npm - genoma-evolution - Versions diffs - 1.0.0 - Mend

genoma-evolution 1.0.0

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 (445) hide show

package/.claude/worktrees/phase-2-mcp/stitch/curator_workspace/code.html ADDED Viewed

@@ -0,0 +1,448 @@
+<!DOCTYPE html>
+<html class="light" lang="en"><head>
+<meta charset="utf-8"/>
+<meta content="width=device-width, initial-scale=1.0" name="viewport"/>
+<title>Hermes - Curator Workspace</title>
+<script src="https://cdn.tailwindcss.com?plugins=forms,container-queries"></script>
+<link href="https://fonts.googleapis.com/css2?family=Inter:wght@400;500;600;700;900&amp;family=JetBrains+Mono:wght@400;500;700&amp;display=swap" rel="stylesheet"/>
+<link href="https://fonts.googleapis.com/css2?family=Material+Symbols+Outlined:wght,FILL@100..700,0..1&amp;display=swap" rel="stylesheet"/>
+<link href="https://fonts.googleapis.com/css2?family=Material+Symbols+Outlined:wght,FILL@100..700,0..1&amp;display=swap" rel="stylesheet"/>
+<script id="tailwind-config">
+        tailwind.config = {
+            darkMode: "class",
+            theme: {
+                extend: {
+                    "colors": {
+                        "on-background": "#1a1c1e",
+                        "surface-tint": "#446083",
+                        "on-tertiary-container": "#8b8a89",
+                        "surface-container": "#eeedf1",
+                        "inverse-surface": "#2f3033",
+                        "tertiary-container": "#232323",
+                        "on-surface": "#1a1c1e",
+                        "on-primary-fixed-variant": "#2c486a",
+                        "outline": "#73777f",
+                        "surface-container-high": "#e9e8eb",
+                        "surface-bright": "#faf9fc",
+                        "on-tertiary-fixed-variant": "#474746",
+                        "on-secondary-fixed": "#380d00",
+                        "tertiary": "#0d0d0d",
+                        "surface": "#faf9fc",
+                        "outline-variant": "#c3c6cf",
+                        "on-primary-fixed": "#001c37",
+                        "surface-container-highest": "#e3e2e5",
+                        "primary-fixed-dim": "#acc9f1",
+                        "primary-fixed": "#d2e4ff",
+                        "on-secondary": "#ffffff",
+                        "on-primary": "#ffffff",
+                        "surface-container-lowest": "#ffffff",
+                        "surface-dim": "#dad9dd",
+                        "primary-container": "#002444",
+                        "error-container": "#ffdad6",
+                        "secondary-fixed-dim": "#ffb59b",
+                        "secondary-fixed": "#ffdbcf",
+                        "on-surface-variant": "#43474e",
+                        "inverse-on-surface": "#f1f0f3",
+                        "secondary-container": "#fe753f",
+                        "on-tertiary": "#ffffff",
+                        "primary": "#000e20",
+                        "on-secondary-container": "#631d00",
+                        "on-error": "#ffffff",
+                        "secondary": "#a93800",
+                        "error": "#ba1a1a",
+                        "surface-variant": "#e3e2e5",
+                        "on-primary-container": "#708cb2",
+                        "tertiary-fixed-dim": "#c8c6c5",
+                        "tertiary-fixed": "#e4e2e1",
+                        "on-tertiary-fixed": "#1b1c1b",
+                        "surface-container-low": "#f4f3f6",
+                        "on-secondary-fixed-variant": "#812900",
+                        "on-error-container": "#93000a",
+                        "background": "#faf9fc",
+                        "inverse-primary": "#acc9f1"
+                    },
+                    "borderRadius": {
+                        "DEFAULT": "0.125rem",
+                        "lg": "0.25rem",
+                        "xl": "0.5rem",
+                        "full": "0.75rem"
+                    },
+                    "spacing": {
+                        "unit": "8px",
+                        "stack-tight": "4px",
+                        "container-padding": "24px",
+                        "max-width": "1440px",
+                        "gutter": "16px",
+                        "component-gap": "8px"
+                    },
+                    "fontFamily": {
+                        "body-sm": ["Inter"],
+                        "body-md": ["Inter"],
+                        "body-lg": ["Inter"],
+                        "mono-data": ["JetBrains Mono"],
+                        "headline-sm": ["Inter"],
+                        "label-caps": ["Inter"],
+                        "headline-lg": ["Inter"],
+                        "mono-label": ["JetBrains Mono"],
+                        "headline-md": ["Inter"]
+                    },
+                    "fontSize": {
+                        "body-sm": ["12px", { "lineHeight": "16px", "fontWeight": "400" }],
+                        "body-md": ["14px", { "lineHeight": "20px", "fontWeight": "400" }],
+                        "body-lg": ["16px", { "lineHeight": "24px", "fontWeight": "400" }],
+                        "mono-data": ["12px", { "lineHeight": "18px", "fontWeight": "400" }],
+                        "headline-sm": ["20px", { "lineHeight": "28px", "fontWeight": "600" }],
+                        "label-caps": ["11px", { "lineHeight": "16px", "letterSpacing": "0.05em", "fontWeight": "700" }],
+                        "headline-lg": ["32px", { "lineHeight": "40px", "letterSpacing": "-0.02em", "fontWeight": "600" }],
+                        "mono-label": ["13px", { "lineHeight": "16px", "letterSpacing": "0.02em", "fontWeight": "500" }],
+                        "headline-md": ["24px", { "lineHeight": "32px", "letterSpacing": "-0.01em", "fontWeight": "600" }]
+                    }
+                }
+            }
+        }
+    </script>
+<style>
+        /* Custom scrollbar for dense data views */
+        ::-webkit-scrollbar { width: 6px; height: 6px; }
+        ::-webkit-scrollbar-track { background: transparent; }
+        ::-webkit-scrollbar-thumb { background: #c3c6cf; border-radius: 4px; }
+        ::-webkit-scrollbar-thumb:hover { background: #73777f; }
+        .sparkline-bar {
+            width: 4px;
+            background-color: #446083;
+            border-radius: 1px;
+            display: inline-block;
+            margin-right: 2px;
+        }
+    </style>
+</head>
+<body class="bg-background text-on-background font-body-md antialiased overflow-hidden h-screen flex w-full">
+<!-- SideNavBar Component -->
+<nav class="hidden md:flex flex-col docked left-0 w-64 h-screen border-r border-outline-variant/30 bg-surface-container-low py-container-padding gap-component-gap shrink-0 z-40">
+<!-- Header -->
+<div class="px-gutter mb-6 flex items-center gap-3">
+<div class="w-8 h-8 rounded bg-primary text-on-primary flex items-center justify-center font-headline-sm font-bold shadow-sm">
+                H
+            </div>
+<div>
+<h1 class="font-headline-sm text-headline-sm font-black text-primary">Hermes Control</h1>
+<p class="font-mono-data text-mono-data text-on-surface-variant opacity-80">v4.2.0-Alpha</p>
+</div>
+</div>
+<!-- CTA -->
+<div class="px-gutter mb-4">
+<button class="w-full bg-primary text-on-primary font-body-sm text-body-sm py-2 rounded-DEFAULT flex items-center justify-center gap-2 hover:bg-surface-tint transition-colors">
+<span class="material-symbols-outlined" style="font-size: 16px;">add</span>
+                New Experiment
+            </button>
+</div>
+<!-- Navigation Links -->
+<div class="flex flex-col flex-1 gap-1 px-2">
+<!-- Inactive -->
+<a class="flex items-center gap-3 px-3 py-2 rounded-DEFAULT text-on-surface-variant hover:text-primary hover:bg-surface-container transition-all font-label-caps text-label-caps tracking-widest group" href="#">
+<span class="material-symbols-outlined text-[18px] opacity-70 group-hover:opacity-100 transition-opacity">dashboard</span>
+                Overview
+            </a>
+<a class="flex items-center gap-3 px-3 py-2 rounded-DEFAULT text-on-surface-variant hover:text-primary hover:bg-surface-container transition-all font-label-caps text-label-caps tracking-widest group" href="#">
+<span class="material-symbols-outlined text-[18px] opacity-70 group-hover:opacity-100 transition-opacity">extension</span>
+                Skill Hub
+            </a>
+<a class="flex items-center gap-3 px-3 py-2 rounded-DEFAULT text-on-surface-variant hover:text-primary hover:bg-surface-container transition-all font-label-caps text-label-caps tracking-widest group" href="#">
+<span class="material-symbols-outlined text-[18px] opacity-70 group-hover:opacity-100 transition-opacity">auto_awesome</span>
+                Evolution
+            </a>
+<a class="flex items-center gap-3 px-3 py-2 rounded-DEFAULT text-on-surface-variant hover:text-primary hover:bg-surface-container transition-all font-label-caps text-label-caps tracking-widest group" href="#">
+<span class="material-symbols-outlined text-[18px] opacity-70 group-hover:opacity-100 transition-opacity">database</span>
+                Datasets
+            </a>
+<a class="flex items-center gap-3 px-3 py-2 rounded-DEFAULT text-on-surface-variant hover:text-primary hover:bg-surface-container transition-all font-label-caps text-label-caps tracking-widest group" href="#">
+<span class="material-symbols-outlined text-[18px] opacity-70 group-hover:opacity-100 transition-opacity">monitoring</span>
+                Metrics
+            </a>
+<a class="flex items-center gap-3 px-3 py-2 rounded-DEFAULT text-on-surface-variant hover:text-primary hover:bg-surface-container transition-all font-label-caps text-label-caps tracking-widest group" href="#">
+<span class="material-symbols-outlined text-[18px] opacity-70 group-hover:opacity-100 transition-opacity">terminal</span>
+                Live Logs
+            </a>
+<!-- ACTIVE: Curator -->
+<a class="flex items-center gap-3 px-3 py-2 text-primary border-r-4 border-secondary bg-surface-container-highest font-label-caps text-label-caps tracking-widest opacity-80 scale-95 transition-all" href="#">
+<span class="material-symbols-outlined text-[18px]" style="font-variation-settings: 'FILL' 1;">inventory_2</span>
+                Curator
+            </a>
+<a class="flex items-center gap-3 px-3 py-2 rounded-DEFAULT text-on-surface-variant hover:text-primary hover:bg-surface-container transition-all font-label-caps text-label-caps tracking-widest group mt-auto" href="#">
+<span class="material-symbols-outlined text-[18px] opacity-70 group-hover:opacity-100 transition-opacity">settings</span>
+                Settings
+            </a>
+</div>
+</nav>
+<!-- Main Content Wrapper -->
+<div class="flex-1 flex flex-col min-w-0 bg-background relative">
+<!-- TopAppBar Component -->
+<header class="docked full-width top-0 border-b border-outline-variant/30 bg-surface flex justify-between items-center w-full px-gutter h-16 z-50 shrink-0">
+<div class="flex items-center gap-4">
+<div class="font-headline-sm text-headline-sm font-bold tracking-tighter text-primary flex items-center gap-2">
+                    Hermes <span class="text-on-surface-variant font-normal opacity-50">/</span> <span class="font-body-lg text-body-lg text-on-surface font-medium">Review Queue</span>
+</div>
+</div>
+<div class="flex items-center gap-4">
+<div class="flex items-center gap-2 text-on-surface-variant font-label-caps text-label-caps bg-surface-container py-1 px-2 rounded-DEFAULT border border-outline-variant/20">
+<span class="w-2 h-2 rounded-full bg-surface-tint animate-pulse"></span>
+                    Syncing
+                </div>
+<button class="w-8 h-8 flex items-center justify-center text-on-surface-variant hover:bg-surface-container-high transition-colors rounded-DEFAULT">
+<span class="material-symbols-outlined">sync</span>
+</button>
+<button class="w-8 h-8 flex items-center justify-center text-on-surface-variant hover:bg-surface-container-high transition-colors rounded-DEFAULT relative">
+<span class="material-symbols-outlined">notifications</span>
+<span class="absolute top-1 right-1 w-2 h-2 bg-secondary rounded-full border border-surface"></span>
+</button>
+<div class="w-8 h-8 rounded-full bg-primary-fixed-dim border border-outline-variant/30 flex items-center justify-center overflow-hidden ml-2">
+<span class="material-symbols-outlined text-primary text-[20px]">person</span>
+</div>
+</div>
+</header>
+<!-- Workspace Split Layout -->
+<main class="flex-1 flex flex-row overflow-hidden relative">
+<!-- Left Pane: Queue List (High Density) -->
+<aside class="w-[380px] flex flex-col border-r border-outline-variant/30 bg-surface-container-lowest shrink-0 z-10">
+<!-- Queue Tools -->
+<div class="p-4 border-b border-outline-variant/20 flex flex-col gap-3 bg-surface">
+<div class="flex items-center justify-between">
+<h2 class="font-mono-label text-mono-label text-on-surface">PENDING_REVIEWS (14)</h2>
+<button class="text-on-surface-variant hover:text-primary">
+<span class="material-symbols-outlined text-[18px]">filter_list</span>
+</button>
+</div>
+<div class="relative">
+<span class="material-symbols-outlined absolute left-2 top-1.5 text-[16px] text-on-surface-variant">search</span>
+<input class="w-full bg-surface-container-low border border-outline-variant/30 rounded-DEFAULT py-1.5 pl-8 pr-3 font-body-sm text-body-sm text-on-surface focus:outline-none focus:border-primary transition-colors" placeholder="Filter by ID or Skill..." type="text"/>
+</div>
+</div>
+<!-- Queue Items -->
+<div class="flex-1 overflow-y-auto p-2 flex flex-col gap-1">
+<!-- Active Item -->
+<div class="p-3 bg-surface-container border border-surface-tint/30 rounded-DEFAULT cursor-pointer relative overflow-hidden group">
+<div class="absolute left-0 top-0 bottom-0 w-1 bg-primary"></div>
+<div class="flex justify-between items-start mb-1">
+<span class="font-mono-data text-mono-data text-on-surface-variant text-[10px]">EVO-9921</span>
+<span class="font-mono-data text-mono-data text-surface-tint text-[10px] bg-primary-fixed-dim/20 px-1 rounded">NLP_CORE</span>
+</div>
+<h3 class="font-body-md text-body-md font-medium text-on-surface mb-2 leading-tight">Optimize Intent Classification Constraints</h3>
+<div class="flex items-center justify-between mt-3 pt-2 border-t border-outline-variant/20">
+<div class="flex flex-col gap-0.5">
+<span class="font-label-caps text-[9px] text-on-surface-variant">CONFIDENCE</span>
+<div class="flex items-center gap-1 font-mono-data text-mono-data text-on-surface">
+<span class="w-1.5 h-1.5 rounded-full bg-surface-tint"></span> 94.2%
+                                </div>
+</div>
+<div class="flex flex-col gap-0.5">
+<span class="font-label-caps text-[9px] text-on-surface-variant">RISK</span>
+<div class="flex items-center gap-1 font-mono-data text-mono-data text-on-surface">
+<span class="material-symbols-outlined text-[12px] text-surface-tint">check_circle</span> Low
+                                </div>
+</div>
+<div class="flex flex-col gap-0.5 items-end">
+<span class="font-label-caps text-[9px] text-on-surface-variant">IMPACT</span>
+<div class="font-mono-data text-mono-data text-surface-tint font-medium">
+                                    +1.4% F1
+                                </div>
+</div>
+</div>
+</div>
+<!-- Inactive Item 1 -->
+<div class="p-3 bg-surface-container-lowest border border-outline-variant/10 rounded-DEFAULT cursor-pointer hover:bg-surface-container-low transition-colors">
+<div class="flex justify-between items-start mb-1">
+<span class="font-mono-data text-mono-data text-on-surface-variant text-[10px] opacity-70">EVO-9918</span>
+<span class="font-mono-data text-mono-data text-on-surface-variant text-[10px] bg-surface-variant px-1 rounded">DATA_PIPE</span>
+</div>
+<h3 class="font-body-md text-body-md text-on-surface mb-2 leading-tight opacity-90">Refactor Entity Extraction Regex Fallbacks</h3>
+<div class="flex items-center justify-between mt-3 pt-2 border-t border-outline-variant/10 opacity-80">
+<div class="flex flex-col gap-0.5">
+<span class="font-label-caps text-[9px] text-on-surface-variant">CONFIDENCE</span>
+<div class="flex items-center gap-1 font-mono-data text-mono-data text-on-surface">
+<span class="w-1.5 h-1.5 rounded-full bg-secondary-container"></span> 82.1%
+                                </div>
+</div>
+<div class="flex flex-col gap-0.5">
+<span class="font-label-caps text-[9px] text-on-surface-variant">RISK</span>
+<div class="flex items-center gap-1 font-mono-data text-mono-data text-secondary-container">
+<span class="material-symbols-outlined text-[12px]">warning</span> Med
+                                </div>
+</div>
+<div class="flex flex-col gap-0.5 items-end">
+<span class="font-label-caps text-[9px] text-on-surface-variant">IMPACT</span>
+<div class="font-mono-data text-mono-data text-on-surface">
+                                    -12ms Lat.
+                                </div>
+</div>
+</div>
+</div>
+<!-- Inactive Item 2 -->
+<div class="p-3 bg-surface-container-lowest border border-outline-variant/10 rounded-DEFAULT cursor-pointer hover:bg-surface-container-low transition-colors">
+<div class="flex justify-between items-start mb-1">
+<span class="font-mono-data text-mono-data text-on-surface-variant text-[10px] opacity-70">EVO-9890</span>
+<span class="font-mono-data text-mono-data text-on-surface-variant text-[10px] bg-surface-variant px-1 rounded">SYS_PROMPT</span>
+</div>
+<h3 class="font-body-md text-body-md text-on-surface mb-2 leading-tight opacity-90">Adjust Tone Constraints for Error States</h3>
+<div class="flex items-center justify-between mt-3 pt-2 border-t border-outline-variant/10 opacity-80">
+<div class="flex flex-col gap-0.5">
+<span class="font-label-caps text-[9px] text-on-surface-variant">CONFIDENCE</span>
+<div class="flex items-center gap-1 font-mono-data text-mono-data text-on-surface">
+<span class="w-1.5 h-1.5 rounded-full bg-surface-tint"></span> 98.9%
+                                </div>
+</div>
+<div class="flex flex-col gap-0.5">
+<span class="font-label-caps text-[9px] text-on-surface-variant">RISK</span>
+<div class="flex items-center gap-1 font-mono-data text-mono-data text-on-surface">
+<span class="material-symbols-outlined text-[12px] text-surface-tint">check_circle</span> Low
+                                </div>
+</div>
+<div class="flex flex-col gap-0.5 items-end">
+<span class="font-label-caps text-[9px] text-on-surface-variant">IMPACT</span>
+<div class="font-mono-data text-mono-data text-on-surface">
+                                    +0.2 CSAT
+                                </div>
+</div>
+</div>
+</div>
+</div>
+</aside>
+<!-- Right Pane: Detail View -->
+<section class="flex-1 flex flex-col bg-surface overflow-hidden relative">
+<!-- Detail Header & Action Bar -->
+<div class="px-8 py-6 border-b border-outline-variant/20 bg-surface-bright flex justify-between items-start shrink-0">
+<div class="max-w-2xl">
+<div class="flex items-center gap-3 mb-2">
+<span class="font-mono-data text-mono-data text-on-surface bg-surface-container px-2 py-0.5 rounded border border-outline-variant/30">EVO-9921</span>
+<span class="font-label-caps text-label-caps text-surface-tint flex items-center gap-1">
+<span class="material-symbols-outlined text-[14px]">auto_fix_high</span> Auto-Generated
+                            </span>
+<span class="text-outline-variant">|</span>
+<span class="font-body-sm text-body-sm text-on-surface-variant">Generated 2 hrs ago by LogicEngine-v2</span>
+</div>
+<h1 class="font-headline-md text-headline-md text-primary mb-2">Optimize Intent Classification Constraints</h1>
+<p class="font-body-md text-body-md text-on-surface-variant">
+                            The system proposes tightening the negative constraints in the core intent routing prompt to reduce false positives in out-of-domain queries. Regression testing shows significant F1 improvement without impacting latency.
+                        </p>
+</div>
+<!-- Actions -->
+<div class="flex items-center gap-3 shrink-0">
+<button class="px-4 py-2 rounded-DEFAULT border border-secondary text-secondary font-mono-label text-mono-label hover:bg-error-container/20 transition-colors flex items-center gap-2">
+<span class="material-symbols-outlined text-[18px]">close</span> Reject
+                        </button>
+<button class="px-4 py-2 rounded-DEFAULT border border-outline-variant text-on-surface hover:bg-surface-container transition-colors font-mono-label text-mono-label flex items-center gap-2">
+<span class="material-symbols-outlined text-[18px]">refresh</span> Retry Test
+                        </button>
+<button class="px-6 py-2 rounded-DEFAULT bg-primary text-on-primary font-mono-label text-mono-label hover:bg-surface-tint transition-colors flex items-center gap-2 shadow-sm">
+<span class="material-symbols-outlined text-[18px]">check</span> Approve Merge
+                        </button>
+</div>
+</div>
+<!-- Detail Content (Scrollable Bento) -->
+<div class="flex-1 overflow-y-auto p-8 flex flex-col gap-6 bg-background">
+<!-- Metrics Row (Stat Cards) -->
+<div class="grid grid-cols-3 gap-4">
+<!-- Stat Card 1 -->
+<div class="bg-surface border border-outline-variant/30 rounded-lg p-4 flex flex-col">
+<span class="font-label-caps text-label-caps text-on-surface-variant mb-1">PREDICTED F1 SCORE</span>
+<div class="flex items-baseline gap-2 mb-2">
+<span class="font-headline-lg font-mono-data text-primary">0.942</span>
+<span class="font-mono-data text-mono-data text-surface-tint bg-primary-fixed-dim/20 px-1 rounded">+0.014</span>
+</div>
+<div class="mt-auto h-4 flex items-end gap-0.5 opacity-50">
+<div class="sparkline-bar h-1"></div><div class="sparkline-bar h-1.5"></div><div class="sparkline-bar h-2"></div>
+<div class="sparkline-bar h-2"></div><div class="sparkline-bar h-3"></div><div class="sparkline-bar h-full bg-primary opacity-100"></div>
+</div>
+</div>
+<!-- Stat Card 2 -->
+<div class="bg-surface border border-outline-variant/30 rounded-lg p-4 flex flex-col">
+<span class="font-label-caps text-label-caps text-on-surface-variant mb-1">P95 LATENCY IMPACT</span>
+<div class="flex items-baseline gap-2 mb-2">
+<span class="font-headline-lg font-mono-data text-primary">124<span class="text-body-md">ms</span></span>
+<span class="font-mono-data text-mono-data text-on-surface-variant bg-surface-container px-1 rounded">No Change</span>
+</div>
+<div class="mt-auto h-4 flex items-end gap-0.5 opacity-30">
+<div class="sparkline-bar h-2"></div><div class="sparkline-bar h-2"></div><div class="sparkline-bar h-2"></div>
+<div class="sparkline-bar h-2"></div><div class="sparkline-bar h-2"></div><div class="sparkline-bar h-2 bg-on-surface"></div>
+</div>
+</div>
+<!-- Stat Card 3 -->
+<div class="bg-surface border border-outline-variant/30 rounded-lg p-4 flex flex-col">
+<span class="font-label-caps text-label-caps text-on-surface-variant mb-1">TEST COVERAGE / OOD</span>
+<div class="flex items-baseline gap-2 mb-2">
+<span class="font-headline-lg font-mono-data text-primary">99.1<span class="text-body-md">%</span></span>
+<span class="font-mono-data text-mono-data text-surface-tint bg-primary-fixed-dim/20 px-1 rounded">-4.2% Err</span>
+</div>
+<div class="mt-auto text-body-sm text-on-surface-variant">
+                                12,400 eval records passed.
+                            </div>
+</div>
+</div>
+<!-- Logic Diff View -->
+<div class="bg-surface border border-outline-variant/30 rounded-lg flex flex-col overflow-hidden">
+<div class="px-4 py-2 bg-surface-container-low border-b border-outline-variant/30 flex justify-between items-center">
+<span class="font-mono-label text-mono-label text-on-surface">src/prompts/core_intent_v2.txt</span>
+<div class="flex items-center gap-4 text-body-sm text-on-surface-variant">
+<span class="flex items-center gap-1"><span class="w-2 h-2 rounded bg-error/50"></span> Removed</span>
+<span class="flex items-center gap-1"><span class="w-2 h-2 rounded bg-surface-tint/50"></span> Added</span>
+</div>
+</div>
+<!-- Dense Code Diff -->
+<div class="font-mono-data text-mono-data leading-relaxed text-on-background bg-surface-bright overflow-x-auto">
+<div class="flex px-4 py-1 hover:bg-surface-container-lowest text-outline">
+<div class="w-8 select-none text-right pr-4 border-r border-outline-variant/30">42</div>
+<div class="pl-4 whitespace-pre">You are an expert system. Classify the user intent.</div>
+</div>
+<div class="flex px-4 py-1 hover:bg-surface-container-lowest text-outline">
+<div class="w-8 select-none text-right pr-4 border-r border-outline-variant/30">43</div>
+<div class="pl-4 whitespace-pre">Rules:</div>
+</div>
+<!-- Removed Line -->
+<div class="flex px-4 py-1 bg-error-container/20 text-error">
+<div class="w-8 select-none text-right pr-4 border-r border-error/30 opacity-70">44</div>
+<div class="pl-4 whitespace-pre">- If the query is unrelated to Hermes capabilities, reply "OOD".</div>
+</div>
+<!-- Added Lines -->
+<div class="flex px-4 py-1 bg-primary-fixed-dim/10 text-on-primary-fixed">
+<div class="w-8 select-none text-right pr-4 border-r border-primary-fixed-dim/50 opacity-70">44</div>
+<div class="pl-4 whitespace-pre">+ If the query explicitly asks about topics outside the defined skill matrix</div>
+</div>
+<div class="flex px-4 py-1 bg-primary-fixed-dim/10 text-on-primary-fixed">
+<div class="w-8 select-none text-right pr-4 border-r border-primary-fixed-dim/50 opacity-70">45</div>
+<div class="pl-4 whitespace-pre">+ (e.g., weather, generic trivia, medical advice), you MUST classify as "OOD".</div>
+</div>
+<div class="flex px-4 py-1 hover:bg-surface-container-lowest text-outline">
+<div class="w-8 select-none text-right pr-4 border-r border-outline-variant/30">46</div>
+<div class="pl-4 whitespace-pre">- Only extract entities that match the schema perfectly.</div>
+</div>
+</div>
+</div>
+<!-- Execution Context / Details -->
+<div class="bg-surface border border-outline-variant/30 rounded-lg p-5">
+<h3 class="font-mono-label text-mono-label text-on-surface mb-4 border-b border-outline-variant/20 pb-2">EVALUATION_TRACE</h3>
+<div class="grid grid-cols-2 gap-x-8 gap-y-4">
+<div class="flex flex-col gap-1">
+<span class="font-label-caps text-on-surface-variant">TEST DATASET</span>
+<span class="font-body-sm text-on-surface flex items-center gap-2">
+<span class="material-symbols-outlined text-[16px] text-surface-tint">database</span>
+                                    golden_intent_set_v4.json (12k rows)
+                                </span>
+</div>
+<div class="flex flex-col gap-1">
+<span class="font-label-caps text-on-surface-variant">TARGET MODEL</span>
+<span class="font-body-sm text-on-surface font-mono-data">hermes-claude-3-haiku-finetune</span>
+</div>
+<div class="flex flex-col gap-1 col-span-2 mt-2">
+<span class="font-label-caps text-on-surface-variant mb-1">NOTABLE REGRESSIONS</span>
+<div class="bg-surface-container-low border border-outline-variant/20 rounded p-3 text-body-sm text-on-surface-variant">
+                                    None detected. False positive rate for OOD queries dropped from 4.1% to 0.8%. Minor shift in confidence scores for edge-case technical queries (-0.02 delta), well within acceptable bounds.
+                                </div>
+</div>
+</div>
+</div>
+</div>
+</section>
+</main>
+</div>
+</body></html>

package/.claude/worktrees/phase-2-mcp/stitch/curator_workspace/screen.png ADDED Viewed

Binary file