@lay111/dsh-plugin-google 1.0.7 → 1.0.9

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (4) hide show
  1. package/README.md +5 -5
  2. package/package.json +1 -1
  3. package/src/api.js +47 -27
  4. package/src/models.js +33 -32
package/README.md CHANGED
@@ -41,12 +41,12 @@
41
41
  | 模型 ID | 模型名称 | 深度思考 (Thinking) | 特点与适用场景 |
42
42
  | :--- | :--- | :---: | :--- |
43
43
  | **`gemini-3.7-flash`** | **Gemini 3.7 Flash High** *(默认旗舰)* | **10,000 Budget 满血思考** | **日常主力推荐**。超强代码逻辑、系统级跨文件重构与复杂 Bug 诊断 |
44
- | **`gemini-3.7-flash-standard`** | Gemini 3.7 Flash (Standard) | 快速低延迟 | 适合单文件快速小改动与日常常规提问 |
45
- | **`gemini-3.6-flash`** | Gemini 3.6 Flash | 4,000 思考 | 速度极快的高性价比轻量模型 |
46
- | **`gemini-3.5-flash`** | Gemini 3.5 Flash | 4,000 思考 | 经典 Flash 系列稳定模型 |
47
- | **`gemini-3.1-pro`** | Gemini 3.1 Pro (Agent) | 8,000 思考 | 专为长上下文深度推演设计的 Pro 级模型 |
44
+ | **`gemini-3.7-flash-medium`** | Gemini 3.7 Flash Medium | 5,000 思考 | 均衡模式，兼顾思考深度与低延迟，配额更充裕 |
45
+ | **`gemini-3.7-flash-low`** | Gemini 3.7 Flash Low (Fast) | 1,000 思考 | 极速响应，适合单文件快速修改与小改动 |
48
46
  | **`claude-sonnet-4.6`** | Claude Sonnet 4.6 (Thinking) | 深度思考 | 逻辑严密、遵循规范性极强的 Anthropic 顶尖编程模型 |
49
- | **`claude-opus-4.6`** | Claude Opus 4.6 (Thinking) | 超强思考 | 适合超大型复杂架构设计与高阶算法推演 |
47
+ | **`gemini-3.1-pro`** | Gemini 3.1 Pro (High Reasoning) | 8,000 思考 | Pro 级旗舰深度推演大模型（100万超大上下文） |
48
+ | **`gemini-3.6-flash`** | Gemini 3.6 Flash High | 4,000 思考 | 经典速度极快的高性价比轻量模型 |
49
+ | **`gemini-3.5-flash`** | Gemini 3.5 Flash | 基础模式 | 经典 Flash 系列稳定模型 |
50
50
  | **`gpt-oss-120b`** | GPT-OSS 120B (Medium) | 基础思考 | 开源生态 120B 大模型 |
51
51
 
52
52
  ---
package/package.json CHANGED
@@ -1,6 +1,6 @@
1
1
  {
2
2
  "name": "@lay111/dsh-plugin-google",
3
- "version": "1.0.7",
3
+ "version": "1.0.9",
4
4
  "description": "Native DeepSeek Harness Plugin for Google Antigravity Pro & Gemini/Claude Models",
5
5
  "type": "module",
6
6
  "main": "src/index.js",
package/src/api.js CHANGED
@@ -68,23 +68,24 @@ export function formatMessagesToAntigravity(messages, modelSpec, projectId) {
68
68
  contents.push({ role: 'user', parts: [{ text: 'Hello' }] });
69
69
  }
70
70
 
71
- const isClaude = modelSpec.id.includes('claude');
71
+ const isClaude = modelSpec.id.includes('claude') || modelSpec.wireId.includes('claude');
72
72
  const payload = {
73
73
  project: projectId || 'aicode-consumers',
74
74
  model: modelSpec.wireId,
75
75
  requestId: `agent/dsh/${Date.now()}/traj_${Date.now()}/1`,
76
76
  request: {
77
77
  contents,
78
- ...(modelSpec.supportsThinking
79
- ? {
80
- generationConfig: {
78
+ generationConfig: {
79
+ maxOutputTokens: modelSpec.maxTokens || 8192,
80
+ ...(modelSpec.supportsThinking
81
+ ? {
81
82
  thinkingConfig: {
82
83
  includeThoughts: true,
83
84
  thinkingBudget: modelSpec.thinkingBudget || 10000,
84
85
  },
85
- },
86
- }
87
- : {}),
86
+ }
87
+ : {}),
88
+ },
88
89
  ...(systemInstruction ? { systemInstruction } : {}),
89
90
  labels: {
90
91
  trajectory_id: `traj_${Date.now()}`,
@@ -116,33 +117,52 @@ export async function* streamAntigravity(options, accessToken, projectId) {
116
117
 
117
118
  let response = null;
118
119
  let lastError = null;
120
+ const maxRetries = 3;
121
+
122
+ for (let attempt = 0; attempt < maxRetries; attempt++) {
123
+ for (const endpoint of ANTIGRAVITY_ENDPOINTS) {
124
+ try {
125
+ const res = await fetch(endpoint, {
126
+ method: 'POST',
127
+ headers: {
128
+ Authorization: `Bearer ${accessToken}`,
129
+ 'Content-Type': 'application/json',
130
+ Accept: 'text/event-stream',
131
+ 'User-Agent': USER_AGENT,
132
+ },
133
+ body: JSON.stringify(payload),
134
+ signal: options.signal,
135
+ });
136
+
137
+ if (res.ok) {
138
+ response = res;
139
+ break;
140
+ }
119
141
 
120
- for (const endpoint of ANTIGRAVITY_ENDPOINTS) {
121
- try {
122
- const res = await fetch(endpoint, {
123
- method: 'POST',
124
- headers: {
125
- Authorization: `Bearer ${accessToken}`,
126
- 'Content-Type': 'application/json',
127
- Accept: 'text/event-stream',
128
- 'User-Agent': USER_AGENT,
129
- },
130
- body: JSON.stringify(payload),
131
- signal: options.signal,
132
- });
142
+ const status = res.status;
143
+ const errText = await res.text();
144
+ lastError = `(HTTP ${status}) ${errText}`;
133
145
 
134
- if (res.ok) {
135
- response = res;
136
- break;
137
- } else {
138
- lastError = await res.text();
146
+ // If rate-limited (429) or temporary server overload (503), wait and retry
147
+ if ((status === 429 || status === 503) && attempt < maxRetries - 1) {
148
+ await new Promise((r) => setTimeout(r, (attempt + 1) * 1500));
149
+ continue;
150
+ }
151
+ } catch (err) {
152
+ lastError = err.message;
153
+ if (attempt < maxRetries - 1) {
154
+ await new Promise((r) => setTimeout(r, (attempt + 1) * 1000));
155
+ }
139
156
  }
140
- } catch (err) {
141
- lastError = err.message;
142
157
  }
158
+
159
+ if (response) break;
143
160
  }
144
161
 
145
162
  if (!response) {
163
+ if (lastError?.includes('429') || lastError?.includes('exhausted')) {
164
+ throw new Error(`Google Quota Exceeded (429): 当前模型并发或配额超限。建议在 TUI 中输入 /model 切换至 claude-sonnet-4.6、gemini-3.7-flash-medium 或 gemini-3.6-flash，或稍等 20 秒后重试。`);
165
+ }
146
166
  throw new Error(`Google Antigravity Upstream Error: ${lastError}`);
147
167
  }
148
168
 
package/src/models.js CHANGED
@@ -10,61 +10,62 @@ export const ANTIGRAVITY_MODELS = [
10
10
  description: 'Google flagship coding model with 10k thinking budget (Fast & Deep)',
11
11
  },
12
12
  {
13
- id: 'gemini-3.7-flash-standard',
14
- name: 'Gemini 3.7 Flash (Standard)',
15
- wireId: 'gemini-3.7-flash',
13
+ id: 'gemini-3.7-flash-medium',
14
+ name: 'Gemini 3.7 Flash Medium',
15
+ wireId: 'gemini-3.7-flash-medium',
16
16
  contextWindow: 1048576,
17
17
  maxTokens: 65536,
18
- supportsThinking: false,
19
- description: 'Standard low-latency mode without extended thinking',
18
+ supportsThinking: true,
19
+ thinkingBudget: 5000,
20
+ description: 'Balanced latency and thinking depth',
20
21
  },
21
22
  {
22
- id: 'gemini-3.6-flash',
23
- name: 'Gemini 3.6 Flash',
24
- wireId: 'gemini-3.6-flash',
23
+ id: 'gemini-3.7-flash-low',
24
+ name: 'Gemini 3.7 Flash Low (Fast)',
25
+ wireId: 'gemini-3.7-flash-low',
25
26
  contextWindow: 1048576,
26
27
  maxTokens: 65536,
27
28
  supportsThinking: true,
28
- thinkingBudget: 4000,
29
- description: 'Fast lightweight Flash model',
29
+ thinkingBudget: 1000,
30
+ description: 'Fast lightweight thinking mode for rapid iterations',
30
31
  },
31
32
  {
32
- id: 'gemini-3.5-flash',
33
- name: 'Gemini 3.5 Flash',
34
- wireId: 'gemini-3.5-flash',
35
- contextWindow: 1048576,
36
- maxTokens: 65536,
33
+ id: 'claude-sonnet-4.6',
34
+ name: 'Claude Sonnet 4.6 (Thinking)',
35
+ wireId: 'claude-sonnet-4-6',
36
+ contextWindow: 200000,
37
+ maxTokens: 64000,
37
38
  supportsThinking: true,
38
- thinkingBudget: 4000,
39
- description: 'Classic Flash model',
39
+ description: 'Anthropic Claude Sonnet 4.6 with reasoning chain via Google Pro',
40
40
  },
41
41
  {
42
42
  id: 'gemini-3.1-pro',
43
- name: 'Gemini 3.1 Pro (Agent)',
43
+ name: 'Gemini 3.1 Pro (High Reasoning)',
44
44
  wireId: 'gemini-3.1-pro-low',
45
45
  contextWindow: 1048576,
46
46
  maxTokens: 65536,
47
47
  supportsThinking: true,
48
48
  thinkingBudget: 8000,
49
- description: 'Pro-tier deep reasoning agent model',
49
+ description: 'Pro-tier flagship deep reasoning model (8k Thinking)',
50
50
  },
51
51
  {
52
- id: 'claude-sonnet-4.6',
53
- name: 'Claude Sonnet 4.6 (Thinking)',
54
- wireId: 'claude-sonnet-4-6-thinking',
55
- contextWindow: 200000,
56
- maxTokens: 64000,
52
+ id: 'gemini-3.6-flash',
53
+ name: 'Gemini 3.6 Flash High',
54
+ wireId: 'gemini-3.6-flash-high',
55
+ contextWindow: 1048576,
56
+ maxTokens: 65536,
57
57
  supportsThinking: true,
58
- description: 'Anthropic Claude Sonnet with reasoning chain via Antigravity',
58
+ thinkingBudget: 4000,
59
+ description: 'Fast lightweight Flash model with reasoning',
59
60
  },
60
61
  {
61
- id: 'claude-opus-4.6',
62
- name: 'Claude Opus 4.6 (Thinking)',
63
- wireId: 'claude-opus-4-6-thinking',
64
- contextWindow: 200000,
65
- maxTokens: 64000,
66
- supportsThinking: true,
67
- description: 'Anthropic Claude Opus for highest-tier architecture and logic modeling',
62
+ id: 'gemini-3.5-flash',
63
+ name: 'Gemini 3.5 Flash',
64
+ wireId: 'gemini-3.5-flash-low',
65
+ contextWindow: 1048576,
66
+ maxTokens: 65536,
67
+ supportsThinking: false,
68
+ description: 'Classic Flash model',
68
69
  },
69
70
  {
70
71
  id: 'gpt-oss-120b',