[cat] = list(seen.values())
+    # Breakpoint dedup by (name|value, origin)
+    seen_bp: set = set()
+    unique_bp: List[Dict[str, Any]] = []
     for bp in all_tokens.get("breakpoints", []):
-        key = bp.get("name", bp.get("value", ""))
-        if key not in seen_bp:
-            seen_bp.add(key)
-            unique_bp.append(bp)
+        key = (bp.get("name", bp.get("value", "")), bp.get("origin", ""))
+        if key in seen_bp:
+            continue
+        seen_bp.add(key)
+        unique_bp.append(bp)
     if "breakpoints" in all_tokens:
         all_tokens["breakpoints"] = unique_bp
-    total = sum(len(v) for v in all_tokens.values())
+    # Coverage gaps → status taxonomy (LED-1010)
+    if not coverage["css_scanned"] and not (tw_path or tw_v4):
+        gaps.append("No CSS variables and no Tailwind config found — project may not use standard design tokens")
+    if (tw_path or tw_v4) and not coverage["css_scanned"]:
+        gaps.append("Tailwind detected but no user CSS variables found — only framework defaults emitted")
+    if gaps:
+        status = STATUS_PARTIAL_COVERAGE
+    else:
+        status = STATUS_OK
+    total = sum(len(v) for v in all_tokens.values() if isinstance(v, list))
     result = {
         "tool": "design.extract_tokens",
-        "status": "ok",
+        "status": status,
         "tokens": all_tokens,
         "total_tokens": total,
         "source_files": sorted(set(source_files)),
+        "coverage": coverage,
+        "gaps": gaps,
         "figma_used": False,
     }
     # If user passed a figma_file_key but no token is available, add a hint
@@ -496,7 +776,13 @@ def design_validate_responsive(
     project_path: str,
     check_types: Optional[List[str]] = None,
 ) -> Dict[str, Any]:
-    """Validate responsive design patterns via static analysis (Playwright optional)."""
+    """Validate responsive design patterns via static analysis.
+    LED-1010: now Tailwind-aware. Scans JSX/TSX/Vue/Svelte for utility-class
+    responsive prefixes (sm:/md:/lg:/xl:/2xl:) in addition to @media CSS.
+    Tailwind is mobile-first by default; previously the tool reported
+    `mobile_first: false` on any Tailwind-only codebase, which was wrong.
+    """
     root = Path(project_path)
     if not root.is_dir():
         return {"tool": "design.validate_responsive", "error": f"Directory not found: {root}"}
@@ -504,14 +790,23 @@ def design_validate_responsive(
     issues: List[Dict[str, str]] = []
     breakpoints_found: List[str] = []
     viewport_meta = False
+    viewport_blocks_zoom = False
     responsive_units_count = 0
+    coverage: Dict[str, Any] = {
+        "css_scanned": False,
+        "jsx_scanned": False,
+        "tailwind_detected": False,
+    }
     # Scan HTML files for viewport meta
     html_files = _find_files(root, [".html", ".htm"])
     for hf in html_files:
         text = _read_text(hf)
-        if _VIEWPORT_META_RE.search(text):
+        m = _VIEWPORT_META_RE.search(text)
+        if m:
             viewport_meta = True
+            if "maximum-scale" in m.group() or "user-scalable=no" in m.group():
+                viewport_blocks_zoom = True
             break
     # Also check Next.js layout files
@@ -522,6 +817,8 @@ def design_validate_responsive(
                 text = _read_text(c)
                 if "viewport" in text.lower():
                     viewport_meta = True
+                    if "maximum-scale" in text.lower() or "userScalable: false" in text:
+                        viewport_blocks_zoom = True
                     break
             if viewport_meta:
                 break
@@ -529,10 +826,20 @@ def design_validate_responsive(
     if not viewport_meta:
         issues.append({"severity": "warning", "message": "No viewport meta tag detected", "fix": "Add <meta name='viewport' content='width=device-width, initial-scale=1'>"})
+    # LED-1010: WCAG2AA (1.4.4) — zoom-blocking viewport
+    if viewport_meta and viewport_blocks_zoom:
+        issues.append({
+            "severity": "error",
+            "message": "Viewport blocks user zoom (maximum-scale=1 or user-scalable=no) — WCAG 1.4.4 violation",
+            "fix": "Remove maximum-scale and user-scalable restrictions so users can zoom to 200%",
+        })
     # Scan CSS for media queries and responsive patterns
     css_files = _find_files(root, [".css", ".scss", ".sass"])
     for cf in css_files:
         text = _read_text(cf)
+        if text:
+            coverage["css_scanned"] = True
         for bp_val in _MEDIA_QUERY_RE.findall(text):
             bp_val = bp_val.strip()
             if bp_val not in breakpoints_found:
@@ -547,6 +854,39 @@ def design_validate_responsive(
         min_width_count += len(re.findall(r"min-width\s*:", text))
         max_width_count += len(re.findall(r"max-width\s*:", text))
+    # LED-1010: Tailwind utility class scan. The old mobile_first heuristic
+    # only considered raw CSS and returned `false` on any Tailwind codebase.
+    # Tailwind prefixes (sm:/md:/...) are mobile-first by design: they apply
+    # AT OR ABOVE the named breakpoint.
+    tw_path = _has_tailwind_config(root)
+    tw_v4 = False if tw_path else _detect_tailwind_v4(root)
+    coverage["tailwind_detected"] = bool(tw_path or tw_v4)
+    tw_stats: Dict[str, Any] = {}
+    if coverage["tailwind_detected"]:
+        tw_stats = _scan_tailwind_utilities(root)
+        coverage["jsx_scanned"] = tw_stats.get("files_scanned", 0) > 0
+        # Add Tailwind default breakpoints that were actually USED in the codebase
+        for bp in tw_stats.get("breakpoints_seen", []):
+            bp_label = f"tailwind:{bp}"
+            if bp_label not in breakpoints_found:
+                breakpoints_found.append(bp_label)
+        # Utility classes count as responsive units (layout-responsive classes
+        # like w-full/h-screen/max-w-*)
+        responsive_units_count += tw_stats.get("utility_count", 0)
+    # Determine mobile_first. Tailwind prefixes → mobile-first. Raw CSS: prefer
+    # min-width over max-width.
+    if coverage["tailwind_detected"] and tw_stats.get("responsive_prefix_count", 0) > 0:
+        mobile_first = True
+    elif min_width_count == 0 and max_width_count == 0:
+        # No explicit breakpoints at all — don't claim mobile_first either way
+        mobile_first = None
+    else:
+        mobile_first = min_width_count >= max_width_count
     if max_width_count > min_width_count * 2 and max_width_count > 3:
         issues.append({
             "severity": "info",
@@ -554,9 +894,7 @@ def design_validate_responsive(
             "fix": "Consider mobile-first approach using min-width media queries",
         })
-    if not breakpoints_found and not any(
-        (root / n).exists() for n in ("tailwind.config.js", "tailwind.config.ts", "tailwind.config.mjs")
-    ):
+    if not breakpoints_found and not coverage["tailwind_detected"]:
         issues.append({
             "severity": "warning",
             "message": "No CSS breakpoints or Tailwind config detected",
@@ -574,15 +912,30 @@ def design_validate_responsive(
                 "fix": "Use max-width or responsive units instead",
             })
-    return {
+    # Status taxonomy (LED-1010)
+    gaps: List[str] = []
+    if coverage["tailwind_detected"] and not coverage["jsx_scanned"]:
+        gaps.append("Tailwind detected but no JSX/TSX/Vue/Svelte files found — responsive coverage may be underreported")
+    if not coverage["css_scanned"] and not coverage["tailwind_detected"]:
+        gaps.append("No CSS and no Tailwind detected — nothing to validate against")
+    status = STATUS_PARTIAL_COVERAGE if gaps else STATUS_OK
+    result = {
         "tool": "design.validate_responsive",
-        "status": "ok",
+        "status": status,
         "breakpoints_found": breakpoints_found,
         "responsive_issues": issues,
         "viewport_meta": viewport_meta,
+        "viewport_blocks_zoom": viewport_blocks_zoom,
         "responsive_units_count": responsive_units_count,
-        "mobile_first": min_width_count >= max_width_count,
+        "mobile_first": mobile_first,
+        "coverage": coverage,
+        "gaps": gaps,
     }
+    if tw_stats:
+        result["tailwind_stats"] = tw_stats
+    return result
 # ---------------------------------------------------------------------------
@@ -764,11 +1117,31 @@ def _puppeteer_screenshot_fallback(url: str, baselines_dir: Path) -> Dict[str, A
             timeout=30,
         )
         if result.returncode != 0:
+            stderr = result.stderr.decode(errors="replace")[:500]
+            # LED-1010: distinguish "module not found" (toolchain missing) from
+            # runtime errors (toolchain present but failed). Consumers need
+            # this distinction to know whether to install or debug.
+            if "Cannot find module 'puppeteer'" in stderr or "Error: Cannot find module" in stderr:
+                return {
+                    "tool": "story.visual_test",
+                    "status": STATUS_TOOLCHAIN_MISSING,
+                    "missing": ["puppeteer", "playwright"],
+                    "error": stderr,
+                    "install_commands": [
+                        "pip install playwright && python -m playwright install chromium",
+                        "npm install -g puppeteer",
+                    ],
+                    "hint": (
+                        "No screenshot engine available. Install Playwright (preferred) "
+                        "or Puppeteer. After install, re-run — the tool auto-detects."
+                    ),
+                }
             return {
                 "tool": "story.visual_test",
-                "status": "puppeteer_error",
-                "error": result.stderr.decode(errors="replace")[:500],
-                "hint": "Puppeteer fallback failed. Install Playwright for better support: pip install playwright && python -m playwright install chromium",
+                "status": "error",
+                "engine": "puppeteer",
+                "error": stderr,
+                "hint": "Puppeteer is installed but the screenshot call failed at runtime. Check the URL reachability, sandbox permissions, or memory limits.",
             }
         return {
@@ -813,9 +1186,14 @@ def story_visual_test(
         return {
             "tool": "story.visual_test",
-            "status": "no_screenshot_tool",
+            "status": STATUS_TOOLCHAIN_MISSING,
+            "missing": ["playwright", "puppeteer"],
+            "install_commands": [
+                "pip install playwright && python -m playwright install chromium",
+                "npm install -g puppeteer",
+            ],
             "message": (
-                "No screenshot tool available. Install one of the following:\n"
+                "No screenshot engine available. Install one:\n"
                 "  - Playwright (recommended): pip install playwright && python -m playwright install chromium\n"
                 "  - Puppeteer (fallback): npm install -g puppeteer"
             ),
@@ -909,19 +1287,76 @@ def story_visual_test(
 # 26. story_accessibility
 # ---------------------------------------------------------------------------
-_IMG_NO_ALT_RE = re.compile(r"<img(?![^>]*alt=)[^>]*>", re.IGNORECASE)
-_INPUT_NO_LABEL_RE = re.compile(r"<input(?![^>]*(?:aria-label|aria-labelledby|id=)[^>]*>)[^>]*>", re.IGNORECASE)
-_BUTTON_EMPTY_RE = re.compile(r"<button[^>]*>\s*</button>", re.IGNORECASE)
-_A_NO_HREF_RE = re.compile(r"<a(?![^>]*href=)[^>]*>", re.IGNORECASE)
-_HEADING_SKIP_RE = re.compile(r"<h([1-6])")
-_ARIA_HIDDEN_FOCUSABLE_RE = re.compile(r'aria-hidden=["\']true["\'][^>]*(?:tabindex=["\']0["\']|<button|<a\s)', re.IGNORECASE)
+# LED-1010 FIX: the original patterns used `<tag` with re.IGNORECASE but no
+# word boundary after the tag name. That meant `<ArrowLeft>` matched `<a` +
+# `rrowLeft>`, producing 128 false-positive link-href "errors" on a single
+# DomainVested scan. Require the character AFTER the tag name to be
+# whitespace, `/`, or `>` so PascalCase React components can't collide with
+# HTML anchors / images / inputs / buttons.
+_IMG_NO_ALT_RE = re.compile(r"<img(?=[\s/>])(?![^>]*\salt=)[^>]*>", re.IGNORECASE)
+_INPUT_NO_LABEL_RE = re.compile(r"<input(?=[\s/>])(?![^>]*(?:\saria-label|\saria-labelledby|\sid=|type=[\"']hidden[\"']))[^>]*>", re.IGNORECASE)
+_BUTTON_EMPTY_RE = re.compile(r"<button(?=[\s>])[^>]*>\s*</button>", re.IGNORECASE)
+_A_NO_HREF_RE = re.compile(r"<a(?=[\s/>])(?![^>]*\shref=)[^>]*>", re.IGNORECASE)
+_HEADING_SKIP_RE = re.compile(r"<h([1-6])(?=[\s/>])")
+_ARIA_HIDDEN_FOCUSABLE_RE = re.compile(r'aria-hidden=["\']true["\'][^>]*(?:tabindex=["\']0["\']|<button(?=[\s>])|<a(?=[\s>]))', re.IGNORECASE)
+# LED-1010: WCAG coverage map. The old tool accepted standards="WCAG2AA"
+# but only implemented 3 of ~50 AA rules, and stamped every issue "WCAG2A"
+# regardless. This map declares exactly which rules the scanner covers so
+# the response can surface `coverage` honestly and `standard_requested` vs
+# `standard_implemented` are separate fields.
+#
+# Source references per rule: https://www.w3.org/WAI/WCAG21/quickref/
+IMPLEMENTED_WCAG_RULES = {
+    "img-alt":                {"criterion": "1.1.1", "level": "A",  "title": "Non-text Content"},
+    "input-label":            {"criterion": "1.3.1", "level": "A",  "title": "Info and Relationships"},
+    "button-content":         {"criterion": "4.1.2", "level": "A",  "title": "Name, Role, Value"},
+    "link-href":              {"criterion": "2.4.4", "level": "A",  "title": "Link Purpose (In Context)"},
+    "heading-order":          {"criterion": "1.3.1", "level": "A",  "title": "Info and Relationships"},
+    "aria-hidden-focusable":  {"criterion": "4.1.2", "level": "A",  "title": "Name, Role, Value"},
+    "viewport-zoom-blocked":  {"criterion": "1.4.4", "level": "AA", "title": "Resize Text"},
+}
+# Rules that exist in each level but we DON'T implement — surfaced as gaps
+# so consumers know coverage is partial.
+UNIMPLEMENTED_WCAG_RULES_AA = [
+    {"criterion": "1.4.3", "level": "AA",  "title": "Contrast (Minimum)"},
+    {"criterion": "2.4.7", "level": "AA",  "title": "Focus Visible"},
+    {"criterion": "2.1.2", "level": "A",   "title": "No Keyboard Trap"},
+    {"criterion": "2.3.3", "level": "AAA", "title": "Animation from Interactions"},
+    {"criterion": "3.1.1", "level": "A",   "title": "Language of Page"},
+    {"criterion": "1.4.1", "level": "A",   "title": "Use of Color"},
+    {"criterion": "1.3.5", "level": "AA",  "title": "Identify Input Purpose"},
+    {"criterion": "3.3.1", "level": "A",   "title": "Error Identification"},
+    {"criterion": "3.3.2", "level": "A",   "title": "Labels or Instructions"},
+    {"criterion": "2.5.3", "level": "A",   "title": "Label in Name"},
+    {"criterion": "2.5.2", "level": "A",   "title": "Pointer Cancellation"},
+    {"criterion": "4.1.3", "level": "AA",  "title": "Status Messages"},
+]
+def _stamp_rule(rule: str) -> str:
+    """Return the actual WCAG level the rule enforces — not the caller's request."""
+    info = IMPLEMENTED_WCAG_RULES.get(rule)
+    if not info:
+        return "WCAG2A"
+    level = info["level"]
+    return "WCAG2A" if level == "A" else ("WCAG2AA" if level == "AA" else "WCAG2AAA")
 def story_accessibility(
     project_path: str,
     standards: str = "WCAG2AA",
 ) -> Dict[str, Any]:
-    """Run accessibility checks by scanning HTML/JSX/TSX for common issues."""
+    """Run accessibility checks by scanning HTML/JSX/TSX for common issues.
+    LED-1010: issues are now stamped with the ACTUAL WCAG level the rule
+    enforces (not the caller's requested level). `standard_requested`,
+    `implemented_rules`, `unimplemented_rules`, and `coverage_percent`
+    surface exactly what ran vs what's still in the standard, so a caller
+    asking for WCAG2AA does not see a false-confident pass.
+    """
     root = Path(project_path)
     if not root.is_dir():
         return {"tool": "story.accessibility", "error": f"Directory not found: {root}"}
@@ -936,21 +1371,21 @@ def story_accessibility(
         files_checked += 1
         rel = str(f.relative_to(root)) if f.is_relative_to(root) else str(f)
-        # Missing alt on images
+        # Missing alt on images (WCAG 1.1.1, Level A)
         for m in _IMG_NO_ALT_RE.finditer(text):
             issues.append({
                 "rule": "img-alt",
                 "severity": "error",
                 "message": "Image missing alt attribute",
                 "file": rel,
-                "standard": "WCAG2A",
+                "standard": _stamp_rule("img-alt"),
+                "wcag": IMPLEMENTED_WCAG_RULES["img-alt"],
                 "snippet": m.group()[:120],
             })
-        # Inputs without labels
+        # Inputs without labels (WCAG 1.3.1, Level A)
         for m in _INPUT_NO_LABEL_RE.finditer(text):
             snippet = m.group()
-            # Skip hidden inputs
             if 'type="hidden"' in snippet or "type='hidden'" in snippet:
                 continue
             issues.append({
@@ -958,33 +1393,37 @@ def story_accessibility(
                 "severity": "error",
                 "message": "Input missing associated label or aria-label",
                 "file": rel,
-                "standard": "WCAG2A",
+                "standard": _stamp_rule("input-label"),
+                "wcag": IMPLEMENTED_WCAG_RULES["input-label"],
                 "snippet": snippet[:120],
             })
-        # Empty buttons
+        # Empty buttons (WCAG 4.1.2, Level A)
         for m in _BUTTON_EMPTY_RE.finditer(text):
             issues.append({
                 "rule": "button-content",
                 "severity": "error",
                 "message": "Button has no text content or aria-label",
                 "file": rel,
-                "standard": "WCAG2A",
+                "standard": _stamp_rule("button-content"),
+                "wcag": IMPLEMENTED_WCAG_RULES["button-content"],
                 "snippet": m.group()[:120],
             })
-        # Links without href
+        # Links without href (WCAG 2.4.4, Level A) — regex fixed to not
+        # false-match PascalCase JSX components (`<ArrowLeft />` etc).
         for m in _A_NO_HREF_RE.finditer(text):
             issues.append({
                 "rule": "link-href",
                 "severity": "warning",
                 "message": "Anchor element missing href attribute",
                 "file": rel,
-                "standard": "WCAG2A",
+                "standard": _stamp_rule("link-href"),
+                "wcag": IMPLEMENTED_WCAG_RULES["link-href"],
                 "snippet": m.group()[:120],
             })
-        # Heading level skips (e.g., h1 -> h3 without h2)
+        # Heading level skips (WCAG 1.3.1, Level A)
         headings = [int(h) for h in _HEADING_SKIP_RE.findall(text)]
         for i in range(1, len(headings)):
             if headings[i] > headings[i - 1] + 1:
@@ -993,21 +1432,23 @@ def story_accessibility(
                     "severity": "warning",
                     "message": f"Heading level skipped: h{headings[i-1]} to h{headings[i]}",
                     "file": rel,
-                    "standard": "WCAG2A",
+                    "standard": _stamp_rule("heading-order"),
+                    "wcag": IMPLEMENTED_WCAG_RULES["heading-order"],
                 })
-        # aria-hidden on focusable elements
+        # aria-hidden on focusable elements (WCAG 4.1.2, Level A)
         for m in _ARIA_HIDDEN_FOCUSABLE_RE.finditer(text):
             issues.append({
                 "rule": "aria-hidden-focusable",
                 "severity": "error",
                 "message": "Focusable element has aria-hidden='true'",
                 "file": rel,
-                "standard": "WCAG2AA",
+                "standard": _stamp_rule("aria-hidden-focusable"),
+                "wcag": IMPLEMENTED_WCAG_RULES["aria-hidden-focusable"],
                 "snippet": m.group()[:120],
             })
-    # Filter by standard level if needed
+    # Filter by requested standard level
     standard_levels = {"WCAG2A": 1, "WCAG2AA": 2, "WCAG2AAA": 3}
     requested_level = standard_levels.get(standards, 2)
     filtered = [i for i in issues if standard_levels.get(i.get("standard", "WCAG2A"), 1) <= requested_level]
@@ -1015,11 +1456,50 @@ def story_accessibility(
     errors = [i for i in filtered if i["severity"] == "error"]
     warnings = [i for i in filtered if i["severity"] == "warning"]
+    # LED-1010: group by (rule, file) so 77 input-label errors across 30+
+    # files can be triaged as ~30 groups with counts rather than 77 individual
+    # call-sites in the caller's inbox.
+    groups: Dict[tuple, Dict[str, Any]] = {}
+    for issue in filtered:
+        key = (issue["rule"], issue.get("file", ""))
+        if key not in groups:
+            groups[key] = {
+                "rule": issue["rule"],
+                "file": issue.get("file", ""),
+                "count": 0,
+                "severity": issue["severity"],
+                "standard": issue.get("standard", "WCAG2A"),
+            }
+        groups[key]["count"] += 1
+    # LED-1010 coverage: we implement ~7 rules; WCAG2AA covers many more.
+    # Count rules at/below the requested level to be honest about coverage.
+    implemented_at_level = [
+        r for r, info in IMPLEMENTED_WCAG_RULES.items()
+        if standard_levels.get("WCAG2" + info["level"], 1) <= requested_level
+    ]
+    unimplemented_at_level = [
+        r for r in UNIMPLEMENTED_WCAG_RULES_AA
+        if standard_levels.get("WCAG2" + r["level"], 1) <= requested_level
+    ]
+    total_rules = max(1, len(implemented_at_level) + len(unimplemented_at_level))
+    coverage_pct = len(implemented_at_level) * 100 // total_rules
+    # Status: partial_coverage when coverage < 100%. A `status: ok` on a scan
+    # that ran 7 of ~50 WCAG2AA rules is exactly the silent-false-confidence
+    # failure mode LED-1010 flagged.
+    status = STATUS_PARTIAL_COVERAGE if coverage_pct < 100 else STATUS_OK
     return {
         "tool": "story.accessibility",
-        "status": "ok",
-        "standard": standards,
+        "status": status,
+        "standard_requested": standards,
+        "standard": standards,  # retained for back-compat
+        "implemented_rules": implemented_at_level,
+        "unimplemented_rules": unimplemented_at_level,
+        "coverage_percent": coverage_pct,
         "issues": filtered,
+        "groups": sorted(groups.values(), key=lambda g: (-g["count"], g["file"])),
         "passed_count": files_checked - len(set(i["file"] for i in errors)),
         "failed_count": len(set(i["file"] for i in errors)),
         "error_count": len(errors),