Spaces:
Sleeping
Sleeping
Mahir commited on
Commit ·
c70f857
1
Parent(s): 40eba49
Complete dashboard rebuild — all 13 tasks, 4 domains
Browse files- README.md +60 -107
- static/index.html +551 -1004
README.md
CHANGED
|
@@ -1,6 +1,6 @@
|
|
| 1 |
---
|
| 2 |
-
title:
|
| 3 |
-
emoji:
|
| 4 |
colorFrom: indigo
|
| 5 |
colorTo: purple
|
| 6 |
sdk: docker
|
|
@@ -9,15 +9,15 @@ license: apache-2.0
|
|
| 9 |
tags:
|
| 10 |
- openenv
|
| 11 |
- reinforcement-learning
|
| 12 |
-
-
|
| 13 |
- nlp
|
| 14 |
-
-
|
| 15 |
-
short_description: OpenEnv environment for
|
| 16 |
---
|
| 17 |
|
| 18 |
-
#
|
| 19 |
|
| 20 |
-
> A
|
| 21 |
|
| 22 |
[](https://openenv.dev)
|
| 23 |
[](https://huggingface.co/spaces)
|
|
@@ -27,86 +27,44 @@ short_description: OpenEnv environment for customer support triage agents
|
|
| 27 |
|
| 28 |
## 🎯 Environment Description
|
| 29 |
|
| 30 |
-
|
| 31 |
|
| 32 |
-
|
| 33 |
-
- **Draft** professional, factually grounded customer responses using a knowledge base
|
| 34 |
-
- **Manage** a live queue of tickets across specialized agents under SLA constraints
|
| 35 |
-
|
| 36 |
-
This environment models realistic support scenarios including billing disputes, API outages, SSO failures, account transfers, feature requests, and abuse reports — with varying customer tiers, sentiment signals, and time pressure.
|
| 37 |
|
| 38 |
---
|
| 39 |
|
| 40 |
-
## 📋 Tasks
|
| 41 |
-
|
| 42 |
-
|
| 43 |
-
|
| 44 |
-
|
| 45 |
-
|
|
| 46 |
-
|
|
| 47 |
-
|
| 48 |
-
|
| 49 |
-
|
| 50 |
-
|
| 51 |
-
|
| 52 |
-
|
| 53 |
-
|
| 54 |
-
-
|
| 55 |
-
|
| 56 |
-
|
| 57 |
-
|
| 58 |
-
|
| 59 |
-
###
|
| 60 |
-
|
| 61 |
-
-
|
| 62 |
-
|
| 63 |
-
|
| 64 |
-
|
| 65 |
-
|
| 66 |
-
|
| 67 |
-
|
| 68 |
-
|
| 69 |
-
|
| 70 |
-
|
| 71 |
-
``
|
| 72 |
-
class Observation(BaseModel):
|
| 73 |
-
task_id: str
|
| 74 |
-
step: int
|
| 75 |
-
current_ticket: Optional[Ticket] # For tasks 1 & 2
|
| 76 |
-
ticket_queue: List[Ticket] # For task 3
|
| 77 |
-
agents: List[AgentInfo] # For task 3
|
| 78 |
-
knowledge_base: List[KBArticle] # For task 2
|
| 79 |
-
sla_status: Dict[str, str] # ticket_id -> ok/warning/breached
|
| 80 |
-
valid_actions: List[str]
|
| 81 |
-
episode_done: bool
|
| 82 |
-
info: Dict[str, Any]
|
| 83 |
-
```
|
| 84 |
-
|
| 85 |
-
### Action
|
| 86 |
-
```python
|
| 87 |
-
class Action(BaseModel):
|
| 88 |
-
action_type: AgentAction # classify | draft_response | assign_ticket | escalate | resolve | close | no_op
|
| 89 |
-
ticket_id: Optional[str]
|
| 90 |
-
category: Optional[TicketCategory] # For classify
|
| 91 |
-
priority: Optional[TicketPriority] # For classify
|
| 92 |
-
response_text: Optional[str] # For draft_response
|
| 93 |
-
target_agent_id: Optional[str] # For assign_ticket / escalate
|
| 94 |
-
resolution_summary: Optional[str] # For resolve
|
| 95 |
-
reasoning: Optional[str] # Optional transparency field
|
| 96 |
-
```
|
| 97 |
-
|
| 98 |
-
### Reward
|
| 99 |
-
```python
|
| 100 |
-
class Reward(BaseModel):
|
| 101 |
-
total: float # -1.0 to 1.0
|
| 102 |
-
classification_accuracy: float
|
| 103 |
-
response_quality: float
|
| 104 |
-
sla_compliance: float
|
| 105 |
-
first_contact_resolution: float
|
| 106 |
-
customer_satisfaction: float
|
| 107 |
-
penalty: float # Negative component
|
| 108 |
-
breakdown: Dict[str, float] # Per-dimension scores
|
| 109 |
-
```
|
| 110 |
|
| 111 |
---
|
| 112 |
|
|
@@ -136,12 +94,10 @@ docker run -p 7860:7860 openenv-support
|
|
| 136 |
### Running the Baseline
|
| 137 |
|
| 138 |
```bash
|
| 139 |
-
export OPENAI_API_KEY="sk-..."
|
| 140 |
export OPENENV_BASE_URL="http://localhost:7860"
|
| 141 |
|
| 142 |
-
|
| 143 |
-
python baseline_inference.py
|
| 144 |
-
python baseline_inference.py --model gpt-4o
|
| 145 |
```
|
| 146 |
|
| 147 |
### OpenEnv Validate
|
|
@@ -152,17 +108,6 @@ openenv validate --url http://localhost:7860
|
|
| 152 |
|
| 153 |
---
|
| 154 |
|
| 155 |
-
## 📊 Baseline Scores
|
| 156 |
-
|
| 157 |
-
| Agent | Classification | Drafting | Queue Mgmt | Overall |
|
| 158 |
-
|-------|---------------|---------|------------|---------|
|
| 159 |
-
| Heuristic (keyword-based) | 0.72 | 0.61 | 0.54 | 0.62 |
|
| 160 |
-
| GPT-4o-mini | ~0.85 | ~0.74 | ~0.61 | ~0.73 |
|
| 161 |
-
|
| 162 |
-
> Scores are deterministic given the same random seed. Run `/baseline` endpoint to reproduce.
|
| 163 |
-
|
| 164 |
-
---
|
| 165 |
-
|
| 166 |
## 🌐 API Reference
|
| 167 |
|
| 168 |
| Endpoint | Method | Description |
|
|
@@ -188,16 +133,24 @@ openenv-support/
|
|
| 188 |
├── tasks/
|
| 189 |
│ ├── task1_classification.py
|
| 190 |
│ ├── task2_drafting.py
|
| 191 |
-
│
|
|
|
|
|
|
|
|
|
|
|
|
|
| 192 |
├── graders/
|
| 193 |
-
│ └── baseline_agent.py #
|
| 194 |
├── data/
|
| 195 |
-
│ ├── tickets.py
|
| 196 |
-
│
|
|
|
|
|
|
|
|
|
|
| 197 |
├── static/
|
| 198 |
-
│ └── index.html # Interactive dashboard
|
| 199 |
├── openenv.yaml # OpenEnv metadata
|
| 200 |
-
├── baseline_inference.py #
|
|
|
|
| 201 |
├── requirements.txt
|
| 202 |
├── Dockerfile
|
| 203 |
└── README.md
|
|
@@ -207,8 +160,8 @@ openenv-support/
|
|
| 207 |
|
| 208 |
## 🏗️ Design Decisions
|
| 209 |
|
| 210 |
-
**Why
|
| 211 |
|
| 212 |
-
**Why deterministic graders?** The rubric uses keyword matching, heuristic NLP signals, and exact comparisons —
|
| 213 |
|
| 214 |
-
**Reward
|
|
|
|
| 1 |
---
|
| 2 |
+
title: OpenEnv — Multi-Domain
|
| 3 |
+
emoji: 🌐
|
| 4 |
colorFrom: indigo
|
| 5 |
colorTo: purple
|
| 6 |
sdk: docker
|
|
|
|
| 9 |
tags:
|
| 10 |
- openenv
|
| 11 |
- reinforcement-learning
|
| 12 |
+
- agents
|
| 13 |
- nlp
|
| 14 |
+
- multi-domain
|
| 15 |
+
short_description: OpenEnv multi-domain evaluation environment for AI agents
|
| 16 |
---
|
| 17 |
|
| 18 |
+
# OpenEnv — Multi-Domain AI Agent Environment
|
| 19 |
|
| 20 |
+
> A comprehensive OpenEnv environment where AI agents are evaluated across 13 complex real-world tasks spanning 4 professional domains: **Customer Support, Legal Review, Clinical Triage, and Software Engineering (PR Review)**.
|
| 21 |
|
| 22 |
[](https://openenv.dev)
|
| 23 |
[](https://huggingface.co/spaces)
|
|
|
|
| 27 |
|
| 28 |
## 🎯 Environment Description
|
| 29 |
|
| 30 |
+
To build general-purpose AI agents, we must evaluate them outside of traditional multiple-choice benchmarks. This environment provides 13 interactive, multi-step tasks across diverse verticals to test an agent's reasoning, tool use, and decision-making capabilities.
|
| 31 |
|
| 32 |
+
Agents must classify data, identify risks, draft domain-specific responses, and manage queues under dynamic constraints.
|
|
|
|
|
|
|
|
|
|
|
|
|
| 33 |
|
| 34 |
---
|
| 35 |
|
| 36 |
+
## 📋 Tasks & Domains
|
| 37 |
+
|
| 38 |
+
This environment features 13 tasks across 4 distinct professional domains:
|
| 39 |
+
|
| 40 |
+
### 🎫 Customer Support
|
| 41 |
+
| Task | ID | Difficulty | Max Steps |
|
| 42 |
+
|------|----|-----------|-----------|
|
| 43 |
+
| Ticket Classification | `ticket_classification` | 🟢 Easy | 10 |
|
| 44 |
+
| Response Drafting | `response_drafting` | 🟡 Medium | 6 |
|
| 45 |
+
| SLA Queue Management | `queue_management` | 🔴 Hard | 40 |
|
| 46 |
+
| Multi-Turn De-escalation | `multi_turn_conversation` | 🟣 Very Hard | 8 |
|
| 47 |
+
|
| 48 |
+
### ⚖️ Legal Review
|
| 49 |
+
| Task | ID | Difficulty | Max Steps |
|
| 50 |
+
|------|----|-----------|-----------|
|
| 51 |
+
| Clause Identification | `legal_clause_identification` | 🟢 Easy | 10 |
|
| 52 |
+
| Risk Flagging | `legal_risk_flagging` | 🟡 Medium | 10 |
|
| 53 |
+
| Clause Redlining | `legal_clause_redlining` | 🔴 Hard | 5 |
|
| 54 |
+
|
| 55 |
+
### 🏥 Clinical Triage
|
| 56 |
+
| Task | ID | Difficulty | Max Steps |
|
| 57 |
+
|------|----|-----------|-----------|
|
| 58 |
+
| Body System Classification | `clinical_triage_classification` | 🟢 Easy | 10 |
|
| 59 |
+
| ESI Level Assignment | `clinical_esi_assignment` | 🟡 Medium | 10 |
|
| 60 |
+
| Triage Note Generation | `clinical_triage_note` | 🔴 Hard | 5 |
|
| 61 |
+
|
| 62 |
+
### 💻 PR Review (Software Engineering)
|
| 63 |
+
| Task | ID | Difficulty | Max Steps |
|
| 64 |
+
|------|----|-----------|-----------|
|
| 65 |
+
| PR Type Classification | `pr_type_classification` | 🟢 Easy | 10 |
|
| 66 |
+
| Bug Identification | `pr_bug_identification` | 🟡 Medium | 5 |
|
| 67 |
+
| Code Review Comment | `pr_review_comment` | 🔴 Hard | 5 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 68 |
|
| 69 |
---
|
| 70 |
|
|
|
|
| 94 |
### Running the Baseline
|
| 95 |
|
| 96 |
```bash
|
|
|
|
| 97 |
export OPENENV_BASE_URL="http://localhost:7860"
|
| 98 |
|
| 99 |
+
# Evaluate heuristic baseline on all tasks
|
| 100 |
+
python baseline_inference.py
|
|
|
|
| 101 |
```
|
| 102 |
|
| 103 |
### OpenEnv Validate
|
|
|
|
| 108 |
|
| 109 |
---
|
| 110 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 111 |
## 🌐 API Reference
|
| 112 |
|
| 113 |
| Endpoint | Method | Description |
|
|
|
|
| 133 |
├── tasks/
|
| 134 |
│ ├── task1_classification.py
|
| 135 |
│ ├── task2_drafting.py
|
| 136 |
+
│ ├── task3_queue.py
|
| 137 |
+
│ ├── task4_multiturn.py
|
| 138 |
+
│ ├── legal_tasks.py
|
| 139 |
+
│ ├── clinical_tasks.py
|
| 140 |
+
│ └── pr_tasks.py
|
| 141 |
├── graders/
|
| 142 |
+
│ └── baseline_agent.py # Auto-grader heuristics
|
| 143 |
├── data/
|
| 144 |
+
│ ├── tickets.py
|
| 145 |
+
│ ├── knowledge_base.py
|
| 146 |
+
│ ├── legal_data.py
|
| 147 |
+
│ ├── clinical_data.py
|
| 148 |
+
│ └── pr_data.py
|
| 149 |
├── static/
|
| 150 |
+
│ └── index.html # Interactive multi-domain dashboard
|
| 151 |
├── openenv.yaml # OpenEnv metadata
|
| 152 |
+
├── baseline_inference.py # Automated test runner for baseline
|
| 153 |
+
├── validate.py # Validation script
|
| 154 |
├── requirements.txt
|
| 155 |
├── Dockerfile
|
| 156 |
└── README.md
|
|
|
|
| 160 |
|
| 161 |
## 🏗️ Design Decisions
|
| 162 |
|
| 163 |
+
**Why multi-domain?** Real-world agents will need to operate across various specific verticals. Providing environments in Support, Legal, Clinical, and Engineering proves that an agent architecture is truly generalizable and robust.
|
| 164 |
|
| 165 |
+
**Why deterministic graders?** The rubric uses keyword matching, heuristic NLP signals (via \`sentence-transformers\`), semantic similarities, and exact comparisons — minimizing LLM-in-the-loop dependencies for grading and ensuring reproducible scores across runs for all 13 tasks.
|
| 166 |
|
| 167 |
+
**Dynamic Reward Shaping:** Every step returns a non-zero reward signal. Even incorrect actions return small penalties rather than zero, giving RL agents a gradient to follow rather than a sparse reward landscape.
|
static/index.html
CHANGED
|
@@ -3,780 +3,294 @@
|
|
| 3 |
<head>
|
| 4 |
<meta charset="UTF-8"/>
|
| 5 |
<meta name="viewport" content="width=device-width, initial-scale=1.0"/>
|
| 6 |
-
<title>
|
| 7 |
<link rel="preconnect" href="https://fonts.googleapis.com">
|
| 8 |
<link href="https://fonts.googleapis.com/css2?family=Space+Mono:wght@400;700&family=Syne:wght@400;600;700;800&display=swap" rel="stylesheet">
|
| 9 |
<style>
|
| 10 |
:root {
|
| 11 |
-
--bg: #
|
| 12 |
-
--surface: #
|
| 13 |
-
--surface2: #
|
| 14 |
-
--border: #
|
| 15 |
--accent: #6c63ff;
|
| 16 |
--accent2: #ff6584;
|
| 17 |
--accent3: #43e97b;
|
| 18 |
--accent4: #f7971e;
|
|
|
|
| 19 |
--text: #e8e8f0;
|
| 20 |
-
--text-dim: #
|
| 21 |
--mono: 'Space Mono', monospace;
|
| 22 |
--sans: 'Syne', sans-serif;
|
| 23 |
-
--radius:
|
| 24 |
-
--glow: 0 0 24px rgba(108,99,255,0.15);
|
| 25 |
}
|
| 26 |
-
|
| 27 |
* { box-sizing: border-box; margin: 0; padding: 0; }
|
| 28 |
-
|
| 29 |
-
body {
|
| 30 |
-
background: var(--bg);
|
| 31 |
-
color: var(--text);
|
| 32 |
-
font-family: var(--sans);
|
| 33 |
-
min-height: 100vh;
|
| 34 |
-
overflow-x: hidden;
|
| 35 |
-
}
|
| 36 |
-
|
| 37 |
-
/* Background grid */
|
| 38 |
body::before {
|
| 39 |
-
content: '';
|
| 40 |
-
|
| 41 |
-
|
| 42 |
-
background-image:
|
| 43 |
-
linear-gradient(rgba(108,99,255,0.04) 1px, transparent 1px),
|
| 44 |
-
linear-gradient(90deg, rgba(108,99,255,0.04) 1px, transparent 1px);
|
| 45 |
-
background-size: 40px 40px;
|
| 46 |
-
pointer-events: none;
|
| 47 |
-
z-index: 0;
|
| 48 |
}
|
| 49 |
|
| 50 |
-
|
| 51 |
-
|
| 52 |
-
|
| 53 |
-
display: grid;
|
| 54 |
-
grid-template-rows: auto 1fr;
|
| 55 |
-
min-height: 100vh;
|
| 56 |
-
}
|
| 57 |
|
| 58 |
/* ── Header ── */
|
| 59 |
header {
|
| 60 |
-
display: flex;
|
| 61 |
-
|
| 62 |
-
|
| 63 |
-
|
| 64 |
-
border-bottom: 1px solid var(--border);
|
| 65 |
-
background: rgba(10,10,15,0.8);
|
| 66 |
-
backdrop-filter: blur(12px);
|
| 67 |
-
position: sticky;
|
| 68 |
-
top: 0;
|
| 69 |
-
z-index: 100;
|
| 70 |
}
|
| 71 |
-
|
| 72 |
-
.logo {
|
| 73 |
-
display: flex;
|
| 74 |
-
align-items: center;
|
| 75 |
-
gap: 12px;
|
| 76 |
-
}
|
| 77 |
-
|
| 78 |
.logo-icon {
|
| 79 |
-
width: 36px;
|
| 80 |
-
height: 36px;
|
| 81 |
background: linear-gradient(135deg, var(--accent), var(--accent2));
|
| 82 |
-
|
| 83 |
-
display: flex;
|
| 84 |
-
align-items: center;
|
| 85 |
-
justify-content: center;
|
| 86 |
-
font-size: 18px;
|
| 87 |
}
|
| 88 |
-
|
| 89 |
-
.logo-
|
| 90 |
-
|
| 91 |
-
|
| 92 |
-
|
|
|
|
|
|
|
|
|
|
| 93 |
}
|
| 94 |
-
|
| 95 |
-
|
| 96 |
-
|
| 97 |
-
font-size:
|
| 98 |
-
color: var(--
|
| 99 |
-
|
| 100 |
-
margin-top: 1px;
|
| 101 |
-
}
|
| 102 |
-
|
| 103 |
-
.header-status {
|
| 104 |
-
display: flex;
|
| 105 |
-
align-items: center;
|
| 106 |
-
gap: 8px;
|
| 107 |
-
font-family: var(--mono);
|
| 108 |
-
font-size: 12px;
|
| 109 |
-
color: var(--accent3);
|
| 110 |
-
}
|
| 111 |
-
|
| 112 |
-
.pulse {
|
| 113 |
-
width: 8px;
|
| 114 |
-
height: 8px;
|
| 115 |
-
background: var(--accent3);
|
| 116 |
-
border-radius: 50%;
|
| 117 |
-
animation: pulse 2s infinite;
|
| 118 |
-
}
|
| 119 |
-
|
| 120 |
-
@keyframes pulse {
|
| 121 |
-
0%, 100% { opacity: 1; transform: scale(1); }
|
| 122 |
-
50% { opacity: 0.5; transform: scale(0.8); }
|
| 123 |
-
}
|
| 124 |
-
|
| 125 |
-
/* ── Main content ── */
|
| 126 |
-
.main {
|
| 127 |
-
display: grid;
|
| 128 |
-
grid-template-columns: 280px 1fr 320px;
|
| 129 |
-
gap: 0;
|
| 130 |
-
height: calc(100vh - 77px);
|
| 131 |
-
overflow: hidden;
|
| 132 |
}
|
| 133 |
|
| 134 |
/* ── Sidebar ── */
|
| 135 |
-
.sidebar {
|
| 136 |
-
|
| 137 |
-
|
| 138 |
-
|
| 139 |
-
display: flex;
|
| 140 |
-
flex-direction: column;
|
| 141 |
-
gap: 24px;
|
| 142 |
-
}
|
| 143 |
-
|
| 144 |
-
.sidebar-section-label {
|
| 145 |
-
font-family: var(--mono);
|
| 146 |
-
font-size: 10px;
|
| 147 |
-
color: var(--text-dim);
|
| 148 |
-
letter-spacing: 2px;
|
| 149 |
-
text-transform: uppercase;
|
| 150 |
-
margin-bottom: 12px;
|
| 151 |
}
|
| 152 |
-
|
| 153 |
.task-card {
|
| 154 |
-
background: var(--surface);
|
| 155 |
-
border:
|
| 156 |
-
|
| 157 |
-
padding: 16px;
|
| 158 |
-
cursor: pointer;
|
| 159 |
-
transition: all 0.2s;
|
| 160 |
-
position: relative;
|
| 161 |
-
overflow: hidden;
|
| 162 |
-
}
|
| 163 |
-
|
| 164 |
-
.task-card::before {
|
| 165 |
-
content: '';
|
| 166 |
-
position: absolute;
|
| 167 |
-
left: 0;
|
| 168 |
-
top: 0;
|
| 169 |
-
bottom: 0;
|
| 170 |
-
width: 3px;
|
| 171 |
-
background: var(--accent);
|
| 172 |
-
opacity: 0;
|
| 173 |
-
transition: opacity 0.2s;
|
| 174 |
-
}
|
| 175 |
-
|
| 176 |
-
.task-card:hover, .task-card.active {
|
| 177 |
-
border-color: var(--accent);
|
| 178 |
-
background: var(--surface2);
|
| 179 |
-
box-shadow: var(--glow);
|
| 180 |
}
|
| 181 |
-
|
|
|
|
|
|
|
| 182 |
.task-card.active::before { opacity: 1; }
|
| 183 |
-
|
| 184 |
-
.task-
|
| 185 |
-
|
| 186 |
-
|
| 187 |
-
|
| 188 |
-
}
|
| 189 |
-
|
| 190 |
-
.task-
|
| 191 |
-
|
| 192 |
-
|
| 193 |
-
|
| 194 |
-
|
| 195 |
-
}
|
| 196 |
-
|
| 197 |
-
|
| 198 |
-
|
| 199 |
-
|
| 200 |
-
padding: 2px 8px;
|
| 201 |
-
border-radius: 4px;
|
| 202 |
-
font-weight: 700;
|
| 203 |
-
letter-spacing: 0.5px;
|
| 204 |
-
}
|
| 205 |
-
|
| 206 |
-
.badge-easy { background: rgba(67,233,123,0.15); color: var(--accent3); }
|
| 207 |
-
.badge-medium { background: rgba(247,151,30,0.15); color: var(--accent4); }
|
| 208 |
-
.badge-hard { background: rgba(255,101,132,0.15); color: var(--accent2); }
|
| 209 |
-
|
| 210 |
-
.score-ring {
|
| 211 |
-
font-family: var(--mono);
|
| 212 |
-
font-size: 11px;
|
| 213 |
-
color: var(--text-dim);
|
| 214 |
-
}
|
| 215 |
-
|
| 216 |
-
/* ── Center panel ── */
|
| 217 |
-
.center {
|
| 218 |
-
display: flex;
|
| 219 |
-
flex-direction: column;
|
| 220 |
-
overflow: hidden;
|
| 221 |
-
}
|
| 222 |
-
|
| 223 |
.center-header {
|
| 224 |
-
padding:
|
| 225 |
-
|
| 226 |
-
|
| 227 |
-
|
| 228 |
-
|
| 229 |
-
|
| 230 |
-
}
|
| 231 |
-
|
| 232 |
-
.center-title {
|
| 233 |
-
font-size: 20px;
|
| 234 |
-
font-weight: 800;
|
| 235 |
-
}
|
| 236 |
-
|
| 237 |
-
.action-bar {
|
| 238 |
-
display: flex;
|
| 239 |
-
gap: 10px;
|
| 240 |
-
}
|
| 241 |
-
|
| 242 |
-
.btn {
|
| 243 |
-
font-family: var(--mono);
|
| 244 |
-
font-size: 12px;
|
| 245 |
-
padding: 8px 18px;
|
| 246 |
-
border-radius: 8px;
|
| 247 |
-
border: none;
|
| 248 |
-
cursor: pointer;
|
| 249 |
-
transition: all 0.15s;
|
| 250 |
-
font-weight: 700;
|
| 251 |
-
letter-spacing: 0.5px;
|
| 252 |
-
}
|
| 253 |
-
|
| 254 |
-
.btn-primary {
|
| 255 |
-
background: var(--accent);
|
| 256 |
-
color: #fff;
|
| 257 |
-
}
|
| 258 |
-
|
| 259 |
.btn-primary:hover { background: #7c74ff; transform: translateY(-1px); }
|
| 260 |
-
|
| 261 |
-
.btn-secondary {
|
| 262 |
-
background: var(--surface2);
|
| 263 |
-
color: var(--text);
|
| 264 |
-
border: 1px solid var(--border);
|
| 265 |
-
}
|
| 266 |
-
|
| 267 |
.btn-secondary:hover { border-color: var(--accent); }
|
| 268 |
-
|
| 269 |
-
|
| 270 |
-
|
| 271 |
-
|
| 272 |
-
|
| 273 |
-
}
|
| 274 |
-
|
| 275 |
-
.
|
| 276 |
-
|
| 277 |
-
|
| 278 |
-
|
| 279 |
-
}
|
| 280 |
-
|
| 281 |
-
|
| 282 |
-
|
| 283 |
-
|
| 284 |
-
|
| 285 |
-
}
|
| 286 |
-
|
| 287 |
-
|
| 288 |
-
.
|
| 289 |
-
|
| 290 |
-
border: 1px solid var(--border);
|
| 291 |
-
border-radius: var(--radius);
|
| 292 |
-
padding: 24px;
|
| 293 |
-
margin-bottom: 16px;
|
| 294 |
-
animation: slideIn 0.3s ease;
|
| 295 |
-
}
|
| 296 |
-
|
| 297 |
-
@keyframes slideIn {
|
| 298 |
-
from { opacity: 0; transform: translateY(12px); }
|
| 299 |
-
to { opacity: 1; transform: translateY(0); }
|
| 300 |
-
}
|
| 301 |
-
|
| 302 |
-
.ticket-header {
|
| 303 |
-
display: flex;
|
| 304 |
-
align-items: flex-start;
|
| 305 |
-
justify-content: space-between;
|
| 306 |
-
margin-bottom: 16px;
|
| 307 |
-
gap: 16px;
|
| 308 |
-
}
|
| 309 |
-
|
| 310 |
-
.ticket-id {
|
| 311 |
-
font-family: var(--mono);
|
| 312 |
-
font-size: 11px;
|
| 313 |
-
color: var(--accent);
|
| 314 |
-
margin-bottom: 6px;
|
| 315 |
-
}
|
| 316 |
-
|
| 317 |
-
.ticket-subject {
|
| 318 |
-
font-size: 18px;
|
| 319 |
-
font-weight: 700;
|
| 320 |
-
line-height: 1.3;
|
| 321 |
-
}
|
| 322 |
-
|
| 323 |
-
.ticket-tier {
|
| 324 |
-
flex-shrink: 0;
|
| 325 |
-
}
|
| 326 |
-
|
| 327 |
-
.tier-badge {
|
| 328 |
-
font-family: var(--mono);
|
| 329 |
-
font-size: 10px;
|
| 330 |
-
padding: 4px 10px;
|
| 331 |
-
border-radius: 6px;
|
| 332 |
-
font-weight: 700;
|
| 333 |
-
}
|
| 334 |
-
|
| 335 |
-
.tier-enterprise { background: rgba(108,99,255,0.2); color: var(--accent); border: 1px solid rgba(108,99,255,0.3); }
|
| 336 |
-
.tier-pro { background: rgba(247,151,30,0.15); color: var(--accent4); border: 1px solid rgba(247,151,30,0.3); }
|
| 337 |
-
.tier-free { background: rgba(120,120,160,0.15); color: var(--text-dim); border: 1px solid var(--border); }
|
| 338 |
-
|
| 339 |
-
.ticket-body {
|
| 340 |
-
font-size: 14px;
|
| 341 |
-
line-height: 1.7;
|
| 342 |
-
color: #c0c0d8;
|
| 343 |
-
padding: 16px;
|
| 344 |
-
background: var(--surface2);
|
| 345 |
-
border-radius: 8px;
|
| 346 |
-
border-left: 3px solid var(--border);
|
| 347 |
-
margin-bottom: 16px;
|
| 348 |
-
}
|
| 349 |
-
|
| 350 |
-
.ticket-meta-row {
|
| 351 |
-
display: flex;
|
| 352 |
-
gap: 12px;
|
| 353 |
-
flex-wrap: wrap;
|
| 354 |
-
}
|
| 355 |
-
|
| 356 |
-
.meta-chip {
|
| 357 |
-
display: flex;
|
| 358 |
-
align-items: center;
|
| 359 |
-
gap: 6px;
|
| 360 |
-
font-family: var(--mono);
|
| 361 |
-
font-size: 11px;
|
| 362 |
-
color: var(--text-dim);
|
| 363 |
-
background: var(--surface2);
|
| 364 |
-
padding: 4px 10px;
|
| 365 |
-
border-radius: 6px;
|
| 366 |
-
border: 1px solid var(--border);
|
| 367 |
-
}
|
| 368 |
-
|
| 369 |
-
.sentiment-bar {
|
| 370 |
-
display: flex;
|
| 371 |
-
align-items: center;
|
| 372 |
-
gap: 8px;
|
| 373 |
-
margin-top: 12px;
|
| 374 |
-
}
|
| 375 |
-
|
| 376 |
-
.sentiment-label {
|
| 377 |
-
font-family: var(--mono);
|
| 378 |
-
font-size: 10px;
|
| 379 |
-
color: var(--text-dim);
|
| 380 |
-
width: 80px;
|
| 381 |
-
flex-shrink: 0;
|
| 382 |
-
}
|
| 383 |
-
|
| 384 |
-
.sentiment-track {
|
| 385 |
-
flex: 1;
|
| 386 |
-
height: 4px;
|
| 387 |
-
background: var(--surface2);
|
| 388 |
-
border-radius: 2px;
|
| 389 |
-
overflow: hidden;
|
| 390 |
-
}
|
| 391 |
-
|
| 392 |
-
.sentiment-fill {
|
| 393 |
-
height: 100%;
|
| 394 |
-
border-radius: 2px;
|
| 395 |
-
transition: width 0.5s ease;
|
| 396 |
-
}
|
| 397 |
-
|
| 398 |
-
/* ── KB articles ── */
|
| 399 |
-
.kb-section { margin-top: 20px; }
|
| 400 |
-
|
| 401 |
-
.kb-label {
|
| 402 |
-
font-family: var(--mono);
|
| 403 |
-
font-size: 10px;
|
| 404 |
-
color: var(--text-dim);
|
| 405 |
-
letter-spacing: 2px;
|
| 406 |
-
text-transform: uppercase;
|
| 407 |
-
margin-bottom: 10px;
|
| 408 |
-
}
|
| 409 |
-
|
| 410 |
-
.kb-article {
|
| 411 |
-
background: var(--surface);
|
| 412 |
-
border: 1px solid var(--border);
|
| 413 |
-
border-radius: 8px;
|
| 414 |
-
padding: 14px;
|
| 415 |
-
margin-bottom: 8px;
|
| 416 |
-
}
|
| 417 |
-
|
| 418 |
-
.kb-article-title {
|
| 419 |
-
font-size: 13px;
|
| 420 |
-
font-weight: 700;
|
| 421 |
-
margin-bottom: 6px;
|
| 422 |
-
display: flex;
|
| 423 |
-
align-items: center;
|
| 424 |
-
gap: 8px;
|
| 425 |
-
}
|
| 426 |
-
|
| 427 |
-
.kb-article-id {
|
| 428 |
-
font-family: var(--mono);
|
| 429 |
-
font-size: 10px;
|
| 430 |
-
color: var(--accent);
|
| 431 |
-
}
|
| 432 |
-
|
| 433 |
-
.kb-article-content {
|
| 434 |
-
font-size: 12px;
|
| 435 |
-
color: var(--text-dim);
|
| 436 |
-
line-height: 1.6;
|
| 437 |
-
}
|
| 438 |
|
| 439 |
/* ── Action form ── */
|
| 440 |
-
.action-form {
|
| 441 |
-
|
| 442 |
-
|
| 443 |
-
|
| 444 |
-
|
| 445 |
-
margin-top: 16px;
|
| 446 |
-
}
|
| 447 |
-
|
| 448 |
-
.form-title {
|
| 449 |
-
font-weight: 700;
|
| 450 |
-
font-size: 15px;
|
| 451 |
-
margin-bottom: 16px;
|
| 452 |
-
display: flex;
|
| 453 |
-
align-items: center;
|
| 454 |
-
gap: 8px;
|
| 455 |
-
}
|
| 456 |
-
|
| 457 |
-
.form-row {
|
| 458 |
-
display: grid;
|
| 459 |
-
grid-template-columns: 1fr 1fr;
|
| 460 |
-
gap: 12px;
|
| 461 |
-
margin-bottom: 12px;
|
| 462 |
-
}
|
| 463 |
-
|
| 464 |
-
.form-group {
|
| 465 |
-
display: flex;
|
| 466 |
-
flex-direction: column;
|
| 467 |
-
gap: 6px;
|
| 468 |
-
}
|
| 469 |
-
|
| 470 |
-
.form-label {
|
| 471 |
-
font-family: var(--mono);
|
| 472 |
-
font-size: 10px;
|
| 473 |
-
color: var(--text-dim);
|
| 474 |
-
letter-spacing: 1px;
|
| 475 |
-
text-transform: uppercase;
|
| 476 |
-
}
|
| 477 |
-
|
| 478 |
.form-select, .form-textarea, .form-input {
|
| 479 |
-
background: var(--surface2);
|
| 480 |
-
|
| 481 |
-
border-
|
| 482 |
-
color: var(--text);
|
| 483 |
-
font-family: var(--mono);
|
| 484 |
-
font-size: 13px;
|
| 485 |
-
padding: 10px 12px;
|
| 486 |
-
transition: border-color 0.2s;
|
| 487 |
-
width: 100%;
|
| 488 |
-
}
|
| 489 |
-
|
| 490 |
-
.form-select:focus, .form-textarea:focus, .form-input:focus {
|
| 491 |
-
outline: none;
|
| 492 |
-
border-color: var(--accent);
|
| 493 |
-
}
|
| 494 |
-
|
| 495 |
-
.form-textarea {
|
| 496 |
-
font-family: var(--sans);
|
| 497 |
-
font-size: 14px;
|
| 498 |
-
resize: vertical;
|
| 499 |
-
min-height: 120px;
|
| 500 |
-
line-height: 1.6;
|
| 501 |
-
}
|
| 502 |
-
|
| 503 |
-
.submit-row {
|
| 504 |
-
display: flex;
|
| 505 |
-
justify-content: flex-end;
|
| 506 |
-
margin-top: 16px;
|
| 507 |
}
|
|
|
|
|
|
|
|
|
|
| 508 |
|
| 509 |
/* ── Right panel ── */
|
| 510 |
-
.right-panel {
|
| 511 |
-
|
| 512 |
-
|
| 513 |
-
|
| 514 |
-
|
| 515 |
-
|
| 516 |
-
gap: 20px;
|
| 517 |
-
}
|
| 518 |
-
|
| 519 |
-
/* ── Score display ── */
|
| 520 |
-
.score-card {
|
| 521 |
-
background: var(--surface);
|
| 522 |
-
border: 1px solid var(--border);
|
| 523 |
-
border-radius: var(--radius);
|
| 524 |
-
padding: 20px;
|
| 525 |
-
}
|
| 526 |
-
|
| 527 |
-
.score-card-title {
|
| 528 |
-
font-family: var(--mono);
|
| 529 |
-
font-size: 10px;
|
| 530 |
-
color: var(--text-dim);
|
| 531 |
-
letter-spacing: 2px;
|
| 532 |
-
text-transform: uppercase;
|
| 533 |
-
margin-bottom: 16px;
|
| 534 |
-
}
|
| 535 |
-
|
| 536 |
-
.big-score {
|
| 537 |
-
font-family: var(--mono);
|
| 538 |
-
font-size: 48px;
|
| 539 |
-
font-weight: 700;
|
| 540 |
-
line-height: 1;
|
| 541 |
-
margin-bottom: 4px;
|
| 542 |
-
}
|
| 543 |
-
|
| 544 |
-
.score-label {
|
| 545 |
-
font-family: var(--mono);
|
| 546 |
-
font-size: 11px;
|
| 547 |
-
color: var(--text-dim);
|
| 548 |
-
}
|
| 549 |
-
|
| 550 |
-
.breakdown-row {
|
| 551 |
-
display: flex;
|
| 552 |
-
justify-content: space-between;
|
| 553 |
-
align-items: center;
|
| 554 |
-
padding: 8px 0;
|
| 555 |
-
border-bottom: 1px solid rgba(42,42,64,0.5);
|
| 556 |
-
}
|
| 557 |
-
|
| 558 |
.breakdown-row:last-child { border-bottom: none; }
|
| 559 |
-
|
| 560 |
-
.breakdown-
|
| 561 |
-
|
| 562 |
-
|
| 563 |
-
|
| 564 |
-
}
|
| 565 |
-
|
| 566 |
-
|
| 567 |
-
|
| 568 |
-
|
| 569 |
-
|
| 570 |
-
}
|
| 571 |
-
|
| 572 |
-
.val-good { color: var(--accent3); }
|
| 573 |
-
.val-mid { color: var(--accent4); }
|
| 574 |
-
.val-bad { color: var(--accent2); }
|
| 575 |
-
|
| 576 |
-
/* ── Reward history chart ── */
|
| 577 |
-
.mini-chart {
|
| 578 |
-
height: 60px;
|
| 579 |
-
display: flex;
|
| 580 |
-
align-items: flex-end;
|
| 581 |
-
gap: 3px;
|
| 582 |
-
padding-top: 8px;
|
| 583 |
-
}
|
| 584 |
-
|
| 585 |
-
.bar {
|
| 586 |
-
flex: 1;
|
| 587 |
-
border-radius: 3px 3px 0 0;
|
| 588 |
-
min-height: 4px;
|
| 589 |
-
transition: height 0.3s ease;
|
| 590 |
-
}
|
| 591 |
-
|
| 592 |
-
/* ── Queue list ── */
|
| 593 |
-
.queue-item {
|
| 594 |
-
display: flex;
|
| 595 |
-
align-items: center;
|
| 596 |
-
gap: 10px;
|
| 597 |
-
padding: 10px;
|
| 598 |
-
background: var(--surface);
|
| 599 |
-
border: 1px solid var(--border);
|
| 600 |
-
border-radius: 8px;
|
| 601 |
-
margin-bottom: 6px;
|
| 602 |
-
font-size: 12px;
|
| 603 |
-
transition: border-color 0.2s;
|
| 604 |
-
}
|
| 605 |
-
|
| 606 |
-
.queue-item:hover { border-color: var(--accent); }
|
| 607 |
-
|
| 608 |
-
.priority-dot {
|
| 609 |
-
width: 8px;
|
| 610 |
-
height: 8px;
|
| 611 |
-
border-radius: 50%;
|
| 612 |
-
flex-shrink: 0;
|
| 613 |
-
}
|
| 614 |
-
|
| 615 |
-
.p1 { background: var(--accent2); box-shadow: 0 0 6px var(--accent2); }
|
| 616 |
-
.p2 { background: var(--accent4); }
|
| 617 |
-
.p3 { background: var(--accent3); }
|
| 618 |
-
.p4 { background: var(--text-dim); }
|
| 619 |
-
|
| 620 |
-
.queue-subject {
|
| 621 |
-
flex: 1;
|
| 622 |
-
overflow: hidden;
|
| 623 |
-
text-overflow: ellipsis;
|
| 624 |
-
white-space: nowrap;
|
| 625 |
-
color: var(--text);
|
| 626 |
-
}
|
| 627 |
-
|
| 628 |
-
.queue-sla {
|
| 629 |
-
font-family: var(--mono);
|
| 630 |
-
font-size: 10px;
|
| 631 |
-
color: var(--text-dim);
|
| 632 |
-
flex-shrink: 0;
|
| 633 |
-
}
|
| 634 |
-
|
| 635 |
-
.sla-breached { color: var(--accent2); }
|
| 636 |
-
.sla-warning { color: var(--accent4); }
|
| 637 |
-
.sla-ok { color: var(--accent3); }
|
| 638 |
-
|
| 639 |
-
/* ── Agent cards ── */
|
| 640 |
-
.agent-card {
|
| 641 |
-
background: var(--surface);
|
| 642 |
-
border: 1px solid var(--border);
|
| 643 |
-
border-radius: 8px;
|
| 644 |
-
padding: 12px;
|
| 645 |
-
margin-bottom: 8px;
|
| 646 |
-
}
|
| 647 |
-
|
| 648 |
-
.agent-name {
|
| 649 |
-
font-size: 13px;
|
| 650 |
-
font-weight: 700;
|
| 651 |
-
margin-bottom: 8px;
|
| 652 |
-
}
|
| 653 |
-
|
| 654 |
-
.load-bar-track {
|
| 655 |
-
height: 6px;
|
| 656 |
-
background: var(--surface2);
|
| 657 |
-
border-radius: 3px;
|
| 658 |
-
overflow: hidden;
|
| 659 |
-
margin-top: 6px;
|
| 660 |
-
}
|
| 661 |
-
|
| 662 |
-
.load-bar-fill {
|
| 663 |
-
height: 100%;
|
| 664 |
-
border-radius: 3px;
|
| 665 |
-
background: linear-gradient(90deg, var(--accent3), var(--accent));
|
| 666 |
-
transition: width 0.4s ease;
|
| 667 |
-
}
|
| 668 |
-
|
| 669 |
-
.load-bar-fill.overloaded { background: linear-gradient(90deg, var(--accent4), var(--accent2)); }
|
| 670 |
-
|
| 671 |
-
/* ── Log ── */
|
| 672 |
-
.log-entry {
|
| 673 |
-
font-family: var(--mono);
|
| 674 |
-
font-size: 11px;
|
| 675 |
-
color: var(--text-dim);
|
| 676 |
-
padding: 6px 10px;
|
| 677 |
-
border-radius: 4px;
|
| 678 |
-
margin-bottom: 4px;
|
| 679 |
-
background: var(--surface);
|
| 680 |
-
}
|
| 681 |
-
|
| 682 |
-
.log-reward-pos { color: var(--accent3); }
|
| 683 |
-
.log-reward-neg { color: var(--accent2); }
|
| 684 |
-
|
| 685 |
-
/* ── Empty state ── */
|
| 686 |
-
.empty-state {
|
| 687 |
-
display: flex;
|
| 688 |
-
flex-direction: column;
|
| 689 |
-
align-items: center;
|
| 690 |
-
justify-content: center;
|
| 691 |
-
height: 300px;
|
| 692 |
-
color: var(--text-dim);
|
| 693 |
-
text-align: center;
|
| 694 |
-
gap: 12px;
|
| 695 |
-
}
|
| 696 |
-
|
| 697 |
-
.empty-icon { font-size: 48px; opacity: 0.3; }
|
| 698 |
-
.empty-text { font-size: 14px; }
|
| 699 |
|
| 700 |
/* ── Episode done ── */
|
| 701 |
-
.episode-done {
|
| 702 |
-
|
| 703 |
-
|
| 704 |
-
|
| 705 |
-
|
| 706 |
-
text-align: center;
|
| 707 |
-
animation: slideIn 0.4s ease;
|
| 708 |
-
}
|
| 709 |
-
|
| 710 |
-
.done-icon { font-size: 48px; margin-bottom: 16px; }
|
| 711 |
-
.done-title { font-size: 24px; font-weight: 800; margin-bottom: 8px; }
|
| 712 |
-
.done-score { font-family: var(--mono); font-size: 64px; font-weight: 700; color: var(--accent); }
|
| 713 |
|
| 714 |
-
/*
|
| 715 |
-
::-
|
| 716 |
-
|
| 717 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 718 |
</style>
|
| 719 |
</head>
|
| 720 |
<body>
|
| 721 |
<div class="layout">
|
| 722 |
<header>
|
| 723 |
<div class="logo">
|
| 724 |
-
<div class="logo-icon">
|
| 725 |
<div>
|
| 726 |
-
<div class="logo-
|
| 727 |
-
<div class="logo-sub">
|
| 728 |
</div>
|
| 729 |
</div>
|
| 730 |
-
<div class="header-
|
| 731 |
-
<div class="
|
| 732 |
-
<span id="server-status">
|
| 733 |
</div>
|
| 734 |
</header>
|
| 735 |
|
| 736 |
<div class="main">
|
| 737 |
<!-- Sidebar -->
|
| 738 |
<aside class="sidebar">
|
| 739 |
-
|
| 740 |
-
|
| 741 |
-
<div class="
|
|
|
|
| 742 |
<div class="task-name">Ticket Classification</div>
|
| 743 |
-
<div
|
| 744 |
-
<div class="task-meta">
|
| 745 |
-
<span class="badge badge-easy">EASY</span>
|
| 746 |
-
<span class="score-ring" id="score-classification">—</span>
|
| 747 |
-
</div>
|
| 748 |
</div>
|
| 749 |
-
<div class="task-card" id="
|
| 750 |
<div class="task-name">Response Drafting</div>
|
| 751 |
-
<div
|
| 752 |
-
<div class="task-meta">
|
| 753 |
-
<span class="badge badge-medium">MEDIUM</span>
|
| 754 |
-
<span class="score-ring" id="score-drafting">—</span>
|
| 755 |
-
</div>
|
| 756 |
</div>
|
| 757 |
-
<div class="task-card" id="
|
| 758 |
-
<div class="task-name">Queue Management</div>
|
| 759 |
-
<div
|
| 760 |
-
|
| 761 |
-
|
| 762 |
-
|
| 763 |
-
</div>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 764 |
</div>
|
| 765 |
</div>
|
| 766 |
|
| 767 |
<div>
|
| 768 |
-
<div class="
|
| 769 |
-
<div id="log-container" style="max-height:
|
| 770 |
</div>
|
| 771 |
|
| 772 |
<div>
|
| 773 |
-
<div class="
|
| 774 |
-
<a href="/docs" target="_blank" style="display:flex;align-items:center;gap:8px;color:var(--accent);font-size:
|
| 775 |
-
|
| 776 |
-
</a>
|
| 777 |
-
<a href="/tasks" target="_blank" style="display:flex;align-items:center;gap:8px;color:var(--text-dim);font-size:13px;text-decoration:none;padding:8px;background:var(--surface);border-radius:8px;border:1px solid var(--border);margin-top:6px;">
|
| 778 |
-
<span>📋</span> /tasks endpoint
|
| 779 |
-
</a>
|
| 780 |
</div>
|
| 781 |
</aside>
|
| 782 |
|
|
@@ -785,20 +299,18 @@
|
|
| 785 |
<div class="center-header">
|
| 786 |
<div>
|
| 787 |
<div class="center-title" id="center-title">Ticket Classification</div>
|
| 788 |
-
<div
|
| 789 |
-
Step <span id="step-counter">0</span> / <span id="max-steps">10</span>
|
| 790 |
-
</div>
|
| 791 |
</div>
|
| 792 |
<div class="action-bar">
|
| 793 |
-
<button class="btn btn-secondary" onclick="runBaseline()">▶ Heuristic Baseline</button>
|
| 794 |
<button class="btn btn-primary" onclick="resetEpisode()">↺ Reset Episode</button>
|
| 795 |
</div>
|
| 796 |
</div>
|
| 797 |
-
|
| 798 |
<div class="center-body" id="center-body">
|
| 799 |
<div class="empty-state">
|
| 800 |
-
<div class="empty-icon">
|
| 801 |
-
<div
|
|
|
|
| 802 |
</div>
|
| 803 |
</div>
|
| 804 |
</main>
|
|
@@ -806,33 +318,28 @@
|
|
| 806 |
<!-- Right panel -->
|
| 807 |
<aside class="right-panel">
|
| 808 |
<div>
|
| 809 |
-
<div class="
|
| 810 |
<div class="score-card">
|
| 811 |
-
<div class="score-card-title">EPISODE SCORE</div>
|
| 812 |
<div class="big-score" id="current-score" style="color:var(--accent)">—</div>
|
| 813 |
<div class="score-label" id="score-status">No active episode</div>
|
| 814 |
-
|
| 815 |
-
<div style="margin-top:16px;" id="score-breakdown"></div>
|
| 816 |
</div>
|
| 817 |
</div>
|
| 818 |
-
|
| 819 |
<div>
|
| 820 |
-
<div class="
|
| 821 |
<div class="score-card">
|
| 822 |
<div class="mini-chart" id="reward-chart"></div>
|
| 823 |
-
<div style="display:flex;justify-content:space-between;margin-top:
|
| 824 |
<span>+1.0</span><span>avg: <span id="avg-reward">—</span></span><span>-1.0</span>
|
| 825 |
</div>
|
| 826 |
</div>
|
| 827 |
</div>
|
| 828 |
-
|
| 829 |
<div id="agents-section" style="display:none;">
|
| 830 |
-
<div class="
|
| 831 |
<div id="agents-container"></div>
|
| 832 |
</div>
|
| 833 |
-
|
| 834 |
<div id="queue-section" style="display:none;">
|
| 835 |
-
<div class="
|
| 836 |
<div id="queue-container"></div>
|
| 837 |
</div>
|
| 838 |
</aside>
|
|
@@ -844,12 +351,21 @@ const API = '';
|
|
| 844 |
let currentTaskId = 'ticket_classification';
|
| 845 |
let currentObs = null;
|
| 846 |
let rewardHistory = [];
|
| 847 |
-
let episodeScore = null;
|
| 848 |
|
| 849 |
const TASK_CONFIG = {
|
| 850 |
-
ticket_classification:
|
| 851 |
-
response_drafting:
|
| 852 |
-
queue_management:
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 853 |
};
|
| 854 |
|
| 855 |
async function api(method, path, body) {
|
|
@@ -863,18 +379,16 @@ async function api(method, path, body) {
|
|
| 863 |
function selectTask(taskId) {
|
| 864 |
currentTaskId = taskId;
|
| 865 |
document.querySelectorAll('.task-card').forEach(c => c.classList.remove('active'));
|
| 866 |
-
document.getElementById('task-' + taskId.split('_')[0]).classList.add('active');
|
| 867 |
-
// Handle 'classification' vs 'ticket_classification'
|
| 868 |
-
const map = { ticket_classification: 'classification', response_drafting: 'drafting', queue_management: 'queue' };
|
| 869 |
-
document.getElementById('task-' + map[taskId]).classList.add('active');
|
| 870 |
-
|
| 871 |
const cfg = TASK_CONFIG[taskId];
|
|
|
|
| 872 |
document.getElementById('center-title').textContent = cfg.title;
|
|
|
|
| 873 |
document.getElementById('max-steps').textContent = cfg.maxSteps;
|
|
|
|
| 874 |
document.getElementById('center-body').innerHTML = `
|
| 875 |
<div class="empty-state">
|
| 876 |
-
<div class="empty-icon">
|
| 877 |
-
<div
|
| 878 |
</div>`;
|
| 879 |
rewardHistory = [];
|
| 880 |
updateRewardChart();
|
|
@@ -885,243 +399,297 @@ function selectTask(taskId) {
|
|
| 885 |
document.getElementById('queue-section').style.display = 'none';
|
| 886 |
}
|
| 887 |
|
| 888 |
-
// Fix task card click handlers
|
| 889 |
-
document.getElementById('task-classification').onclick = () => selectTask('ticket_classification');
|
| 890 |
-
document.getElementById('task-drafting').onclick = () => selectTask('response_drafting');
|
| 891 |
-
document.getElementById('task-queue').onclick = () => selectTask('queue_management');
|
| 892 |
-
|
| 893 |
async function resetEpisode() {
|
| 894 |
rewardHistory = [];
|
| 895 |
-
episodeScore = null;
|
| 896 |
document.getElementById('log-container').innerHTML = '';
|
| 897 |
-
|
| 898 |
try {
|
| 899 |
const data = await api('POST', `/reset?task_id=${currentTaskId}`);
|
| 900 |
currentObs = data.observation;
|
| 901 |
document.getElementById('step-counter').textContent = currentObs.step || 0;
|
| 902 |
renderObservation(currentObs);
|
| 903 |
-
addLog(`
|
| 904 |
updateAgents(currentObs.agents);
|
| 905 |
updateQueue(currentObs.ticket_queue, currentObs.sla_status);
|
| 906 |
-
} catch(e) {
|
| 907 |
-
addLog(`❌ Reset failed: ${e.message}`, 'error');
|
| 908 |
-
}
|
| 909 |
}
|
| 910 |
|
| 911 |
function renderObservation(obs) {
|
| 912 |
-
if (obs.episode_done) {
|
| 913 |
-
fetchAndShowGrader();
|
| 914 |
-
return;
|
| 915 |
-
}
|
| 916 |
-
|
| 917 |
const body = document.getElementById('center-body');
|
|
|
|
|
|
|
|
|
|
|
|
|
| 918 |
|
| 919 |
-
if (
|
| 920 |
-
|
| 921 |
-
|
| 922 |
-
|
| 923 |
-
|
| 924 |
-
renderQueueView(obs, body);
|
| 925 |
-
}
|
| 926 |
|
| 927 |
updateAgents(obs.agents);
|
| 928 |
updateQueue(obs.ticket_queue, obs.sla_status);
|
| 929 |
}
|
| 930 |
|
| 931 |
-
function
|
| 932 |
-
|
| 933 |
-
const
|
| 934 |
-
const
|
| 935 |
-
const
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 936 |
|
| 937 |
-
|
| 938 |
<div class="ticket-card">
|
| 939 |
-
<div class="ticket-
|
| 940 |
-
|
| 941 |
-
|
| 942 |
-
|
| 943 |
-
|
| 944 |
-
<
|
| 945 |
-
|
| 946 |
-
</
|
| 947 |
-
|
| 948 |
-
<div class="ticket-body">${ticket.body}</div>
|
| 949 |
-
<div class="ticket-meta-row">
|
| 950 |
-
${ticket.category ? `<span class="meta-chip">📁 ${ticket.category}</span>` : ''}
|
| 951 |
-
${ticket.priority ? `<span class="meta-chip">🔺 ${ticket.priority}</span>` : ''}
|
| 952 |
-
${ticket.status ? `<span class="meta-chip">📌 ${ticket.status}</span>` : ''}
|
| 953 |
-
<span class="meta-chip">🕐 SLA: ${ticket.sla_deadline ? new Date(ticket.sla_deadline).toLocaleTimeString() : '—'}</span>
|
| 954 |
</div>
|
| 955 |
-
|
| 956 |
-
|
| 957 |
-
|
| 958 |
-
|
| 959 |
-
|
| 960 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 961 |
</div>
|
| 962 |
-
|
|
|
|
|
|
|
|
|
|
| 963 |
}
|
| 964 |
|
| 965 |
-
function
|
| 966 |
-
|
| 967 |
-
|
| 968 |
-
<
|
| 969 |
-
|
| 970 |
-
|
| 971 |
-
|
| 972 |
-
|
| 973 |
-
|
| 974 |
-
|
| 975 |
-
|
| 976 |
-
|
| 977 |
-
|
| 978 |
-
|
|
|
|
|
|
|
|
|
|
| 979 |
}
|
| 980 |
|
| 981 |
-
function
|
| 982 |
-
|
| 983 |
-
|
| 984 |
-
|
| 985 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 986 |
<div class="form-row">
|
| 987 |
-
<div class="form-group">
|
| 988 |
-
<label class="form-label">Category</label>
|
| 989 |
<select class="form-select" id="f-category">
|
| 990 |
-
<option value="">
|
| 991 |
-
|
| 992 |
-
|
| 993 |
-
).join('')}
|
| 994 |
-
</select>
|
| 995 |
-
</div>
|
| 996 |
-
<div class="form-group">
|
| 997 |
-
<label class="form-label">Priority</label>
|
| 998 |
<select class="form-select" id="f-priority">
|
| 999 |
-
<option value="">
|
| 1000 |
-
|
| 1001 |
-
`<option value="${p}">${p} — ${{'P1':'Critical','P2':'High','P3':'Medium','P4':'Low'}[p]}</option>`
|
| 1002 |
-
).join('')}
|
| 1003 |
-
</select>
|
| 1004 |
-
</div>
|
| 1005 |
-
</div>
|
| 1006 |
-
<div class="form-group" style="margin-bottom:12px;">
|
| 1007 |
-
<label class="form-label">Reasoning (optional)</label>
|
| 1008 |
-
<input class="form-input" id="f-reasoning" placeholder="Why did you choose this category/priority?"/>
|
| 1009 |
-
</div>
|
| 1010 |
-
<div class="submit-row">
|
| 1011 |
-
<button class="btn btn-primary" onclick="submitClassify('${ticket?.ticket_id}')">Submit Classification →</button>
|
| 1012 |
</div>
|
|
|
|
| 1013 |
</div>`;
|
| 1014 |
-
}
|
| 1015 |
|
| 1016 |
-
|
| 1017 |
-
|
| 1018 |
-
|
| 1019 |
-
|
| 1020 |
-
|
| 1021 |
-
<div class="
|
| 1022 |
-
<label class="form-label">Customer Response</label>
|
| 1023 |
-
<textarea class="form-textarea" id="f-response" rows="8" style="min-height:200px;"
|
| 1024 |
-
placeholder="Write a professional, empathetic response that addresses the customer's issue with concrete next steps..."></textarea>
|
| 1025 |
-
</div>
|
| 1026 |
-
<div class="submit-row">
|
| 1027 |
-
<button class="btn btn-primary" onclick="submitDraft('${ticket?.ticket_id}')">Send Response →</button>
|
| 1028 |
-
</div>
|
| 1029 |
</div>`;
|
| 1030 |
-
}
|
| 1031 |
-
|
| 1032 |
-
function renderQueueView(obs, body) {
|
| 1033 |
-
const queue = obs.ticket_queue || [];
|
| 1034 |
-
const info = obs.info || {};
|
| 1035 |
|
| 1036 |
-
|
| 1037 |
-
|
| 1038 |
-
|
| 1039 |
-
<div class="score-card-title">OPEN</div>
|
| 1040 |
-
<div style="font-family:var(--mono);font-size:32px;font-weight:700;color:var(--accent)">${info.open_count ?? queue.length}</div>
|
| 1041 |
-
</div>
|
| 1042 |
-
<div class="score-card" style="text-align:center;">
|
| 1043 |
-
<div class="score-card-title">RESOLVED</div>
|
| 1044 |
-
<div style="font-family:var(--mono);font-size:32px;font-weight:700;color:var(--accent3)">${info.resolved ?? 0}</div>
|
| 1045 |
-
</div>
|
| 1046 |
-
<div class="score-card" style="text-align:center;">
|
| 1047 |
-
<div class="score-card-title">SLA BREACHES</div>
|
| 1048 |
-
<div style="font-family:var(--mono);font-size:32px;font-weight:700;color:var(--accent2)">${info.sla_breaches ?? 0}</div>
|
| 1049 |
-
</div>
|
| 1050 |
-
</div>
|
| 1051 |
-
<div class="action-form">
|
| 1052 |
-
<div class="form-title">⚙️ Queue Action</div>
|
| 1053 |
<div class="form-row">
|
| 1054 |
-
<div class="form-group">
|
| 1055 |
-
<label class="form-label">Ticket ID</label>
|
| 1056 |
<select class="form-select" id="f-ticket-id">
|
| 1057 |
-
<option value="">
|
| 1058 |
-
|
| 1059 |
-
|
| 1060 |
-
|
| 1061 |
-
|
| 1062 |
-
<
|
| 1063 |
-
<select class="form-select" id="f-action-type" onchange="toggleQueueFields()">
|
| 1064 |
-
${['assign_ticket','escalate','resolve','close','no_op'].map(a =>
|
| 1065 |
-
`<option value="${a}">${a.replace('_',' ').toUpperCase()}</option>`
|
| 1066 |
-
).join('')}
|
| 1067 |
-
</select>
|
| 1068 |
-
</div>
|
| 1069 |
</div>
|
| 1070 |
-
<div id="agent-row" class="form-group" style="margin-bottom:
|
| 1071 |
-
<label class="form-label">Assign To Agent</label>
|
| 1072 |
<select class="form-select" id="f-agent-id">
|
| 1073 |
-
<option value="agent_billing">agent_billing —
|
| 1074 |
-
<option value="agent_tech">agent_tech —
|
| 1075 |
-
<option value="agent_general">agent_general —
|
| 1076 |
-
</select>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1077 |
</div>
|
| 1078 |
-
|
| 1079 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1080 |
</div>
|
|
|
|
| 1081 |
</div>`;
|
| 1082 |
-
}
|
| 1083 |
|
| 1084 |
-
|
| 1085 |
-
|
| 1086 |
-
|
| 1087 |
-
|
| 1088 |
-
|
|
|
|
| 1089 |
}
|
| 1090 |
-
}
|
| 1091 |
|
| 1092 |
-
|
| 1093 |
-
|
| 1094 |
-
|
| 1095 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1096 |
|
| 1097 |
-
if (
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1098 |
|
| 1099 |
-
|
| 1100 |
-
|
| 1101 |
-
|
| 1102 |
-
|
| 1103 |
-
|
| 1104 |
-
|
| 1105 |
-
}
|
| 1106 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1107 |
|
| 1108 |
-
|
| 1109 |
-
|
| 1110 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1111 |
|
| 1112 |
-
|
| 1113 |
-
action_type: 'draft_response',
|
| 1114 |
-
ticket_id: ticketId,
|
| 1115 |
-
response_text,
|
| 1116 |
-
});
|
| 1117 |
}
|
| 1118 |
|
| 1119 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1120 |
const action_type = document.getElementById('f-action-type').value;
|
| 1121 |
const ticket_id = document.getElementById('f-ticket-id')?.value;
|
| 1122 |
const target_agent_id = document.getElementById('f-agent-id')?.value;
|
| 1123 |
-
|
| 1124 |
-
await submitAction({ action_type, ticket_id, target_agent_id });
|
| 1125 |
}
|
| 1126 |
|
| 1127 |
async function submitAction(actionBody) {
|
|
@@ -1129,63 +697,48 @@ async function submitAction(actionBody) {
|
|
| 1129 |
const data = await api('POST', `/step?task_id=${currentTaskId}`, actionBody);
|
| 1130 |
currentObs = data.observation;
|
| 1131 |
const reward = data.reward;
|
| 1132 |
-
|
| 1133 |
rewardHistory.push(reward.total);
|
| 1134 |
document.getElementById('step-counter').textContent = currentObs.step || rewardHistory.length;
|
| 1135 |
-
|
| 1136 |
const sign = reward.total >= 0 ? '+' : '';
|
| 1137 |
-
addLog(`Step ${rewardHistory.length}: ${sign}${reward.total.toFixed(3)}
|
| 1138 |
-
|
| 1139 |
updateRewardChart();
|
| 1140 |
updateScoreDisplay(reward);
|
| 1141 |
renderObservation(currentObs);
|
| 1142 |
-
|
| 1143 |
-
|
| 1144 |
-
fetchAndShowGrader();
|
| 1145 |
-
}
|
| 1146 |
-
} catch(e) {
|
| 1147 |
-
addLog(`❌ Action failed: ${e.message}`, 'error');
|
| 1148 |
-
}
|
| 1149 |
}
|
| 1150 |
|
| 1151 |
-
async function
|
| 1152 |
try {
|
| 1153 |
const result = await api('POST', `/grader?task_id=${currentTaskId}`);
|
| 1154 |
-
|
| 1155 |
-
|
| 1156 |
-
|
| 1157 |
-
} catch(e) {
|
| 1158 |
-
console.error(e);
|
| 1159 |
-
}
|
| 1160 |
}
|
| 1161 |
|
| 1162 |
-
function
|
| 1163 |
-
const passed = result.passed;
|
| 1164 |
const score = result.final_score;
|
| 1165 |
const color = score >= 0.7 ? 'var(--accent3)' : score >= 0.4 ? 'var(--accent4)' : 'var(--accent2)';
|
| 1166 |
const emoji = score >= 0.7 ? '🏆' : score >= 0.4 ? '📊' : '📉';
|
| 1167 |
-
|
| 1168 |
document.getElementById('center-body').innerHTML = `
|
| 1169 |
<div class="episode-done">
|
| 1170 |
-
<div
|
| 1171 |
<div class="done-title">Episode Complete</div>
|
| 1172 |
<div class="done-score" style="color:${color}">${score.toFixed(4)}</div>
|
| 1173 |
-
<div style="font-family:var(--mono);font-size:
|
| 1174 |
-
${passed ? '✓ PASSED' : '✗ DID NOT PASS'} | ${rewardHistory.length} steps
|
| 1175 |
</div>
|
| 1176 |
-
|
| 1177 |
-
|
| 1178 |
-
<
|
| 1179 |
-
|
| 1180 |
-
|
| 1181 |
-
|
| 1182 |
-
</div>
|
| 1183 |
-
<button class="btn btn-primary" style="margin-top:24px;" onclick="resetEpisode()">↺ New Episode</button>
|
| 1184 |
</div>`;
|
| 1185 |
-
|
| 1186 |
document.getElementById('current-score').textContent = score.toFixed(4);
|
| 1187 |
document.getElementById('current-score').style.color = color;
|
| 1188 |
-
document.getElementById('score-status').textContent = passed ? '✓ Passed' : '✗ Did not pass';
|
| 1189 |
}
|
| 1190 |
|
| 1191 |
function updateScoreDisplay(reward) {
|
|
@@ -1193,107 +746,101 @@ function updateScoreDisplay(reward) {
|
|
| 1193 |
const color = total >= 0.6 ? 'var(--accent3)' : total >= 0.2 ? 'var(--accent4)' : 'var(--accent2)';
|
| 1194 |
document.getElementById('current-score').textContent = total.toFixed(4);
|
| 1195 |
document.getElementById('current-score').style.color = color;
|
| 1196 |
-
document.getElementById('score-status').textContent =
|
| 1197 |
-
|
| 1198 |
const bd = reward.breakdown || {};
|
| 1199 |
document.getElementById('score-breakdown').innerHTML = Object.entries(bd).map(([k,v]) => `
|
| 1200 |
<div class="breakdown-row">
|
| 1201 |
<span class="breakdown-label">${k.replace(/_/g,' ')}</span>
|
| 1202 |
-
<span class="breakdown-val ${v
|
| 1203 |
</div>`).join('');
|
| 1204 |
}
|
| 1205 |
|
| 1206 |
function updateRewardChart() {
|
| 1207 |
const container = document.getElementById('reward-chart');
|
| 1208 |
-
if (!rewardHistory.length) { container.innerHTML
|
| 1209 |
-
|
| 1210 |
-
const avg = (rewardHistory.reduce((a,b) => a+b, 0) / rewardHistory.length).toFixed(3);
|
| 1211 |
document.getElementById('avg-reward').textContent = avg;
|
| 1212 |
-
|
| 1213 |
container.innerHTML = rewardHistory.map(r => {
|
| 1214 |
-
const pct = ((r
|
| 1215 |
-
const color = r
|
| 1216 |
return `<div class="bar" style="height:${pct}%;background:${color};opacity:0.8;"></div>`;
|
| 1217 |
}).join('');
|
| 1218 |
}
|
| 1219 |
|
| 1220 |
function updateAgents(agents) {
|
| 1221 |
-
if (!agents
|
| 1222 |
-
|
| 1223 |
-
return;
|
| 1224 |
-
}
|
| 1225 |
-
document.getElementById('agents-section').style.display = 'block';
|
| 1226 |
document.getElementById('agents-container').innerHTML = agents.map(a => {
|
| 1227 |
-
const pct = (a.current_load
|
| 1228 |
-
|
| 1229 |
-
|
| 1230 |
-
<div
|
| 1231 |
-
|
| 1232 |
-
|
| 1233 |
-
Load: ${a.current_load}/${a.max_load} | ${a.specialization.join(', ')}
|
| 1234 |
-
</div>
|
| 1235 |
-
<div class="load-bar-track">
|
| 1236 |
-
<div class="load-bar-fill ${overloaded ? 'overloaded' : ''}" style="width:${pct}%"></div>
|
| 1237 |
-
</div>
|
| 1238 |
-
</div>`;
|
| 1239 |
}).join('');
|
| 1240 |
}
|
| 1241 |
|
| 1242 |
function updateQueue(queue, slaStatus) {
|
| 1243 |
-
if (!queue
|
| 1244 |
-
|
| 1245 |
-
return;
|
| 1246 |
-
}
|
| 1247 |
-
document.getElementById('queue-section').style.display = 'block';
|
| 1248 |
document.getElementById('queue-container').innerHTML = queue.slice(0,8).map(t => {
|
| 1249 |
-
const
|
| 1250 |
-
|
| 1251 |
-
|
| 1252 |
-
<
|
| 1253 |
-
|
| 1254 |
-
<span class="queue-subject">${t.subject}</span>
|
| 1255 |
-
<span class="queue-sla sla-${sla}">${sla.toUpperCase()}</span>
|
| 1256 |
-
</div>`;
|
| 1257 |
}).join('');
|
| 1258 |
}
|
| 1259 |
|
| 1260 |
function addLog(message, type) {
|
| 1261 |
const container = document.getElementById('log-container');
|
| 1262 |
const div = document.createElement('div');
|
| 1263 |
-
div.className = `log-entry ${type
|
| 1264 |
div.textContent = message;
|
| 1265 |
container.insertBefore(div, container.firstChild);
|
| 1266 |
-
if (container.children.length >
|
| 1267 |
}
|
| 1268 |
|
| 1269 |
async function runBaseline() {
|
| 1270 |
-
addLog('⚡ Running heuristic baseline...'
|
| 1271 |
try {
|
| 1272 |
const data = await api('POST', '/baseline');
|
| 1273 |
-
|
| 1274 |
-
|
| 1275 |
-
|
| 1276 |
-
|
| 1277 |
-
|
| 1278 |
-
}
|
| 1279 |
-
addLog(`${tid}: ${result.final_score.toFixed(4)} ${result.passed ? '✓' : '✗'}`, result.final_score > 0.5 ? 'pos' : 'neg');
|
| 1280 |
});
|
| 1281 |
-
|
| 1282 |
-
|
| 1283 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1284 |
}
|
| 1285 |
}
|
| 1286 |
|
| 1287 |
-
//
|
| 1288 |
(async () => {
|
| 1289 |
try {
|
| 1290 |
-
await api('GET', '/health');
|
| 1291 |
document.getElementById('server-status').textContent = 'LIVE';
|
|
|
|
| 1292 |
} catch(e) {
|
| 1293 |
document.getElementById('server-status').textContent = 'OFFLINE';
|
| 1294 |
document.getElementById('server-status').style.color = 'var(--accent2)';
|
|
|
|
|
|
|
|
|
|
| 1295 |
}
|
| 1296 |
})();
|
| 1297 |
</script>
|
| 1298 |
</body>
|
| 1299 |
-
</html>
|
|
|
|
| 3 |
<head>
|
| 4 |
<meta charset="UTF-8"/>
|
| 5 |
<meta name="viewport" content="width=device-width, initial-scale=1.0"/>
|
| 6 |
+
<title>OpenEnv — Multi-Domain AI Agent Environment</title>
|
| 7 |
<link rel="preconnect" href="https://fonts.googleapis.com">
|
| 8 |
<link href="https://fonts.googleapis.com/css2?family=Space+Mono:wght@400;700&family=Syne:wght@400;600;700;800&display=swap" rel="stylesheet">
|
| 9 |
<style>
|
| 10 |
:root {
|
| 11 |
+
--bg: #07070f;
|
| 12 |
+
--surface: #0f0f1a;
|
| 13 |
+
--surface2: #181826;
|
| 14 |
+
--border: #252538;
|
| 15 |
--accent: #6c63ff;
|
| 16 |
--accent2: #ff6584;
|
| 17 |
--accent3: #43e97b;
|
| 18 |
--accent4: #f7971e;
|
| 19 |
+
--accent5: #00d2ff;
|
| 20 |
--text: #e8e8f0;
|
| 21 |
+
--text-dim: #6868a0;
|
| 22 |
--mono: 'Space Mono', monospace;
|
| 23 |
--sans: 'Syne', sans-serif;
|
| 24 |
+
--radius: 10px;
|
|
|
|
| 25 |
}
|
|
|
|
| 26 |
* { box-sizing: border-box; margin: 0; padding: 0; }
|
| 27 |
+
body { background: var(--bg); color: var(--text); font-family: var(--sans); min-height: 100vh; overflow-x: hidden; }
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 28 |
body::before {
|
| 29 |
+
content: ''; position: fixed; inset: 0;
|
| 30 |
+
background-image: linear-gradient(rgba(108,99,255,0.03) 1px, transparent 1px), linear-gradient(90deg, rgba(108,99,255,0.03) 1px, transparent 1px);
|
| 31 |
+
background-size: 40px 40px; pointer-events: none; z-index: 0;
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 32 |
}
|
| 33 |
|
| 34 |
+
/* ── Layout ── */
|
| 35 |
+
.layout { position: relative; z-index: 1; display: grid; grid-template-rows: auto 1fr; min-height: 100vh; }
|
| 36 |
+
.main { display: grid; grid-template-columns: 260px 1fr 300px; height: calc(100vh - 72px); overflow: hidden; }
|
|
|
|
|
|
|
|
|
|
|
|
|
| 37 |
|
| 38 |
/* ── Header ── */
|
| 39 |
header {
|
| 40 |
+
display: flex; align-items: center; justify-content: space-between;
|
| 41 |
+
padding: 16px 28px; border-bottom: 1px solid var(--border);
|
| 42 |
+
background: rgba(7,7,15,0.9); backdrop-filter: blur(12px);
|
| 43 |
+
position: sticky; top: 0; z-index: 100;
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 44 |
}
|
| 45 |
+
.logo { display: flex; align-items: center; gap: 12px; }
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 46 |
.logo-icon {
|
| 47 |
+
width: 36px; height: 36px; border-radius: 8px;
|
|
|
|
| 48 |
background: linear-gradient(135deg, var(--accent), var(--accent2));
|
| 49 |
+
display: flex; align-items: center; justify-content: center; font-size: 18px;
|
|
|
|
|
|
|
|
|
|
|
|
|
| 50 |
}
|
| 51 |
+
.logo-name { font-size: 15px; font-weight: 800; letter-spacing: -0.3px; }
|
| 52 |
+
.logo-sub { font-family: var(--mono); font-size: 10px; color: var(--text-dim); letter-spacing: 1.5px; margin-top: 2px; }
|
| 53 |
+
.header-right { display: flex; align-items: center; gap: 16px; }
|
| 54 |
+
.status-pill {
|
| 55 |
+
display: flex; align-items: center; gap: 6px;
|
| 56 |
+
font-family: var(--mono); font-size: 11px; color: var(--accent3);
|
| 57 |
+
background: rgba(67,233,123,0.08); border: 1px solid rgba(67,233,123,0.2);
|
| 58 |
+
padding: 4px 12px; border-radius: 20px;
|
| 59 |
}
|
| 60 |
+
.pulse { width: 7px; height: 7px; background: var(--accent3); border-radius: 50%; animation: pulse 2s infinite; }
|
| 61 |
+
@keyframes pulse { 0%,100%{opacity:1;transform:scale(1)} 50%{opacity:0.4;transform:scale(0.7)} }
|
| 62 |
+
.overall-badge {
|
| 63 |
+
font-family: var(--mono); font-size: 12px; font-weight: 700;
|
| 64 |
+
color: var(--accent); background: rgba(108,99,255,0.1);
|
| 65 |
+
border: 1px solid rgba(108,99,255,0.3); padding: 4px 12px; border-radius: 20px;
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 66 |
}
|
| 67 |
|
| 68 |
/* ── Sidebar ── */
|
| 69 |
+
.sidebar { border-right: 1px solid var(--border); padding: 16px 14px; overflow-y: auto; display: flex; flex-direction: column; gap: 16px; }
|
| 70 |
+
.domain-label {
|
| 71 |
+
font-family: var(--mono); font-size: 9px; color: var(--text-dim);
|
| 72 |
+
letter-spacing: 2px; text-transform: uppercase; margin-bottom: 6px;
|
| 73 |
+
display: flex; align-items: center; gap: 6px;
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 74 |
}
|
| 75 |
+
.domain-label::after { content: ''; flex: 1; height: 1px; background: var(--border); }
|
| 76 |
.task-card {
|
| 77 |
+
background: var(--surface); border: 1px solid var(--border);
|
| 78 |
+
border-radius: var(--radius); padding: 10px 12px; cursor: pointer;
|
| 79 |
+
transition: all 0.15s; position: relative; overflow: hidden; margin-bottom: 4px;
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 80 |
}
|
| 81 |
+
.task-card::before { content: ''; position: absolute; left: 0; top: 0; bottom: 0; width: 3px; opacity: 0; transition: opacity 0.15s; }
|
| 82 |
+
.task-card:hover { border-color: rgba(108,99,255,0.4); background: var(--surface2); }
|
| 83 |
+
.task-card.active { border-color: var(--accent); background: var(--surface2); }
|
| 84 |
.task-card.active::before { opacity: 1; }
|
| 85 |
+
.task-name { font-size: 12px; font-weight: 700; margin-bottom: 4px; }
|
| 86 |
+
.task-meta { display: flex; align-items: center; gap: 6px; }
|
| 87 |
+
.badge { font-family: var(--mono); font-size: 9px; padding: 2px 6px; border-radius: 4px; font-weight: 700; letter-spacing: 0.5px; }
|
| 88 |
+
.badge-easy { background: rgba(67,233,123,0.12); color: var(--accent3); }
|
| 89 |
+
.badge-medium { background: rgba(247,151,30,0.12); color: var(--accent4); }
|
| 90 |
+
.badge-hard { background: rgba(255,101,132,0.12); color: var(--accent2); }
|
| 91 |
+
.badge-veryhard { background: rgba(108,99,255,0.15); color: var(--accent); }
|
| 92 |
+
.task-score { font-family: var(--mono); font-size: 10px; color: var(--text-dim); margin-left: auto; }
|
| 93 |
+
|
| 94 |
+
/* Domain colors for left bar */
|
| 95 |
+
.domain-support .task-card::before { background: var(--accent); }
|
| 96 |
+
.domain-legal .task-card::before { background: var(--accent4); }
|
| 97 |
+
.domain-clinical .task-card::before { background: var(--accent2); }
|
| 98 |
+
.domain-engineering .task-card::before { background: var(--accent5); }
|
| 99 |
+
|
| 100 |
+
/* ── Center ── */
|
| 101 |
+
.center { display: flex; flex-direction: column; overflow: hidden; }
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 102 |
.center-header {
|
| 103 |
+
padding: 16px 24px; border-bottom: 1px solid var(--border);
|
| 104 |
+
display: flex; align-items: center; justify-content: space-between; flex-shrink: 0;
|
| 105 |
+
}
|
| 106 |
+
.center-title { font-size: 18px; font-weight: 800; }
|
| 107 |
+
.center-sub { font-family: var(--mono); font-size: 11px; color: var(--text-dim); margin-top: 3px; }
|
| 108 |
+
.action-bar { display: flex; gap: 8px; }
|
| 109 |
+
.btn { font-family: var(--mono); font-size: 11px; padding: 7px 16px; border-radius: 8px; border: none; cursor: pointer; transition: all 0.15s; font-weight: 700; letter-spacing: 0.5px; }
|
| 110 |
+
.btn-primary { background: var(--accent); color: #fff; }
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 111 |
.btn-primary:hover { background: #7c74ff; transform: translateY(-1px); }
|
| 112 |
+
.btn-secondary { background: var(--surface2); color: var(--text); border: 1px solid var(--border); }
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 113 |
.btn-secondary:hover { border-color: var(--accent); }
|
| 114 |
+
.center-body { flex: 1; overflow-y: auto; padding: 20px 24px; }
|
| 115 |
+
|
| 116 |
+
/* ── Ticket card ── */
|
| 117 |
+
.ticket-card { background: var(--surface); border: 1px solid var(--border); border-radius: var(--radius); padding: 20px; margin-bottom: 14px; animation: slideIn 0.25s ease; }
|
| 118 |
+
@keyframes slideIn { from{opacity:0;transform:translateY(10px)} to{opacity:1;transform:translateY(0)} }
|
| 119 |
+
.ticket-id { font-family: var(--mono); font-size: 10px; color: var(--accent); margin-bottom: 4px; }
|
| 120 |
+
.ticket-subject { font-size: 16px; font-weight: 700; line-height: 1.3; margin-bottom: 12px; }
|
| 121 |
+
.ticket-body { font-size: 13px; line-height: 1.7; color: #b0b0cc; padding: 14px; background: var(--surface2); border-radius: 8px; border-left: 3px solid var(--border); margin-bottom: 14px; }
|
| 122 |
+
.ticket-meta { display: flex; gap: 10px; flex-wrap: wrap; }
|
| 123 |
+
.meta-chip { font-family: var(--mono); font-size: 10px; color: var(--text-dim); background: var(--surface2); padding: 3px 8px; border-radius: 5px; border: 1px solid var(--border); }
|
| 124 |
+
.tier-enterprise { color: var(--accent); border-color: rgba(108,99,255,0.3); }
|
| 125 |
+
.tier-pro { color: var(--accent4); border-color: rgba(247,151,30,0.3); }
|
| 126 |
+
|
| 127 |
+
/* ── Domain cards (legal/clinical/pr) ── */
|
| 128 |
+
.domain-card { background: var(--surface); border: 1px solid var(--border); border-radius: var(--radius); padding: 20px; margin-bottom: 14px; animation: slideIn 0.25s ease; }
|
| 129 |
+
.domain-card-label { font-family: var(--mono); font-size: 10px; color: var(--text-dim); letter-spacing: 1.5px; text-transform: uppercase; margin-bottom: 6px; }
|
| 130 |
+
.domain-card-title { font-size: 15px; font-weight: 700; margin-bottom: 12px; line-height: 1.4; }
|
| 131 |
+
.domain-card-body { font-size: 13px; line-height: 1.7; color: #b0b0cc; padding: 14px; background: var(--surface2); border-radius: 8px; border-left: 3px solid var(--border); }
|
| 132 |
+
.code-diff { font-family: var(--mono); font-size: 11px; background: #0a0a14; border: 1px solid var(--border); border-radius: 8px; padding: 14px; overflow-x: auto; line-height: 1.6; white-space: pre; }
|
| 133 |
+
.diff-add { color: var(--accent3); }
|
| 134 |
+
.diff-remove { color: var(--accent2); }
|
| 135 |
+
.diff-header { color: var(--text-dim); }
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 136 |
|
| 137 |
/* ── Action form ── */
|
| 138 |
+
.action-form { background: var(--surface); border: 1px solid var(--border); border-radius: var(--radius); padding: 20px; margin-top: 14px; }
|
| 139 |
+
.form-title { font-weight: 700; font-size: 14px; margin-bottom: 14px; }
|
| 140 |
+
.form-row { display: grid; grid-template-columns: 1fr 1fr; gap: 10px; margin-bottom: 10px; }
|
| 141 |
+
.form-group { display: flex; flex-direction: column; gap: 5px; }
|
| 142 |
+
.form-label { font-family: var(--mono); font-size: 9px; color: var(--text-dim); letter-spacing: 1px; text-transform: uppercase; }
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 143 |
.form-select, .form-textarea, .form-input {
|
| 144 |
+
background: var(--surface2); border: 1px solid var(--border); border-radius: 7px;
|
| 145 |
+
color: var(--text); font-family: var(--mono); font-size: 12px; padding: 8px 10px;
|
| 146 |
+
transition: border-color 0.15s; width: 100%;
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 147 |
}
|
| 148 |
+
.form-select:focus, .form-textarea:focus, .form-input:focus { outline: none; border-color: var(--accent); }
|
| 149 |
+
.form-textarea { font-family: var(--sans); font-size: 13px; resize: vertical; min-height: 100px; line-height: 1.6; }
|
| 150 |
+
.submit-row { display: flex; justify-content: flex-end; margin-top: 14px; }
|
| 151 |
|
| 152 |
/* ── Right panel ── */
|
| 153 |
+
.right-panel { border-left: 1px solid var(--border); padding: 16px 14px; overflow-y: auto; display: flex; flex-direction: column; gap: 16px; }
|
| 154 |
+
.panel-label { font-family: var(--mono); font-size: 9px; color: var(--text-dim); letter-spacing: 2px; text-transform: uppercase; margin-bottom: 10px; }
|
| 155 |
+
.score-card { background: var(--surface); border: 1px solid var(--border); border-radius: var(--radius); padding: 16px; }
|
| 156 |
+
.big-score { font-family: var(--mono); font-size: 42px; font-weight: 700; line-height: 1; }
|
| 157 |
+
.score-label { font-family: var(--mono); font-size: 10px; color: var(--text-dim); margin-top: 4px; }
|
| 158 |
+
.breakdown-row { display: flex; justify-content: space-between; align-items: center; padding: 6px 0; border-bottom: 1px solid rgba(37,37,56,0.5); }
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 159 |
.breakdown-row:last-child { border-bottom: none; }
|
| 160 |
+
.breakdown-label { font-family: var(--mono); font-size: 10px; color: var(--text-dim); }
|
| 161 |
+
.breakdown-val { font-family: var(--mono); font-size: 11px; font-weight: 700; }
|
| 162 |
+
.val-good { color: var(--accent3); } .val-mid { color: var(--accent4); } .val-bad { color: var(--accent2); }
|
| 163 |
+
.mini-chart { height: 50px; display: flex; align-items: flex-end; gap: 2px; }
|
| 164 |
+
.bar { flex: 1; border-radius: 2px 2px 0 0; min-height: 3px; transition: height 0.3s ease; }
|
| 165 |
+
.log-entry { font-family: var(--mono); font-size: 10px; color: var(--text-dim); padding: 5px 8px; border-radius: 4px; margin-bottom: 3px; background: var(--surface); }
|
| 166 |
+
.log-pos { color: var(--accent3); } .log-neg { color: var(--accent2); }
|
| 167 |
+
|
| 168 |
+
/* ── Queue ── */
|
| 169 |
+
.queue-item { display: flex; align-items: center; gap: 8px; padding: 8px 10px; background: var(--surface); border: 1px solid var(--border); border-radius: 7px; margin-bottom: 5px; font-size: 11px; }
|
| 170 |
+
.priority-dot { width: 7px; height: 7px; border-radius: 50%; flex-shrink: 0; }
|
| 171 |
+
.p1 { background: var(--accent2); box-shadow: 0 0 5px var(--accent2); } .p2 { background: var(--accent4); } .p3 { background: var(--accent3); } .p4 { background: var(--text-dim); }
|
| 172 |
+
.queue-subject { flex: 1; overflow: hidden; text-overflow: ellipsis; white-space: nowrap; }
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 173 |
|
| 174 |
/* ── Episode done ── */
|
| 175 |
+
.episode-done { background: linear-gradient(135deg, var(--surface2), rgba(108,99,255,0.06)); border: 1px solid rgba(108,99,255,0.25); border-radius: var(--radius); padding: 28px; text-align: center; animation: slideIn 0.35s ease; }
|
| 176 |
+
.done-score { font-family: var(--mono); font-size: 56px; font-weight: 700; color: var(--accent); }
|
| 177 |
+
.done-title { font-size: 20px; font-weight: 800; margin-bottom: 6px; }
|
| 178 |
+
.empty-state { display: flex; flex-direction: column; align-items: center; justify-content: center; height: 280px; color: var(--text-dim); text-align: center; gap: 10px; }
|
| 179 |
+
.empty-icon { font-size: 44px; opacity: 0.25; }
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 180 |
|
| 181 |
+
/* ── Agent cards ── */
|
| 182 |
+
.agent-card { background: var(--surface); border: 1px solid var(--border); border-radius: 8px; padding: 10px; margin-bottom: 6px; }
|
| 183 |
+
.agent-name { font-size: 12px; font-weight: 700; margin-bottom: 6px; }
|
| 184 |
+
.load-bar-track { height: 5px; background: var(--surface2); border-radius: 3px; overflow: hidden; margin-top: 5px; }
|
| 185 |
+
.load-bar-fill { height: 100%; border-radius: 3px; background: linear-gradient(90deg, var(--accent3), var(--accent)); transition: width 0.4s; }
|
| 186 |
+
.load-bar-fill.over { background: linear-gradient(90deg, var(--accent4), var(--accent2)); }
|
| 187 |
+
|
| 188 |
+
/* ── Stats bar ── */
|
| 189 |
+
.stats-bar { display: grid; grid-template-columns: repeat(3, 1fr); gap: 8px; margin-bottom: 14px; }
|
| 190 |
+
.stat-box { background: var(--surface); border: 1px solid var(--border); border-radius: 8px; padding: 12px; text-align: center; }
|
| 191 |
+
.stat-num { font-family: var(--mono); font-size: 26px; font-weight: 700; }
|
| 192 |
+
.stat-label { font-family: var(--mono); font-size: 9px; color: var(--text-dim); letter-spacing: 1px; margin-top: 3px; }
|
| 193 |
+
|
| 194 |
+
::-webkit-scrollbar { width: 4px; } ::-webkit-scrollbar-track { background: transparent; } ::-webkit-scrollbar-thumb { background: var(--border); border-radius: 2px; }
|
| 195 |
</style>
|
| 196 |
</head>
|
| 197 |
<body>
|
| 198 |
<div class="layout">
|
| 199 |
<header>
|
| 200 |
<div class="logo">
|
| 201 |
+
<div class="logo-icon">🌐</div>
|
| 202 |
<div>
|
| 203 |
+
<div class="logo-name">OpenEnv — Multi-Domain</div>
|
| 204 |
+
<div class="logo-sub">AI AGENT EVALUATION ENVIRONMENT</div>
|
| 205 |
</div>
|
| 206 |
</div>
|
| 207 |
+
<div class="header-right">
|
| 208 |
+
<div class="overall-badge">OVERALL: <span id="overall-score">—</span></div>
|
| 209 |
+
<div class="status-pill"><div class="pulse"></div><span id="server-status">CONNECTING...</span></div>
|
| 210 |
</div>
|
| 211 |
</header>
|
| 212 |
|
| 213 |
<div class="main">
|
| 214 |
<!-- Sidebar -->
|
| 215 |
<aside class="sidebar">
|
| 216 |
+
|
| 217 |
+
<div class="domain-support">
|
| 218 |
+
<div class="domain-label">🎫 Customer Support</div>
|
| 219 |
+
<div class="task-card active" id="tc-card" onclick="selectTask('ticket_classification')">
|
| 220 |
<div class="task-name">Ticket Classification</div>
|
| 221 |
+
<div class="task-meta"><span class="badge badge-easy">EASY</span><span class="task-score" id="score-ticket_classification">—</span></div>
|
|
|
|
|
|
|
|
|
|
|
|
|
| 222 |
</div>
|
| 223 |
+
<div class="task-card" id="rd-card" onclick="selectTask('response_drafting')">
|
| 224 |
<div class="task-name">Response Drafting</div>
|
| 225 |
+
<div class="task-meta"><span class="badge badge-medium">MEDIUM</span><span class="task-score" id="score-response_drafting">—</span></div>
|
|
|
|
|
|
|
|
|
|
|
|
|
| 226 |
</div>
|
| 227 |
+
<div class="task-card" id="qm-card" onclick="selectTask('queue_management')">
|
| 228 |
+
<div class="task-name">SLA Queue Management</div>
|
| 229 |
+
<div class="task-meta"><span class="badge badge-hard">HARD</span><span class="task-score" id="score-queue_management">—</span></div>
|
| 230 |
+
</div>
|
| 231 |
+
<div class="task-card" id="mt-card" onclick="selectTask('multi_turn_conversation')">
|
| 232 |
+
<div class="task-name">Multi-Turn De-escalation</div>
|
| 233 |
+
<div class="task-meta"><span class="badge badge-veryhard">VERY HARD</span><span class="task-score" id="score-multi_turn_conversation">—</span></div>
|
| 234 |
+
</div>
|
| 235 |
+
</div>
|
| 236 |
+
|
| 237 |
+
<div class="domain-legal">
|
| 238 |
+
<div class="domain-label">⚖️ Legal Review</div>
|
| 239 |
+
<div class="task-card" id="lci-card" onclick="selectTask('legal_clause_identification')">
|
| 240 |
+
<div class="task-name">Clause Identification</div>
|
| 241 |
+
<div class="task-meta"><span class="badge badge-easy">EASY</span><span class="task-score" id="score-legal_clause_identification">—</span></div>
|
| 242 |
+
</div>
|
| 243 |
+
<div class="task-card" id="lrf-card" onclick="selectTask('legal_risk_flagging')">
|
| 244 |
+
<div class="task-name">Risk Flagging</div>
|
| 245 |
+
<div class="task-meta"><span class="badge badge-medium">MEDIUM</span><span class="task-score" id="score-legal_risk_flagging">—</span></div>
|
| 246 |
+
</div>
|
| 247 |
+
<div class="task-card" id="lcr-card" onclick="selectTask('legal_clause_redlining')">
|
| 248 |
+
<div class="task-name">Clause Redlining</div>
|
| 249 |
+
<div class="task-meta"><span class="badge badge-hard">HARD</span><span class="task-score" id="score-legal_clause_redlining">—</span></div>
|
| 250 |
+
</div>
|
| 251 |
+
</div>
|
| 252 |
+
|
| 253 |
+
<div class="domain-clinical">
|
| 254 |
+
<div class="domain-label">🏥 Clinical Triage</div>
|
| 255 |
+
<div class="task-card" id="ctc-card" onclick="selectTask('clinical_triage_classification')">
|
| 256 |
+
<div class="task-name">Body System Classification</div>
|
| 257 |
+
<div class="task-meta"><span class="badge badge-easy">EASY</span><span class="task-score" id="score-clinical_triage_classification">—</span></div>
|
| 258 |
+
</div>
|
| 259 |
+
<div class="task-card" id="esi-card" onclick="selectTask('clinical_esi_assignment')">
|
| 260 |
+
<div class="task-name">ESI Level Assignment</div>
|
| 261 |
+
<div class="task-meta"><span class="badge badge-medium">MEDIUM</span><span class="task-score" id="score-clinical_esi_assignment">—</span></div>
|
| 262 |
+
</div>
|
| 263 |
+
<div class="task-card" id="ctn-card" onclick="selectTask('clinical_triage_note')">
|
| 264 |
+
<div class="task-name">Triage Note Generation</div>
|
| 265 |
+
<div class="task-meta"><span class="badge badge-hard">HARD</span><span class="task-score" id="score-clinical_triage_note">—</span></div>
|
| 266 |
+
</div>
|
| 267 |
+
</div>
|
| 268 |
+
|
| 269 |
+
<div class="domain-engineering">
|
| 270 |
+
<div class="domain-label">💻 PR Review</div>
|
| 271 |
+
<div class="task-card" id="ptc-card" onclick="selectTask('pr_type_classification')">
|
| 272 |
+
<div class="task-name">PR Type Classification</div>
|
| 273 |
+
<div class="task-meta"><span class="badge badge-easy">EASY</span><span class="task-score" id="score-pr_type_classification">—</span></div>
|
| 274 |
+
</div>
|
| 275 |
+
<div class="task-card" id="pbi-card" onclick="selectTask('pr_bug_identification')">
|
| 276 |
+
<div class="task-name">Bug Identification</div>
|
| 277 |
+
<div class="task-meta"><span class="badge badge-medium">MEDIUM</span><span class="task-score" id="score-pr_bug_identification">—</span></div>
|
| 278 |
+
</div>
|
| 279 |
+
<div class="task-card" id="prc-card" onclick="selectTask('pr_review_comment')">
|
| 280 |
+
<div class="task-name">Code Review Comment</div>
|
| 281 |
+
<div class="task-meta"><span class="badge badge-hard">HARD</span><span class="task-score" id="score-pr_review_comment">—</span></div>
|
| 282 |
</div>
|
| 283 |
</div>
|
| 284 |
|
| 285 |
<div>
|
| 286 |
+
<div class="domain-label">Episode Log</div>
|
| 287 |
+
<div id="log-container" style="max-height:200px;overflow-y:auto;"></div>
|
| 288 |
</div>
|
| 289 |
|
| 290 |
<div>
|
| 291 |
+
<div class="domain-label">API Links</div>
|
| 292 |
+
<a href="/docs" target="_blank" style="display:flex;align-items:center;gap:8px;color:var(--accent);font-size:12px;text-decoration:none;padding:8px;background:var(--surface);border-radius:8px;border:1px solid var(--border);margin-bottom:5px;">📖 Swagger API Docs</a>
|
| 293 |
+
<a href="/tasks" target="_blank" style="display:flex;align-items:center;gap:8px;color:var(--text-dim);font-size:12px;text-decoration:none;padding:8px;background:var(--surface);border-radius:8px;border:1px solid var(--border);">📋 /tasks endpoint</a>
|
|
|
|
|
|
|
|
|
|
|
|
|
| 294 |
</div>
|
| 295 |
</aside>
|
| 296 |
|
|
|
|
| 299 |
<div class="center-header">
|
| 300 |
<div>
|
| 301 |
<div class="center-title" id="center-title">Ticket Classification</div>
|
| 302 |
+
<div class="center-sub">Step <span id="step-counter">0</span> / <span id="max-steps">10</span> | <span id="domain-label">Customer Support</span></div>
|
|
|
|
|
|
|
| 303 |
</div>
|
| 304 |
<div class="action-bar">
|
| 305 |
+
<button class="btn btn-secondary" onclick="runBaseline()">▶ Run Heuristic Baseline</button>
|
| 306 |
<button class="btn btn-primary" onclick="resetEpisode()">↺ Reset Episode</button>
|
| 307 |
</div>
|
| 308 |
</div>
|
|
|
|
| 309 |
<div class="center-body" id="center-body">
|
| 310 |
<div class="empty-state">
|
| 311 |
+
<div class="empty-icon">🌐</div>
|
| 312 |
+
<div style="font-size:14px;font-weight:700;">Multi-Domain AI Agent Environment</div>
|
| 313 |
+
<div style="font-size:12px;max-width:300px;line-height:1.6;">13 tasks across 4 professional domains. Select a task from the sidebar and click Reset Episode to begin.</div>
|
| 314 |
</div>
|
| 315 |
</div>
|
| 316 |
</main>
|
|
|
|
| 318 |
<!-- Right panel -->
|
| 319 |
<aside class="right-panel">
|
| 320 |
<div>
|
| 321 |
+
<div class="panel-label">Episode Score</div>
|
| 322 |
<div class="score-card">
|
|
|
|
| 323 |
<div class="big-score" id="current-score" style="color:var(--accent)">—</div>
|
| 324 |
<div class="score-label" id="score-status">No active episode</div>
|
| 325 |
+
<div style="margin-top:12px;" id="score-breakdown"></div>
|
|
|
|
| 326 |
</div>
|
| 327 |
</div>
|
|
|
|
| 328 |
<div>
|
| 329 |
+
<div class="panel-label">Reward History</div>
|
| 330 |
<div class="score-card">
|
| 331 |
<div class="mini-chart" id="reward-chart"></div>
|
| 332 |
+
<div style="display:flex;justify-content:space-between;margin-top:5px;font-family:var(--mono);font-size:9px;color:var(--text-dim);">
|
| 333 |
<span>+1.0</span><span>avg: <span id="avg-reward">—</span></span><span>-1.0</span>
|
| 334 |
</div>
|
| 335 |
</div>
|
| 336 |
</div>
|
|
|
|
| 337 |
<div id="agents-section" style="display:none;">
|
| 338 |
+
<div class="panel-label">Agent Status</div>
|
| 339 |
<div id="agents-container"></div>
|
| 340 |
</div>
|
|
|
|
| 341 |
<div id="queue-section" style="display:none;">
|
| 342 |
+
<div class="panel-label">Ticket Queue</div>
|
| 343 |
<div id="queue-container"></div>
|
| 344 |
</div>
|
| 345 |
</aside>
|
|
|
|
| 351 |
let currentTaskId = 'ticket_classification';
|
| 352 |
let currentObs = null;
|
| 353 |
let rewardHistory = [];
|
|
|
|
| 354 |
|
| 355 |
const TASK_CONFIG = {
|
| 356 |
+
ticket_classification: { title: 'Ticket Classification', domain: 'Customer Support', maxSteps: 10, cardId: 'tc-card' },
|
| 357 |
+
response_drafting: { title: 'Response Drafting', domain: 'Customer Support', maxSteps: 6, cardId: 'rd-card' },
|
| 358 |
+
queue_management: { title: 'SLA Queue Management', domain: 'Customer Support', maxSteps: 40, cardId: 'qm-card' },
|
| 359 |
+
multi_turn_conversation: { title: 'Multi-Turn De-escalation', domain: 'Customer Support', maxSteps: 8, cardId: 'mt-card' },
|
| 360 |
+
legal_clause_identification: { title: 'Legal Clause Identification', domain: 'Legal Review', maxSteps: 10, cardId: 'lci-card' },
|
| 361 |
+
legal_risk_flagging: { title: 'Legal Risk Flagging', domain: 'Legal Review', maxSteps: 10, cardId: 'lrf-card' },
|
| 362 |
+
legal_clause_redlining: { title: 'Legal Clause Redlining', domain: 'Legal Review', maxSteps: 5, cardId: 'lcr-card' },
|
| 363 |
+
clinical_triage_classification: { title: 'Clinical Triage Classification', domain: 'Clinical Triage', maxSteps: 10, cardId: 'ctc-card' },
|
| 364 |
+
clinical_esi_assignment: { title: 'ESI Level Assignment', domain: 'Clinical Triage', maxSteps: 10, cardId: 'esi-card' },
|
| 365 |
+
clinical_triage_note: { title: 'Triage Note Generation', domain: 'Clinical Triage', maxSteps: 5, cardId: 'ctn-card' },
|
| 366 |
+
pr_type_classification: { title: 'PR Type Classification', domain: 'PR Review', maxSteps: 10, cardId: 'ptc-card' },
|
| 367 |
+
pr_bug_identification: { title: 'PR Bug Identification', domain: 'PR Review', maxSteps: 5, cardId: 'pbi-card' },
|
| 368 |
+
pr_review_comment: { title: 'PR Code Review', domain: 'PR Review', maxSteps: 5, cardId: 'prc-card' },
|
| 369 |
};
|
| 370 |
|
| 371 |
async function api(method, path, body) {
|
|
|
|
| 379 |
function selectTask(taskId) {
|
| 380 |
currentTaskId = taskId;
|
| 381 |
document.querySelectorAll('.task-card').forEach(c => c.classList.remove('active'));
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 382 |
const cfg = TASK_CONFIG[taskId];
|
| 383 |
+
document.getElementById(cfg.cardId).classList.add('active');
|
| 384 |
document.getElementById('center-title').textContent = cfg.title;
|
| 385 |
+
document.getElementById('domain-label').textContent = cfg.domain;
|
| 386 |
document.getElementById('max-steps').textContent = cfg.maxSteps;
|
| 387 |
+
document.getElementById('step-counter').textContent = 0;
|
| 388 |
document.getElementById('center-body').innerHTML = `
|
| 389 |
<div class="empty-state">
|
| 390 |
+
<div class="empty-icon">▶</div>
|
| 391 |
+
<div style="font-size:13px;">Click "Reset Episode" to start this task</div>
|
| 392 |
</div>`;
|
| 393 |
rewardHistory = [];
|
| 394 |
updateRewardChart();
|
|
|
|
| 399 |
document.getElementById('queue-section').style.display = 'none';
|
| 400 |
}
|
| 401 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 402 |
async function resetEpisode() {
|
| 403 |
rewardHistory = [];
|
|
|
|
| 404 |
document.getElementById('log-container').innerHTML = '';
|
|
|
|
| 405 |
try {
|
| 406 |
const data = await api('POST', `/reset?task_id=${currentTaskId}`);
|
| 407 |
currentObs = data.observation;
|
| 408 |
document.getElementById('step-counter').textContent = currentObs.step || 0;
|
| 409 |
renderObservation(currentObs);
|
| 410 |
+
addLog(`↺ Reset — ${currentTaskId}`);
|
| 411 |
updateAgents(currentObs.agents);
|
| 412 |
updateQueue(currentObs.ticket_queue, currentObs.sla_status);
|
| 413 |
+
} catch(e) { addLog(`❌ ${e.message}`, 'neg'); }
|
|
|
|
|
|
|
| 414 |
}
|
| 415 |
|
| 416 |
function renderObservation(obs) {
|
| 417 |
+
if (obs.episode_done) { fetchGrader(); return; }
|
|
|
|
|
|
|
|
|
|
|
|
|
| 418 |
const body = document.getElementById('center-body');
|
| 419 |
+
const ticket = obs.current_ticket;
|
| 420 |
+
const clause = obs.current_clause;
|
| 421 |
+
const patient = obs.current_patient;
|
| 422 |
+
const pr = obs.current_pr;
|
| 423 |
|
| 424 |
+
if (ticket) renderTicketView(obs, body);
|
| 425 |
+
else if (clause) renderLegalView(obs, body);
|
| 426 |
+
else if (patient) renderClinicalView(obs, body);
|
| 427 |
+
else if (pr) renderPRView(obs, body);
|
| 428 |
+
else body.innerHTML = `<div class="empty-state"><div class="empty-icon">✓</div><div>Step complete</div></div>`;
|
|
|
|
|
|
|
| 429 |
|
| 430 |
updateAgents(obs.agents);
|
| 431 |
updateQueue(obs.ticket_queue, obs.sla_status);
|
| 432 |
}
|
| 433 |
|
| 434 |
+
function renderTicketView(obs, body) {
|
| 435 |
+
const t = obs.current_ticket;
|
| 436 |
+
const isQueue = currentTaskId === 'queue_management';
|
| 437 |
+
const isMT = currentTaskId === 'multi_turn_conversation';
|
| 438 |
+
const history = t.previous_interactions || [];
|
| 439 |
+
|
| 440 |
+
let historyHTML = '';
|
| 441 |
+
if (history.length > 0) {
|
| 442 |
+
historyHTML = `<div style="margin-bottom:12px;"><div style="font-family:var(--mono);font-size:10px;color:var(--text-dim);margin-bottom:8px;">CONVERSATION HISTORY</div>` +
|
| 443 |
+
history.map(m => `<div style="padding:8px;margin-bottom:6px;border-radius:6px;background:${m.role==='customer'?'var(--surface2)':'rgba(108,99,255,0.08)'};border:1px solid var(--border);font-size:12px;"><span style="font-family:var(--mono);font-size:9px;color:var(--text-dim);text-transform:uppercase;">${m.role}:</span> ${m.content}</div>`).join('') + `</div>`;
|
| 444 |
+
}
|
| 445 |
|
| 446 |
+
body.innerHTML = `
|
| 447 |
<div class="ticket-card">
|
| 448 |
+
<div class="ticket-id">${t.ticket_id}</div>
|
| 449 |
+
<div class="ticket-subject">${t.subject}</div>
|
| 450 |
+
${historyHTML}
|
| 451 |
+
<div class="ticket-body">${t.body}</div>
|
| 452 |
+
<div class="ticket-meta">
|
| 453 |
+
<span class="meta-chip tier-${t.customer_tier}">${t.customer_tier?.toUpperCase()}</span>
|
| 454 |
+
${t.category ? `<span class="meta-chip">📁 ${t.category}</span>` : ''}
|
| 455 |
+
${t.priority ? `<span class="meta-chip">🔺 ${t.priority}</span>` : ''}
|
| 456 |
+
<span class="meta-chip">😶 Sentiment: ${(t.sentiment_score||0).toFixed(2)}</span>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 457 |
</div>
|
| 458 |
+
</div>
|
| 459 |
+
${obs.knowledge_base?.length ? renderKB(obs.knowledge_base) : ''}
|
| 460 |
+
${renderActionForm(obs)}`;
|
| 461 |
+
}
|
| 462 |
+
|
| 463 |
+
function renderLegalView(obs, body) {
|
| 464 |
+
const c = obs.current_clause;
|
| 465 |
+
body.innerHTML = `
|
| 466 |
+
<div class="domain-card">
|
| 467 |
+
<div class="domain-card-label">${c.contract_type} — ${c.counterparty}</div>
|
| 468 |
+
<div class="domain-card-title">Clause ${c.clause_id}</div>
|
| 469 |
+
<div class="domain-card-body">${c.text}</div>
|
| 470 |
+
${c.true_risk_level ? `<div style="margin-top:10px;font-family:var(--mono);font-size:11px;color:var(--accent4);">Known Risk Level: ${c.true_risk_level}</div>` : ''}
|
| 471 |
+
</div>
|
| 472 |
+
${renderActionForm(obs)}`;
|
| 473 |
+
}
|
| 474 |
+
|
| 475 |
+
function renderClinicalView(obs, body) {
|
| 476 |
+
const p = obs.current_patient;
|
| 477 |
+
const vitals = p.vitals || {};
|
| 478 |
+
body.innerHTML = `
|
| 479 |
+
<div class="domain-card">
|
| 480 |
+
<div class="domain-card-label">Patient ${p.patient_id} — Age ${p.age} ${p.gender}</div>
|
| 481 |
+
<div class="domain-card-title">${p.chief_complaint}</div>
|
| 482 |
+
<div style="display:grid;grid-template-columns:repeat(5,1fr);gap:8px;margin:12px 0;">
|
| 483 |
+
${Object.entries(vitals).map(([k,v]) => `<div style="background:var(--surface2);border:1px solid var(--border);border-radius:6px;padding:8px;text-align:center;"><div style="font-family:var(--mono);font-size:14px;font-weight:700;">${v}</div><div style="font-family:var(--mono);font-size:9px;color:var(--text-dim);margin-top:2px;">${k}</div></div>`).join('')}
|
| 484 |
</div>
|
| 485 |
+
<div class="domain-card-body"><strong>History:</strong> ${p.history}</div>
|
| 486 |
+
${p.true_esi_level ? `<div style="margin-top:10px;font-family:var(--mono);font-size:11px;color:var(--accent4);">Known ESI Level: ${p.true_esi_level}</div>` : ''}
|
| 487 |
+
</div>
|
| 488 |
+
${renderActionForm(obs)}`;
|
| 489 |
}
|
| 490 |
|
| 491 |
+
function renderPRView(obs, body) {
|
| 492 |
+
const pr = obs.current_pr;
|
| 493 |
+
const diffLines = (pr.diff||'').split('\n').map(line => {
|
| 494 |
+
if (line.startsWith('+')) return `<span class="diff-add">${escHtml(line)}</span>`;
|
| 495 |
+
if (line.startsWith('-')) return `<span class="diff-remove">${escHtml(line)}</span>`;
|
| 496 |
+
if (line.startsWith('@@')) return `<span class="diff-header">${escHtml(line)}</span>`;
|
| 497 |
+
return escHtml(line);
|
| 498 |
+
}).join('\n');
|
| 499 |
+
body.innerHTML = `
|
| 500 |
+
<div class="domain-card">
|
| 501 |
+
<div class="domain-card-label">PR ${pr.pr_id} — by ${pr.author}</div>
|
| 502 |
+
<div class="domain-card-title">${pr.title}</div>
|
| 503 |
+
<div class="domain-card-body" style="margin-bottom:12px;">${pr.description}</div>
|
| 504 |
+
<div class="code-diff">${diffLines}</div>
|
| 505 |
+
${pr.true_bug_description ? `<div style="margin-top:10px;font-family:var(--mono);font-size:11px;color:var(--accent4);">Known Bug: ${pr.true_bug_description}</div>` : ''}
|
| 506 |
+
</div>
|
| 507 |
+
${renderActionForm(obs)}`;
|
| 508 |
}
|
| 509 |
|
| 510 |
+
function escHtml(s) { return s.replace(/&/g,'&').replace(/</g,'<').replace(/>/g,'>'); }
|
| 511 |
+
|
| 512 |
+
function renderKB(articles) {
|
| 513 |
+
return `<div style="margin-bottom:14px;">
|
| 514 |
+
<div style="font-family:var(--mono);font-size:9px;color:var(--text-dim);letter-spacing:2px;text-transform:uppercase;margin-bottom:8px;">Knowledge Base</div>
|
| 515 |
+
${articles.slice(0,2).map(a => `
|
| 516 |
+
<div style="background:var(--surface);border:1px solid var(--border);border-radius:8px;padding:12px;margin-bottom:6px;">
|
| 517 |
+
<div style="font-size:12px;font-weight:700;margin-bottom:4px;"><span style="font-family:var(--mono);font-size:10px;color:var(--accent);margin-right:8px;">${a.article_id}</span>${a.title}</div>
|
| 518 |
+
<div style="font-size:11px;color:var(--text-dim);line-height:1.5;">${a.content?.substring(0,200)}...</div>
|
| 519 |
+
</div>`).join('')}
|
| 520 |
+
</div>`;
|
| 521 |
+
}
|
| 522 |
+
|
| 523 |
+
function renderActionForm(obs) {
|
| 524 |
+
const tid = currentTaskId;
|
| 525 |
+
const actions = obs.valid_actions || [];
|
| 526 |
+
|
| 527 |
+
if (tid === 'ticket_classification') {
|
| 528 |
+
const t = obs.current_ticket;
|
| 529 |
+
return `<div class="action-form"><div class="form-title">⚡ Classify Ticket</div>
|
| 530 |
<div class="form-row">
|
| 531 |
+
<div class="form-group"><label class="form-label">Category</label>
|
|
|
|
| 532 |
<select class="form-select" id="f-category">
|
| 533 |
+
${['billing','technical','account','feature_request','abuse','unknown'].map(c => `<option value="${c}">${c.replace('_',' ').toUpperCase()}</option>`).join('')}
|
| 534 |
+
</select></div>
|
| 535 |
+
<div class="form-group"><label class="form-label">Priority</label>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 536 |
<select class="form-select" id="f-priority">
|
| 537 |
+
${['P1','P2','P3','P4'].map(p => `<option value="${p}">${p}</option>`).join('')}
|
| 538 |
+
</select></div>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 539 |
</div>
|
| 540 |
+
<div class="submit-row"><button class="btn btn-primary" onclick="submitAction({action_type:'classify',ticket_id:'${t?.ticket_id}',category:document.getElementById('f-category').value,priority:document.getElementById('f-priority').value})">Submit →</button></div>
|
| 541 |
</div>`;
|
| 542 |
+
}
|
| 543 |
|
| 544 |
+
if (tid === 'response_drafting') {
|
| 545 |
+
const t = obs.current_ticket;
|
| 546 |
+
return `<div class="action-form"><div class="form-title">✍️ Draft Response</div>
|
| 547 |
+
<div class="form-group" style="margin-bottom:10px;"><label class="form-label">Response Text</label>
|
| 548 |
+
<textarea class="form-textarea" id="f-response" placeholder="Write a professional customer response..."></textarea></div>
|
| 549 |
+
<div class="submit-row"><button class="btn btn-primary" onclick="submitAction({action_type:'draft_response',ticket_id:'${t?.ticket_id}',response_text:document.getElementById('f-response').value})">Send →</button></div>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 550 |
</div>`;
|
| 551 |
+
}
|
|
|
|
|
|
|
|
|
|
|
|
|
| 552 |
|
| 553 |
+
if (tid === 'queue_management') {
|
| 554 |
+
const queue = obs.ticket_queue || [];
|
| 555 |
+
return `<div class="action-form"><div class="form-title">⚙️ Queue Action</div>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 556 |
<div class="form-row">
|
| 557 |
+
<div class="form-group"><label class="form-label">Ticket</label>
|
|
|
|
| 558 |
<select class="form-select" id="f-ticket-id">
|
| 559 |
+
${queue.map(t => `<option value="${t.ticket_id}">${t.ticket_id} | ${t.priority||'?'} | ${t.subject?.substring(0,30)}</option>`).join('')}
|
| 560 |
+
</select></div>
|
| 561 |
+
<div class="form-group"><label class="form-label">Action</label>
|
| 562 |
+
<select class="form-select" id="f-action-type" onchange="toggleAgentRow()">
|
| 563 |
+
${['assign_ticket','resolve','escalate','close','no_op'].map(a => `<option value="${a}">${a.replace('_',' ').toUpperCase()}</option>`).join('')}
|
| 564 |
+
</select></div>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 565 |
</div>
|
| 566 |
+
<div id="agent-row" class="form-group" style="margin-bottom:10px;"><label class="form-label">Assign To</label>
|
|
|
|
| 567 |
<select class="form-select" id="f-agent-id">
|
| 568 |
+
<option value="agent_billing">agent_billing — Billing Specialist</option>
|
| 569 |
+
<option value="agent_tech">agent_tech — Technical Support</option>
|
| 570 |
+
<option value="agent_general">agent_general — General Support</option>
|
| 571 |
+
</select></div>
|
| 572 |
+
<div class="submit-row"><button class="btn btn-primary" onclick="submitQueueAction()">Execute →</button></div>
|
| 573 |
+
</div>`;
|
| 574 |
+
}
|
| 575 |
+
|
| 576 |
+
if (tid === 'multi_turn_conversation') {
|
| 577 |
+
const t = obs.current_ticket;
|
| 578 |
+
return `<div class="action-form"><div class="form-title">💬 Respond or Escalate</div>
|
| 579 |
+
<div class="form-group" style="margin-bottom:10px;"><label class="form-label">Response</label>
|
| 580 |
+
<textarea class="form-textarea" id="f-mt-response" placeholder="Write your response to the customer..."></textarea></div>
|
| 581 |
+
<div class="submit-row" style="gap:8px;">
|
| 582 |
+
<button class="btn btn-secondary" onclick="submitAction({action_type:'escalate',ticket_id:'${t?.ticket_id}'})">↑ Escalate</button>
|
| 583 |
+
<button class="btn btn-secondary" onclick="submitAction({action_type:'resolve',ticket_id:'${t?.ticket_id}'})">✓ Resolve</button>
|
| 584 |
+
<button class="btn btn-primary" onclick="submitAction({action_type:'draft_response',ticket_id:'${t?.ticket_id}',response_text:document.getElementById('f-mt-response').value})">Send →</button>
|
| 585 |
</div>
|
| 586 |
+
</div>`;
|
| 587 |
+
}
|
| 588 |
+
|
| 589 |
+
if (tid === 'legal_clause_identification') {
|
| 590 |
+
return `<div class="action-form"><div class="form-title">⚖️ Identify Clause Type</div>
|
| 591 |
+
<div class="form-group" style="margin-bottom:10px;"><label class="form-label">Clause Type</label>
|
| 592 |
+
<select class="form-select" id="f-clause-type">
|
| 593 |
+
${['indemnity','liability','ip','termination','unknown'].map(c => `<option value="${c}">${c.toUpperCase()}</option>`).join('')}
|
| 594 |
+
</select></div>
|
| 595 |
+
<div class="submit-row"><button class="btn btn-primary" onclick="submitAction({action_type:'identify_clause',clause_type:document.getElementById('f-clause-type').value})">Submit →</button></div>
|
| 596 |
+
</div>`;
|
| 597 |
+
}
|
| 598 |
+
|
| 599 |
+
if (tid === 'legal_risk_flagging') {
|
| 600 |
+
return `<div class="action-form"><div class="form-title">🚩 Flag Risk Level</div>
|
| 601 |
+
<div class="form-row">
|
| 602 |
+
<div class="form-group"><label class="form-label">Risk Level</label>
|
| 603 |
+
<select class="form-select" id="f-risk-level">
|
| 604 |
+
${['low','medium','high','critical'].map(r => `<option value="${r}">${r.toUpperCase()}</option>`).join('')}
|
| 605 |
+
</select></div>
|
| 606 |
+
<div class="form-group"><label class="form-label">Reasoning</label>
|
| 607 |
+
<input class="form-input" id="f-risk-reason" placeholder="Brief justification..."/></div>
|
| 608 |
</div>
|
| 609 |
+
<div class="submit-row"><button class="btn btn-primary" onclick="submitAction({action_type:'flag_risk',risk_level:document.getElementById('f-risk-level').value,reasoning:document.getElementById('f-risk-reason').value})">Submit →</button></div>
|
| 610 |
</div>`;
|
| 611 |
+
}
|
| 612 |
|
| 613 |
+
if (tid === 'legal_clause_redlining') {
|
| 614 |
+
return `<div class="action-form"><div class="form-title">✏️ Redline Clause</div>
|
| 615 |
+
<div class="form-group" style="margin-bottom:10px;"><label class="form-label">Improved Clause Text</label>
|
| 616 |
+
<textarea class="form-textarea" id="f-redline" placeholder="Write the improved, lower-risk version of this clause..."></textarea></div>
|
| 617 |
+
<div class="submit-row"><button class="btn btn-primary" onclick="submitAction({action_type:'redline',redline_text:document.getElementById('f-redline').value})">Submit →</button></div>
|
| 618 |
+
</div>`;
|
| 619 |
}
|
|
|
|
| 620 |
|
| 621 |
+
if (tid === 'clinical_triage_classification') {
|
| 622 |
+
return `<div class="action-form"><div class="form-title">🏥 Classify Body System</div>
|
| 623 |
+
<div class="form-group" style="margin-bottom:10px;"><label class="form-label">Body System</label>
|
| 624 |
+
<select class="form-select" id="f-body-system">
|
| 625 |
+
${['cardiac','respiratory','neurologic','gi','musculoskeletal','other'].map(s => `<option value="${s}">${s.toUpperCase()}</option>`).join('')}
|
| 626 |
+
</select></div>
|
| 627 |
+
<div class="submit-row"><button class="btn btn-primary" onclick="submitAction({action_type:'classify_triage',body_system:document.getElementById('f-body-system').value})">Submit →</button></div>
|
| 628 |
+
</div>`;
|
| 629 |
+
}
|
| 630 |
|
| 631 |
+
if (tid === 'clinical_esi_assignment') {
|
| 632 |
+
return `<div class="action-form"><div class="form-title">🔢 Assign ESI Level</div>
|
| 633 |
+
<div class="form-row">
|
| 634 |
+
<div class="form-group"><label class="form-label">ESI Level (1=Critical, 5=Non-urgent)</label>
|
| 635 |
+
<select class="form-select" id="f-esi">
|
| 636 |
+
${[1,2,3,4,5].map(n => `<option value="${n}">ESI ${n}</option>`).join('')}
|
| 637 |
+
</select></div>
|
| 638 |
+
<div class="form-group"><label class="form-label">Reasoning</label>
|
| 639 |
+
<input class="form-input" id="f-esi-reason" placeholder="Clinical justification..."/></div>
|
| 640 |
+
</div>
|
| 641 |
+
<div class="submit-row"><button class="btn btn-primary" onclick="submitAction({action_type:'assign_esi',esi_level:parseInt(document.getElementById('f-esi').value),reasoning:document.getElementById('f-esi-reason').value})">Submit →</button></div>
|
| 642 |
+
</div>`;
|
| 643 |
+
}
|
| 644 |
|
| 645 |
+
if (tid === 'clinical_triage_note') {
|
| 646 |
+
return `<div class="action-form"><div class="form-title">📋 Write Triage Note</div>
|
| 647 |
+
<div class="form-group" style="margin-bottom:10px;"><label class="form-label">Clinical Triage Note</label>
|
| 648 |
+
<textarea class="form-textarea" id="f-note" placeholder="Patient presents with... Immediate actions: stat IV access, oxygen, continuous monitor..."></textarea></div>
|
| 649 |
+
<div class="submit-row"><button class="btn btn-primary" onclick="submitAction({action_type:'write_triage_note',triage_note:document.getElementById('f-note').value})">Submit →</button></div>
|
| 650 |
+
</div>`;
|
| 651 |
+
}
|
| 652 |
+
|
| 653 |
+
if (tid === 'pr_type_classification') {
|
| 654 |
+
return `<div class="action-form"><div class="form-title">🔍 Classify PR Type</div>
|
| 655 |
+
<div class="form-group" style="margin-bottom:10px;"><label class="form-label">PR Type</label>
|
| 656 |
+
<select class="form-select" id="f-pr-type">
|
| 657 |
+
${['bug_fix','feature','refactor','security'].map(t => `<option value="${t}">${t.replace('_',' ').toUpperCase()}</option>`).join('')}
|
| 658 |
+
</select></div>
|
| 659 |
+
<div class="submit-row"><button class="btn btn-primary" onclick="submitAction({action_type:'classify_pr',pr_type:document.getElementById('f-pr-type').value})">Submit →</button></div>
|
| 660 |
+
</div>`;
|
| 661 |
+
}
|
| 662 |
+
|
| 663 |
+
if (tid === 'pr_bug_identification') {
|
| 664 |
+
return `<div class="action-form"><div class="form-title">🐛 Identify Bug</div>
|
| 665 |
+
<div class="form-group" style="margin-bottom:10px;"><label class="form-label">Bug Description</label>
|
| 666 |
+
<textarea class="form-textarea" id="f-bug" placeholder="Describe the specific vulnerability or bug found in this diff..."></textarea></div>
|
| 667 |
+
<div class="submit-row"><button class="btn btn-primary" onclick="submitAction({action_type:'identify_bug',bug_description:document.getElementById('f-bug').value})">Submit →</button></div>
|
| 668 |
+
</div>`;
|
| 669 |
+
}
|
| 670 |
|
| 671 |
+
if (tid === 'pr_review_comment') {
|
| 672 |
+
return `<div class="action-form"><div class="form-title">💬 Write Review Comment</div>
|
| 673 |
+
<div class="form-group" style="margin-bottom:10px;"><label class="form-label">Review Comment</label>
|
| 674 |
+
<textarea class="form-textarea" id="f-review" placeholder="Approve or block this PR with your code review comment..."></textarea></div>
|
| 675 |
+
<div class="submit-row"><button class="btn btn-primary" onclick="submitAction({action_type:'review_pr',review_comment:document.getElementById('f-review').value})">Submit →</button></div>
|
| 676 |
+
</div>`;
|
| 677 |
+
}
|
| 678 |
|
| 679 |
+
return '';
|
|
|
|
|
|
|
|
|
|
|
|
|
| 680 |
}
|
| 681 |
|
| 682 |
+
function toggleAgentRow() {
|
| 683 |
+
const action = document.getElementById('f-action-type')?.value;
|
| 684 |
+
const row = document.getElementById('agent-row');
|
| 685 |
+
if (row) row.style.display = action === 'assign_ticket' ? 'flex' : 'none';
|
| 686 |
+
}
|
| 687 |
+
|
| 688 |
+
function submitQueueAction() {
|
| 689 |
const action_type = document.getElementById('f-action-type').value;
|
| 690 |
const ticket_id = document.getElementById('f-ticket-id')?.value;
|
| 691 |
const target_agent_id = document.getElementById('f-agent-id')?.value;
|
| 692 |
+
submitAction({ action_type, ticket_id, target_agent_id });
|
|
|
|
| 693 |
}
|
| 694 |
|
| 695 |
async function submitAction(actionBody) {
|
|
|
|
| 697 |
const data = await api('POST', `/step?task_id=${currentTaskId}`, actionBody);
|
| 698 |
currentObs = data.observation;
|
| 699 |
const reward = data.reward;
|
|
|
|
| 700 |
rewardHistory.push(reward.total);
|
| 701 |
document.getElementById('step-counter').textContent = currentObs.step || rewardHistory.length;
|
|
|
|
| 702 |
const sign = reward.total >= 0 ? '+' : '';
|
| 703 |
+
addLog(`Step ${rewardHistory.length}: ${sign}${reward.total.toFixed(3)}`, reward.total >= 0 ? 'pos' : 'neg');
|
|
|
|
| 704 |
updateRewardChart();
|
| 705 |
updateScoreDisplay(reward);
|
| 706 |
renderObservation(currentObs);
|
| 707 |
+
if (data.done) fetchGrader();
|
| 708 |
+
} catch(e) { addLog(`❌ ${e.message}`, 'neg'); }
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 709 |
}
|
| 710 |
|
| 711 |
+
async function fetchGrader() {
|
| 712 |
try {
|
| 713 |
const result = await api('POST', `/grader?task_id=${currentTaskId}`);
|
| 714 |
+
showDone(result);
|
| 715 |
+
document.getElementById(`score-${currentTaskId}`).textContent = result.final_score.toFixed(3);
|
| 716 |
+
updateOverallScore();
|
| 717 |
+
} catch(e) { console.error(e); }
|
|
|
|
|
|
|
| 718 |
}
|
| 719 |
|
| 720 |
+
function showDone(result) {
|
|
|
|
| 721 |
const score = result.final_score;
|
| 722 |
const color = score >= 0.7 ? 'var(--accent3)' : score >= 0.4 ? 'var(--accent4)' : 'var(--accent2)';
|
| 723 |
const emoji = score >= 0.7 ? '🏆' : score >= 0.4 ? '📊' : '📉';
|
|
|
|
| 724 |
document.getElementById('center-body').innerHTML = `
|
| 725 |
<div class="episode-done">
|
| 726 |
+
<div style="font-size:36px;margin-bottom:12px;">${emoji}</div>
|
| 727 |
<div class="done-title">Episode Complete</div>
|
| 728 |
<div class="done-score" style="color:${color}">${score.toFixed(4)}</div>
|
| 729 |
+
<div style="font-family:var(--mono);font-size:12px;color:var(--text-dim);margin:8px 0 20px">
|
| 730 |
+
${result.passed ? '✓ PASSED' : '✗ DID NOT PASS'} | ${rewardHistory.length} steps
|
| 731 |
</div>
|
| 732 |
+
${Object.entries(result.metrics||{}).filter(([k,v])=>typeof v==='number').map(([k,v])=>`
|
| 733 |
+
<div class="breakdown-row">
|
| 734 |
+
<span class="breakdown-label">${k.replace(/_/g,' ')}</span>
|
| 735 |
+
<span class="breakdown-val ${v>=0.7?'val-good':v>=0.4?'val-mid':'val-bad'}">${v.toFixed(4)}</span>
|
| 736 |
+
</div>`).join('')}
|
| 737 |
+
<button class="btn btn-primary" style="margin-top:20px;" onclick="resetEpisode()">↺ New Episode</button>
|
|
|
|
|
|
|
| 738 |
</div>`;
|
|
|
|
| 739 |
document.getElementById('current-score').textContent = score.toFixed(4);
|
| 740 |
document.getElementById('current-score').style.color = color;
|
| 741 |
+
document.getElementById('score-status').textContent = result.passed ? '✓ Passed' : '✗ Did not pass';
|
| 742 |
}
|
| 743 |
|
| 744 |
function updateScoreDisplay(reward) {
|
|
|
|
| 746 |
const color = total >= 0.6 ? 'var(--accent3)' : total >= 0.2 ? 'var(--accent4)' : 'var(--accent2)';
|
| 747 |
document.getElementById('current-score').textContent = total.toFixed(4);
|
| 748 |
document.getElementById('current-score').style.color = color;
|
| 749 |
+
document.getElementById('score-status').textContent = 'Step reward';
|
|
|
|
| 750 |
const bd = reward.breakdown || {};
|
| 751 |
document.getElementById('score-breakdown').innerHTML = Object.entries(bd).map(([k,v]) => `
|
| 752 |
<div class="breakdown-row">
|
| 753 |
<span class="breakdown-label">${k.replace(/_/g,' ')}</span>
|
| 754 |
+
<span class="breakdown-val ${v>=0.6?'val-good':v<0?'val-bad':'val-mid'}">${typeof v==='number'?v.toFixed(3):v}</span>
|
| 755 |
</div>`).join('');
|
| 756 |
}
|
| 757 |
|
| 758 |
function updateRewardChart() {
|
| 759 |
const container = document.getElementById('reward-chart');
|
| 760 |
+
if (!rewardHistory.length) { container.innerHTML=''; return; }
|
| 761 |
+
const avg = (rewardHistory.reduce((a,b)=>a+b,0)/rewardHistory.length).toFixed(3);
|
|
|
|
| 762 |
document.getElementById('avg-reward').textContent = avg;
|
|
|
|
| 763 |
container.innerHTML = rewardHistory.map(r => {
|
| 764 |
+
const pct = ((r+1)/2*100).toFixed(1);
|
| 765 |
+
const color = r>=0.3?'var(--accent3)':r>=0?'var(--accent4)':'var(--accent2)';
|
| 766 |
return `<div class="bar" style="height:${pct}%;background:${color};opacity:0.8;"></div>`;
|
| 767 |
}).join('');
|
| 768 |
}
|
| 769 |
|
| 770 |
function updateAgents(agents) {
|
| 771 |
+
if (!agents?.length) { document.getElementById('agents-section').style.display='none'; return; }
|
| 772 |
+
document.getElementById('agents-section').style.display='block';
|
|
|
|
|
|
|
|
|
|
| 773 |
document.getElementById('agents-container').innerHTML = agents.map(a => {
|
| 774 |
+
const pct = (a.current_load/a.max_load*100).toFixed(0);
|
| 775 |
+
return `<div class="agent-card">
|
| 776 |
+
<div class="agent-name">${a.name}</div>
|
| 777 |
+
<div style="font-family:var(--mono);font-size:9px;color:var(--text-dim);">Load: ${a.current_load}/${a.max_load}</div>
|
| 778 |
+
<div class="load-bar-track"><div class="load-bar-fill ${a.current_load>=a.max_load?'over':''}" style="width:${pct}%"></div></div>
|
| 779 |
+
</div>`;
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 780 |
}).join('');
|
| 781 |
}
|
| 782 |
|
| 783 |
function updateQueue(queue, slaStatus) {
|
| 784 |
+
if (!queue?.length) { document.getElementById('queue-section').style.display='none'; return; }
|
| 785 |
+
document.getElementById('queue-section').style.display='block';
|
|
|
|
|
|
|
|
|
|
| 786 |
document.getElementById('queue-container').innerHTML = queue.slice(0,8).map(t => {
|
| 787 |
+
const p = (t.priority||'P3').replace('_CRITICAL','').replace('_HIGH','').replace('_MEDIUM','').replace('_LOW','').toLowerCase();
|
| 788 |
+
return `<div class="queue-item">
|
| 789 |
+
<div class="priority-dot ${p}"></div>
|
| 790 |
+
<span class="queue-subject">${t.subject||t.ticket_id}</span>
|
| 791 |
+
</div>`;
|
|
|
|
|
|
|
|
|
|
| 792 |
}).join('');
|
| 793 |
}
|
| 794 |
|
| 795 |
function addLog(message, type) {
|
| 796 |
const container = document.getElementById('log-container');
|
| 797 |
const div = document.createElement('div');
|
| 798 |
+
div.className = `log-entry ${type==='pos'?'log-pos':type==='neg'?'log-neg':''}`;
|
| 799 |
div.textContent = message;
|
| 800 |
container.insertBefore(div, container.firstChild);
|
| 801 |
+
if (container.children.length > 40) container.removeChild(container.lastChild);
|
| 802 |
}
|
| 803 |
|
| 804 |
async function runBaseline() {
|
| 805 |
+
addLog('⚡ Running heuristic baseline...');
|
| 806 |
try {
|
| 807 |
const data = await api('POST', '/baseline');
|
| 808 |
+
Object.entries(data.tasks||{}).forEach(([tid, result]) => {
|
| 809 |
+
const el = document.getElementById(`score-${tid}`);
|
| 810 |
+
if (el) el.textContent = result.final_score.toFixed(3);
|
| 811 |
+
addLog(`${tid.replace(/_/g,' ')}: ${result.final_score.toFixed(3)} ${result.passed?'✓':'✗'}`,
|
| 812 |
+
result.final_score >= 0.5 ? 'pos' : 'neg');
|
|
|
|
|
|
|
| 813 |
});
|
| 814 |
+
document.getElementById('overall-score').textContent = data.overall_score.toFixed(3);
|
| 815 |
+
addLog(`Overall: ${data.overall_score.toFixed(3)}`, data.overall_score > 0.6 ? 'pos' : 'neg');
|
| 816 |
+
} catch(e) { addLog(`Baseline error: ${e.message}`, 'neg'); }
|
| 817 |
+
}
|
| 818 |
+
|
| 819 |
+
function updateOverallScore() {
|
| 820 |
+
const scores = Object.keys(TASK_CONFIG).map(tid => {
|
| 821 |
+
const el = document.getElementById(`score-${tid}`);
|
| 822 |
+
return el && el.textContent !== '—' ? parseFloat(el.textContent) : null;
|
| 823 |
+
}).filter(s => s !== null);
|
| 824 |
+
if (scores.length > 0) {
|
| 825 |
+
const avg = scores.reduce((a,b)=>a+b,0)/scores.length;
|
| 826 |
+
document.getElementById('overall-score').textContent = avg.toFixed(3);
|
| 827 |
}
|
| 828 |
}
|
| 829 |
|
| 830 |
+
// Health check on load
|
| 831 |
(async () => {
|
| 832 |
try {
|
| 833 |
+
const h = await api('GET', '/health');
|
| 834 |
document.getElementById('server-status').textContent = 'LIVE';
|
| 835 |
+
addLog(`✓ Server online — ${h.tasks?.length||0} tasks`);
|
| 836 |
} catch(e) {
|
| 837 |
document.getElementById('server-status').textContent = 'OFFLINE';
|
| 838 |
document.getElementById('server-status').style.color = 'var(--accent2)';
|
| 839 |
+
document.querySelector('.status-pill').style.color = 'var(--accent2)';
|
| 840 |
+
document.querySelector('.status-pill').style.borderColor = 'rgba(255,101,132,0.3)';
|
| 841 |
+
document.querySelector('.pulse').style.background = 'var(--accent2)';
|
| 842 |
}
|
| 843 |
})();
|
| 844 |
</script>
|
| 845 |
</body>
|
| 846 |
+
</html>
|