@pasko70/pibo 1.4.0 → 1.4.2
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/README.md +183 -183
- package/context/codex-base-prompt.md +148 -148
- package/context/compute-worker.md +23 -23
- package/context/pibo-compaction-prompt.md +100 -100
- package/context/pibo-native-tooling.md +18 -18
- package/context/pibo-system-prompt.md +77 -77
- package/dist/apps/chat/agent-store.js +82 -82
- package/dist/apps/chat/data/project-service.js +166 -166
- package/dist/apps/chat/data/read-state-service.js +18 -18
- package/dist/apps/chat/data/timeline-query-service.js +8 -8
- package/dist/apps/chat/model-catalog.js +5 -1
- package/dist/apps/chat/static-assets.js +853 -853
- package/dist/apps/chat/workflow-persistence.js +255 -255
- package/dist/apps/chat-ui/assets/{dist-7YaJd19a.js → dist-B2BEpL7n.js} +1 -1
- package/dist/apps/chat-ui/assets/{dist-CiYeO8nN.js → dist-BAGS_xkV.js} +1 -1
- package/dist/apps/chat-ui/assets/{dist-BIvPnn_C.js → dist-BAXNalar.js} +1 -1
- package/dist/apps/chat-ui/assets/{dist-CjSI6y5z.js → dist-BwUvs6Ph.js} +1 -1
- package/dist/apps/chat-ui/assets/{dist-Ubstha8t.js → dist-C0zsJ8II.js} +1 -1
- package/dist/apps/chat-ui/assets/{dist-KXCMNKIL.js → dist-C3PnEkhb.js} +1 -1
- package/dist/apps/chat-ui/assets/{dist-CVQU42Fn.js → dist-CYPL-B2Z.js} +1 -1
- package/dist/apps/chat-ui/assets/{dist-BbNE72h8.js → dist-CiDSXgtg.js} +1 -1
- package/dist/apps/chat-ui/assets/{dist-D7TCkoFT.js → dist-DnACFKyO.js} +1 -1
- package/dist/apps/chat-ui/assets/{dist-ChSZNqKE.js → dist-ZB1-ui2y.js} +1 -1
- package/dist/apps/chat-ui/assets/{dist-8Noo5eCN.js → dist-vKlxFkTa.js} +1 -1
- package/dist/apps/chat-ui/assets/{index-CmxtUVG1.js → index-0x7tuTNX.js} +3 -3
- package/dist/apps/chat-ui/assets/{index-C25VYnyb.css → index-B-qaya1G.css} +1 -1
- package/dist/apps/chat-ui/index.html +18 -18
- package/dist/apps/chat-ui/manifest.webmanifest +25 -25
- package/dist/apps/chat-ui/sw.js +41 -41
- package/dist/apps/chat-vscode-web/index.html +12 -12
- package/dist/apps/cli-ui/cliSessionsCommand.js +23 -23
- package/dist/apps/context-files-ui/index.html +11 -11
- package/dist/apps/vscode-artifacts/latest.vsix +0 -0
- package/dist/apps/vscode-artifacts/pibo-vscode-ext-1.4.2.vsix +0 -0
- package/dist/bin/pibo.js +0 -0
- package/dist/bin/rg.js +0 -0
- package/dist/cli.js +35 -35
- package/dist/compute/cli.js +54 -54
- package/dist/core/runtime.js +2 -0
- package/dist/core/session-router.js +1 -0
- package/dist/cron/cli.js +15 -15
- package/dist/cron/store.js +49 -49
- package/dist/data/cli.js +23 -23
- package/dist/data/event-log.js +23 -23
- package/dist/data/message-store.js +20 -20
- package/dist/data/navigation-store.js +9 -9
- package/dist/data/observation-store.js +4 -4
- package/dist/data/payload-store.js +17 -17
- package/dist/data/schema.js +429 -429
- package/dist/data/session-store.js +4 -4
- package/dist/data/telemetry-queries.js +54 -54
- package/dist/data/telemetry.js +156 -156
- package/dist/debug/events.js +12 -12
- package/dist/debug/failures.js +6 -6
- package/dist/debug/index.js +207 -207
- package/dist/debug/messages.js +6 -6
- package/dist/debug/pty.js +124 -124
- package/dist/debug/session.js +29 -29
- package/dist/debug/tools.js +5 -5
- package/dist/debug/web-snapshot-browser-scripts.js +294 -294
- package/dist/debug/web-streaming-browser-library.js +925 -925
- package/dist/debug/web-streaming-browser-scripts.js +232 -232
- package/dist/debug/web-streaming-provider-telemetry.js +4 -4
- package/dist/debug/web.js +93 -93
- package/dist/gateway/cli.js +19 -19
- package/dist/mcp/config-command.js +53 -53
- package/dist/mcp/index.js +21 -21
- package/dist/mcp/registry.js +11 -11
- package/dist/pi-packages/cli.js +11 -11
- package/dist/plugins/context-files-store.js +110 -110
- package/dist/plugins/context-files.js +4 -4
- package/dist/providers/glm.js +59 -0
- package/dist/ralph/cli.js +18 -18
- package/dist/ralph/templates.js +140 -140
- package/dist/reliability/store.js +226 -226
- package/dist/sessions/pibo-data-store.js +16 -16
- package/dist/sessions/sqlite-store.js +53 -53
- package/dist/setup/cli.js +58 -58
- package/dist/tools/agent-browser-wrapper.js +80 -80
- package/dist/tools/browser-use-cdp.js +12 -12
- package/dist/tools/browser-use-wrapper.js +762 -762
- package/dist/tools/guides.js +455 -455
- package/dist/tools/index.js +99 -99
- package/dist/tools/runtime/node-worker-source.js +205 -205
- package/dist/tools/runtime/python-worker-source.js +177 -177
- package/dist/vscode/cli.js +9 -9
- package/dist/web-annotations/cdp.js +900 -900
- package/dist/web-annotations/store.js +96 -96
- package/docs/README.md +23 -23
- package/docs/ops/install-developer-host.md +112 -112
- package/docs/ops/install-user-host.md +96 -96
- package/docs/ops/upgrade-user-to-developer-host.md +69 -69
- package/docs/ops/vscode-extension-release.md +160 -160
- package/package.json +1 -1
- package/skills/builtin/pi-agent-harness/SKILL.md +319 -319
- package/skills/builtin/pi-agent-harness/agents/openai.yaml +4 -4
- package/skills/builtin/pibo-docker-system/SKILL.md +170 -170
- package/skills/builtin/pibo-spec-writing/SKILL.md +330 -330
- package/skills/builtin/prd/SKILL.md +143 -143
- package/skills/builtin/ralph-loop/SKILL.md +359 -359
- package/skills/builtin/ralph-prd-json/SKILL.md +123 -123
- package/skills/builtin/skill-creator/LICENSE.txt +201 -201
- package/skills/builtin/skill-creator/SKILL.md +513 -513
- package/skills/builtin/skill-creator/agents/analyzer.md +274 -274
- package/skills/builtin/skill-creator/agents/comparator.md +202 -202
- package/skills/builtin/skill-creator/agents/grader.md +223 -223
- package/skills/builtin/skill-creator/assets/eval_review.html +146 -146
- package/skills/builtin/skill-creator/eval-viewer/generate_review.py +471 -471
- package/skills/builtin/skill-creator/eval-viewer/viewer.html +1325 -1325
- package/skills/builtin/skill-creator/references/schemas.md +430 -430
- package/skills/builtin/skill-creator/scripts/aggregate_benchmark.py +401 -401
- package/skills/builtin/skill-creator/scripts/generate_report.py +326 -326
- package/skills/builtin/skill-creator/scripts/improve_description.py +247 -247
- package/skills/builtin/skill-creator/scripts/package_skill.py +136 -136
- package/skills/builtin/skill-creator/scripts/quick_validate.py +102 -102
- package/skills/builtin/skill-creator/scripts/run_eval.py +310 -310
- package/skills/builtin/skill-creator/scripts/run_loop.py +328 -328
- package/skills/builtin/skill-creator/scripts/utils.py +47 -47
- package/skills/builtin/web-annotations/SKILL.md +93 -93
- package/src/mcp/LICENSE.mcp-cli +21 -21
- package/dist/apps/vscode-artifacts/pibo-vscode-1.3.0.vsix +0 -0
- package/dist/apps/vscode-artifacts/pibo-vscode-ext-1.3.3.vsix +0 -0
- package/dist/apps/vscode-artifacts/pibo-vscode-ext-1.3.4.vsix +0 -0
- package/dist/apps/vscode-artifacts/pibo-vscode-ext-1.3.5.vsix +0 -0
- package/dist/core/shared-app.js +0 -17
- package/dist/data/final-app-space-cutover-migration.js +0 -728
- package/dist/data/shared-app-migration.js +0 -757
- package/dist/session-ui/ownerViewModel.js +0 -27
- package/dist/shared-app.js +0 -4
|
@@ -1,146 +1,146 @@
|
|
|
1
|
-
<!DOCTYPE html>
|
|
2
|
-
<html lang="en">
|
|
3
|
-
<head>
|
|
4
|
-
<meta charset="UTF-8">
|
|
5
|
-
<meta name="viewport" content="width=device-width, initial-scale=1.0">
|
|
6
|
-
<title>Eval Set Review - __SKILL_NAME_PLACEHOLDER__</title>
|
|
7
|
-
<link rel="preconnect" href="https://fonts.googleapis.com">
|
|
8
|
-
<link rel="preconnect" href="https://fonts.gstatic.com" crossorigin>
|
|
9
|
-
<link href="https://fonts.googleapis.com/css2?family=Poppins:wght@500;600&family=Lora:wght@400;500&display=swap" rel="stylesheet">
|
|
10
|
-
<style>
|
|
11
|
-
* { box-sizing: border-box; margin: 0; padding: 0; }
|
|
12
|
-
body { font-family: 'Lora', Georgia, serif; background: #faf9f5; padding: 2rem; color: #141413; }
|
|
13
|
-
h1 { font-family: 'Poppins', sans-serif; margin-bottom: 0.5rem; font-size: 1.5rem; }
|
|
14
|
-
.description { color: #b0aea5; margin-bottom: 1.5rem; font-style: italic; max-width: 900px; }
|
|
15
|
-
.controls { margin-bottom: 1rem; display: flex; gap: 0.5rem; }
|
|
16
|
-
.btn { font-family: 'Poppins', sans-serif; padding: 0.5rem 1rem; border: none; border-radius: 6px; cursor: pointer; font-size: 0.875rem; font-weight: 500; }
|
|
17
|
-
.btn-add { background: #6a9bcc; color: white; }
|
|
18
|
-
.btn-add:hover { background: #5889b8; }
|
|
19
|
-
.btn-export { background: #d97757; color: white; }
|
|
20
|
-
.btn-export:hover { background: #c4613f; }
|
|
21
|
-
table { width: 100%; max-width: 1100px; border-collapse: collapse; background: white; border-radius: 6px; overflow: hidden; box-shadow: 0 1px 3px rgba(0,0,0,0.08); }
|
|
22
|
-
th { font-family: 'Poppins', sans-serif; background: #141413; color: #faf9f5; padding: 0.75rem 1rem; text-align: left; font-size: 0.875rem; }
|
|
23
|
-
td { padding: 0.75rem 1rem; border-bottom: 1px solid #e8e6dc; vertical-align: top; }
|
|
24
|
-
tr:nth-child(even) td { background: #faf9f5; }
|
|
25
|
-
tr:hover td { background: #f3f1ea; }
|
|
26
|
-
.section-header td { background: #e8e6dc; font-family: 'Poppins', sans-serif; font-weight: 500; font-size: 0.8rem; color: #141413; text-transform: uppercase; letter-spacing: 0.05em; }
|
|
27
|
-
.query-input { width: 100%; padding: 0.4rem; border: 1px solid #e8e6dc; border-radius: 4px; font-size: 0.875rem; font-family: 'Lora', Georgia, serif; resize: vertical; min-height: 60px; }
|
|
28
|
-
.query-input:focus { outline: none; border-color: #d97757; box-shadow: 0 0 0 2px rgba(217,119,87,0.15); }
|
|
29
|
-
.toggle { position: relative; display: inline-block; width: 44px; height: 24px; }
|
|
30
|
-
.toggle input { opacity: 0; width: 0; height: 0; }
|
|
31
|
-
.toggle .slider { position: absolute; inset: 0; background: #b0aea5; border-radius: 24px; cursor: pointer; transition: 0.2s; }
|
|
32
|
-
.toggle .slider::before { content: ""; position: absolute; width: 18px; height: 18px; left: 3px; bottom: 3px; background: white; border-radius: 50%; transition: 0.2s; }
|
|
33
|
-
.toggle input:checked + .slider { background: #d97757; }
|
|
34
|
-
.toggle input:checked + .slider::before { transform: translateX(20px); }
|
|
35
|
-
.btn-delete { background: #c44; color: white; padding: 0.3rem 0.6rem; border: none; border-radius: 4px; cursor: pointer; font-size: 0.75rem; font-family: 'Poppins', sans-serif; }
|
|
36
|
-
.btn-delete:hover { background: #a33; }
|
|
37
|
-
.summary { margin-top: 1rem; color: #b0aea5; font-size: 0.875rem; }
|
|
38
|
-
</style>
|
|
39
|
-
</head>
|
|
40
|
-
<body>
|
|
41
|
-
<h1>Eval Set Review: <span id="skill-name">__SKILL_NAME_PLACEHOLDER__</span></h1>
|
|
42
|
-
<p class="description">Current description: <span id="skill-desc">__SKILL_DESCRIPTION_PLACEHOLDER__</span></p>
|
|
43
|
-
|
|
44
|
-
<div class="controls">
|
|
45
|
-
<button class="btn btn-add" onclick="addRow()">+ Add Query</button>
|
|
46
|
-
<button class="btn btn-export" onclick="exportEvalSet()">Export Eval Set</button>
|
|
47
|
-
</div>
|
|
48
|
-
|
|
49
|
-
<table>
|
|
50
|
-
<thead>
|
|
51
|
-
<tr>
|
|
52
|
-
<th style="width:65%">Query</th>
|
|
53
|
-
<th style="width:18%">Should Trigger</th>
|
|
54
|
-
<th style="width:10%">Actions</th>
|
|
55
|
-
</tr>
|
|
56
|
-
</thead>
|
|
57
|
-
<tbody id="eval-body"></tbody>
|
|
58
|
-
</table>
|
|
59
|
-
|
|
60
|
-
<p class="summary" id="summary"></p>
|
|
61
|
-
|
|
62
|
-
<script>
|
|
63
|
-
const EVAL_DATA = __EVAL_DATA_PLACEHOLDER__;
|
|
64
|
-
|
|
65
|
-
let evalItems = [...EVAL_DATA];
|
|
66
|
-
|
|
67
|
-
function render() {
|
|
68
|
-
const tbody = document.getElementById('eval-body');
|
|
69
|
-
tbody.innerHTML = '';
|
|
70
|
-
|
|
71
|
-
// Sort: should-trigger first, then should-not-trigger
|
|
72
|
-
const sorted = evalItems
|
|
73
|
-
.map((item, origIdx) => ({ ...item, origIdx }))
|
|
74
|
-
.sort((a, b) => (b.should_trigger ? 1 : 0) - (a.should_trigger ? 1 : 0));
|
|
75
|
-
|
|
76
|
-
let lastGroup = null;
|
|
77
|
-
sorted.forEach(item => {
|
|
78
|
-
const group = item.should_trigger ? 'trigger' : 'no-trigger';
|
|
79
|
-
if (group !== lastGroup) {
|
|
80
|
-
const headerRow = document.createElement('tr');
|
|
81
|
-
headerRow.className = 'section-header';
|
|
82
|
-
headerRow.innerHTML = `<td colspan="3">${item.should_trigger ? 'Should Trigger' : 'Should NOT Trigger'}</td>`;
|
|
83
|
-
tbody.appendChild(headerRow);
|
|
84
|
-
lastGroup = group;
|
|
85
|
-
}
|
|
86
|
-
|
|
87
|
-
const idx = item.origIdx;
|
|
88
|
-
const tr = document.createElement('tr');
|
|
89
|
-
tr.innerHTML = `
|
|
90
|
-
<td><textarea class="query-input" onchange="updateQuery(${idx}, this.value)">${escapeHtml(item.query)}</textarea></td>
|
|
91
|
-
<td>
|
|
92
|
-
<label class="toggle">
|
|
93
|
-
<input type="checkbox" ${item.should_trigger ? 'checked' : ''} onchange="updateTrigger(${idx}, this.checked)">
|
|
94
|
-
<span class="slider"></span>
|
|
95
|
-
</label>
|
|
96
|
-
<span style="margin-left:8px;font-size:0.8rem;color:#b0aea5">${item.should_trigger ? 'Yes' : 'No'}</span>
|
|
97
|
-
</td>
|
|
98
|
-
<td><button class="btn-delete" onclick="deleteRow(${idx})">Delete</button></td>
|
|
99
|
-
`;
|
|
100
|
-
tbody.appendChild(tr);
|
|
101
|
-
});
|
|
102
|
-
updateSummary();
|
|
103
|
-
}
|
|
104
|
-
|
|
105
|
-
function escapeHtml(text) {
|
|
106
|
-
const div = document.createElement('div');
|
|
107
|
-
div.textContent = text;
|
|
108
|
-
return div.innerHTML;
|
|
109
|
-
}
|
|
110
|
-
|
|
111
|
-
function updateQuery(idx, value) { evalItems[idx].query = value; updateSummary(); }
|
|
112
|
-
function updateTrigger(idx, value) { evalItems[idx].should_trigger = value; render(); }
|
|
113
|
-
function deleteRow(idx) { evalItems.splice(idx, 1); render(); }
|
|
114
|
-
|
|
115
|
-
function addRow() {
|
|
116
|
-
evalItems.push({ query: '', should_trigger: true });
|
|
117
|
-
render();
|
|
118
|
-
const inputs = document.querySelectorAll('.query-input');
|
|
119
|
-
inputs[inputs.length - 1].focus();
|
|
120
|
-
}
|
|
121
|
-
|
|
122
|
-
function updateSummary() {
|
|
123
|
-
const trigger = evalItems.filter(i => i.should_trigger).length;
|
|
124
|
-
const noTrigger = evalItems.filter(i => !i.should_trigger).length;
|
|
125
|
-
document.getElementById('summary').textContent =
|
|
126
|
-
`${evalItems.length} queries total: ${trigger} should trigger, ${noTrigger} should not trigger`;
|
|
127
|
-
}
|
|
128
|
-
|
|
129
|
-
function exportEvalSet() {
|
|
130
|
-
const valid = evalItems.filter(i => i.query.trim() !== '');
|
|
131
|
-
const data = valid.map(i => ({ query: i.query.trim(), should_trigger: i.should_trigger }));
|
|
132
|
-
const blob = new Blob([JSON.stringify(data, null, 2)], { type: 'application/json' });
|
|
133
|
-
const url = URL.createObjectURL(blob);
|
|
134
|
-
const a = document.createElement('a');
|
|
135
|
-
a.href = url;
|
|
136
|
-
a.download = 'eval_set.json';
|
|
137
|
-
document.body.appendChild(a);
|
|
138
|
-
a.click();
|
|
139
|
-
document.body.removeChild(a);
|
|
140
|
-
URL.revokeObjectURL(url);
|
|
141
|
-
}
|
|
142
|
-
|
|
143
|
-
render();
|
|
144
|
-
</script>
|
|
145
|
-
</body>
|
|
146
|
-
</html>
|
|
1
|
+
<!DOCTYPE html>
|
|
2
|
+
<html lang="en">
|
|
3
|
+
<head>
|
|
4
|
+
<meta charset="UTF-8">
|
|
5
|
+
<meta name="viewport" content="width=device-width, initial-scale=1.0">
|
|
6
|
+
<title>Eval Set Review - __SKILL_NAME_PLACEHOLDER__</title>
|
|
7
|
+
<link rel="preconnect" href="https://fonts.googleapis.com">
|
|
8
|
+
<link rel="preconnect" href="https://fonts.gstatic.com" crossorigin>
|
|
9
|
+
<link href="https://fonts.googleapis.com/css2?family=Poppins:wght@500;600&family=Lora:wght@400;500&display=swap" rel="stylesheet">
|
|
10
|
+
<style>
|
|
11
|
+
* { box-sizing: border-box; margin: 0; padding: 0; }
|
|
12
|
+
body { font-family: 'Lora', Georgia, serif; background: #faf9f5; padding: 2rem; color: #141413; }
|
|
13
|
+
h1 { font-family: 'Poppins', sans-serif; margin-bottom: 0.5rem; font-size: 1.5rem; }
|
|
14
|
+
.description { color: #b0aea5; margin-bottom: 1.5rem; font-style: italic; max-width: 900px; }
|
|
15
|
+
.controls { margin-bottom: 1rem; display: flex; gap: 0.5rem; }
|
|
16
|
+
.btn { font-family: 'Poppins', sans-serif; padding: 0.5rem 1rem; border: none; border-radius: 6px; cursor: pointer; font-size: 0.875rem; font-weight: 500; }
|
|
17
|
+
.btn-add { background: #6a9bcc; color: white; }
|
|
18
|
+
.btn-add:hover { background: #5889b8; }
|
|
19
|
+
.btn-export { background: #d97757; color: white; }
|
|
20
|
+
.btn-export:hover { background: #c4613f; }
|
|
21
|
+
table { width: 100%; max-width: 1100px; border-collapse: collapse; background: white; border-radius: 6px; overflow: hidden; box-shadow: 0 1px 3px rgba(0,0,0,0.08); }
|
|
22
|
+
th { font-family: 'Poppins', sans-serif; background: #141413; color: #faf9f5; padding: 0.75rem 1rem; text-align: left; font-size: 0.875rem; }
|
|
23
|
+
td { padding: 0.75rem 1rem; border-bottom: 1px solid #e8e6dc; vertical-align: top; }
|
|
24
|
+
tr:nth-child(even) td { background: #faf9f5; }
|
|
25
|
+
tr:hover td { background: #f3f1ea; }
|
|
26
|
+
.section-header td { background: #e8e6dc; font-family: 'Poppins', sans-serif; font-weight: 500; font-size: 0.8rem; color: #141413; text-transform: uppercase; letter-spacing: 0.05em; }
|
|
27
|
+
.query-input { width: 100%; padding: 0.4rem; border: 1px solid #e8e6dc; border-radius: 4px; font-size: 0.875rem; font-family: 'Lora', Georgia, serif; resize: vertical; min-height: 60px; }
|
|
28
|
+
.query-input:focus { outline: none; border-color: #d97757; box-shadow: 0 0 0 2px rgba(217,119,87,0.15); }
|
|
29
|
+
.toggle { position: relative; display: inline-block; width: 44px; height: 24px; }
|
|
30
|
+
.toggle input { opacity: 0; width: 0; height: 0; }
|
|
31
|
+
.toggle .slider { position: absolute; inset: 0; background: #b0aea5; border-radius: 24px; cursor: pointer; transition: 0.2s; }
|
|
32
|
+
.toggle .slider::before { content: ""; position: absolute; width: 18px; height: 18px; left: 3px; bottom: 3px; background: white; border-radius: 50%; transition: 0.2s; }
|
|
33
|
+
.toggle input:checked + .slider { background: #d97757; }
|
|
34
|
+
.toggle input:checked + .slider::before { transform: translateX(20px); }
|
|
35
|
+
.btn-delete { background: #c44; color: white; padding: 0.3rem 0.6rem; border: none; border-radius: 4px; cursor: pointer; font-size: 0.75rem; font-family: 'Poppins', sans-serif; }
|
|
36
|
+
.btn-delete:hover { background: #a33; }
|
|
37
|
+
.summary { margin-top: 1rem; color: #b0aea5; font-size: 0.875rem; }
|
|
38
|
+
</style>
|
|
39
|
+
</head>
|
|
40
|
+
<body>
|
|
41
|
+
<h1>Eval Set Review: <span id="skill-name">__SKILL_NAME_PLACEHOLDER__</span></h1>
|
|
42
|
+
<p class="description">Current description: <span id="skill-desc">__SKILL_DESCRIPTION_PLACEHOLDER__</span></p>
|
|
43
|
+
|
|
44
|
+
<div class="controls">
|
|
45
|
+
<button class="btn btn-add" onclick="addRow()">+ Add Query</button>
|
|
46
|
+
<button class="btn btn-export" onclick="exportEvalSet()">Export Eval Set</button>
|
|
47
|
+
</div>
|
|
48
|
+
|
|
49
|
+
<table>
|
|
50
|
+
<thead>
|
|
51
|
+
<tr>
|
|
52
|
+
<th style="width:65%">Query</th>
|
|
53
|
+
<th style="width:18%">Should Trigger</th>
|
|
54
|
+
<th style="width:10%">Actions</th>
|
|
55
|
+
</tr>
|
|
56
|
+
</thead>
|
|
57
|
+
<tbody id="eval-body"></tbody>
|
|
58
|
+
</table>
|
|
59
|
+
|
|
60
|
+
<p class="summary" id="summary"></p>
|
|
61
|
+
|
|
62
|
+
<script>
|
|
63
|
+
const EVAL_DATA = __EVAL_DATA_PLACEHOLDER__;
|
|
64
|
+
|
|
65
|
+
let evalItems = [...EVAL_DATA];
|
|
66
|
+
|
|
67
|
+
function render() {
|
|
68
|
+
const tbody = document.getElementById('eval-body');
|
|
69
|
+
tbody.innerHTML = '';
|
|
70
|
+
|
|
71
|
+
// Sort: should-trigger first, then should-not-trigger
|
|
72
|
+
const sorted = evalItems
|
|
73
|
+
.map((item, origIdx) => ({ ...item, origIdx }))
|
|
74
|
+
.sort((a, b) => (b.should_trigger ? 1 : 0) - (a.should_trigger ? 1 : 0));
|
|
75
|
+
|
|
76
|
+
let lastGroup = null;
|
|
77
|
+
sorted.forEach(item => {
|
|
78
|
+
const group = item.should_trigger ? 'trigger' : 'no-trigger';
|
|
79
|
+
if (group !== lastGroup) {
|
|
80
|
+
const headerRow = document.createElement('tr');
|
|
81
|
+
headerRow.className = 'section-header';
|
|
82
|
+
headerRow.innerHTML = `<td colspan="3">${item.should_trigger ? 'Should Trigger' : 'Should NOT Trigger'}</td>`;
|
|
83
|
+
tbody.appendChild(headerRow);
|
|
84
|
+
lastGroup = group;
|
|
85
|
+
}
|
|
86
|
+
|
|
87
|
+
const idx = item.origIdx;
|
|
88
|
+
const tr = document.createElement('tr');
|
|
89
|
+
tr.innerHTML = `
|
|
90
|
+
<td><textarea class="query-input" onchange="updateQuery(${idx}, this.value)">${escapeHtml(item.query)}</textarea></td>
|
|
91
|
+
<td>
|
|
92
|
+
<label class="toggle">
|
|
93
|
+
<input type="checkbox" ${item.should_trigger ? 'checked' : ''} onchange="updateTrigger(${idx}, this.checked)">
|
|
94
|
+
<span class="slider"></span>
|
|
95
|
+
</label>
|
|
96
|
+
<span style="margin-left:8px;font-size:0.8rem;color:#b0aea5">${item.should_trigger ? 'Yes' : 'No'}</span>
|
|
97
|
+
</td>
|
|
98
|
+
<td><button class="btn-delete" onclick="deleteRow(${idx})">Delete</button></td>
|
|
99
|
+
`;
|
|
100
|
+
tbody.appendChild(tr);
|
|
101
|
+
});
|
|
102
|
+
updateSummary();
|
|
103
|
+
}
|
|
104
|
+
|
|
105
|
+
function escapeHtml(text) {
|
|
106
|
+
const div = document.createElement('div');
|
|
107
|
+
div.textContent = text;
|
|
108
|
+
return div.innerHTML;
|
|
109
|
+
}
|
|
110
|
+
|
|
111
|
+
function updateQuery(idx, value) { evalItems[idx].query = value; updateSummary(); }
|
|
112
|
+
function updateTrigger(idx, value) { evalItems[idx].should_trigger = value; render(); }
|
|
113
|
+
function deleteRow(idx) { evalItems.splice(idx, 1); render(); }
|
|
114
|
+
|
|
115
|
+
function addRow() {
|
|
116
|
+
evalItems.push({ query: '', should_trigger: true });
|
|
117
|
+
render();
|
|
118
|
+
const inputs = document.querySelectorAll('.query-input');
|
|
119
|
+
inputs[inputs.length - 1].focus();
|
|
120
|
+
}
|
|
121
|
+
|
|
122
|
+
function updateSummary() {
|
|
123
|
+
const trigger = evalItems.filter(i => i.should_trigger).length;
|
|
124
|
+
const noTrigger = evalItems.filter(i => !i.should_trigger).length;
|
|
125
|
+
document.getElementById('summary').textContent =
|
|
126
|
+
`${evalItems.length} queries total: ${trigger} should trigger, ${noTrigger} should not trigger`;
|
|
127
|
+
}
|
|
128
|
+
|
|
129
|
+
function exportEvalSet() {
|
|
130
|
+
const valid = evalItems.filter(i => i.query.trim() !== '');
|
|
131
|
+
const data = valid.map(i => ({ query: i.query.trim(), should_trigger: i.should_trigger }));
|
|
132
|
+
const blob = new Blob([JSON.stringify(data, null, 2)], { type: 'application/json' });
|
|
133
|
+
const url = URL.createObjectURL(blob);
|
|
134
|
+
const a = document.createElement('a');
|
|
135
|
+
a.href = url;
|
|
136
|
+
a.download = 'eval_set.json';
|
|
137
|
+
document.body.appendChild(a);
|
|
138
|
+
a.click();
|
|
139
|
+
document.body.removeChild(a);
|
|
140
|
+
URL.revokeObjectURL(url);
|
|
141
|
+
}
|
|
142
|
+
|
|
143
|
+
render();
|
|
144
|
+
</script>
|
|
145
|
+
</body>
|
|
146
|
+
</html>
|