dsh-vertex-argon 2.2.4 → 2.2.6

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/CHANGELOG.md CHANGED
@@ -2,6 +2,27 @@
2
2
 
3
3
  All notable changes to **Gemini 4 Adapter** (`dsh-vertex-argon`) are documented in this file.
4
4
 
5
+ ## [2.2.6] - 2026-11-19
6
+
7
+ ### Fixed
8
+ - **Fail-Closed Unsupported Part Validation (`google-signatures.js`)**:
9
+ - Validated every candidate Part inside `observe(chunk)` against the supported `pi-ai` replay contract (`text`, `thought`, `thoughtSignature` / `thought_signature`, `functionCall` / `function_call`); unrepresentable parts (`executableCode`, `codeExecutionResult`, `functionResponse`, `inlineData`, `fileData`, mixed `text + functionCall`, `functionCall` with `thought: true`, or bare `{ thought: true }`) now fail closed with `ARGON_SIGNATURE_REPLAY` instead of being silently dropped.
10
+ - **Symmetric Cross-Chunk & Same-Chunk Text/Thinking Merge (`google-signatures.js`)**:
11
+ - Made `splitSigned` symmetric between `text` and `thinking`: cross-chunk continuations from a non-empty signed block to an unsigned block merge while preserving the signature, whereas same-chunk signature boundaries, conflicting signatures (`S1` → `S2`), and explicit empty signed carriers split 1:1.
12
+ - **Immutable Deep-Frozen Model Descriptors (`adapter.js`)**:
13
+ - Added `freezeDescriptor` deep-freezing returned/registered model descriptors (`input`, `thinkingLevelMap`, `cost`, and the descriptor object itself while preserving `MAX_TOKENS_CEILING`) in both `getProfiles()` and `argonProvider()`, preventing callers from mutating cached descriptor state via `getModels()`.
14
+
15
+ ## [2.2.5] - 2026-11-19
16
+
17
+ ### Fixed
18
+ - **Same-Chunk Unsigned-to-Signed Part Boundaries (`google-signatures.js`)**:
19
+ - Tracked `currentFromThisChunk` inside `observe(chunk)` so an unsigned part followed immediately by a signed part (`text` or `thinking`) within the same `candidates[0].content.parts` array always splits 1:1 into separate blocks while preserving cross-chunk delta merging.
20
+ - **Strict `extraBody` Effort/Null Hardening & Direct-Provider Origin Pinning (`adapter.js`)**:
21
+ - Added `'reasoningEffort'`, `'reasoning_effort'`, `'thinkingLevel'`, and `'thinking_level'` to `EXTRA_BODY_STRIPPED_KEYS` (and stripped `reasoningEffort` / `reasoning_effort` on `body` & `cleanConfig` in `gemini4StreamSimple`), and rejected explicit `null` / non-plain-object values for `httpOptions.extraBody`, `extraBody.generationConfig`, and `extraBody.generation_config` with `INVALID_REQUEST`.
22
+ - Pinned `baseUrl` and `contextWindow` on `effectiveModel` in `argonProvider.streamSimple` from `configured` alongside `id`, `provider`, `api`, `maxTokens`, `thinkingLevelMap`, and `[MAX_TOKENS_CEILING]`.
23
+ - **Two-Stage `apply()` Cleanup Rollback (`index.js`)**:
24
+ - Rolled back both `registration?.()` and `unpatchPiAi()` in `try/finally` inside both the `ctx.effect` disposer and `catch` handler if `ctx.effect(...)` throws after `ctx.llm.registerAdapter()` has succeeded.
25
+
5
26
  ## [2.2.4] - 2026-11-19
6
27
 
7
28
  ### Fixed
package/README.md CHANGED
@@ -4,7 +4,7 @@
4
4
 
5
5
  **面向 [DeepSeek Harness](<https://github.com/deepseek-ai/deepseek-harness>) 的 Google 官方双通道(Vertex AI + Gemini Developer API)Gemini 4 适配器,原生支持 Argon、Carbon、Hydrogen 与 Neon 八模型路由、GenerateContent / Interactions 双协议、65K+ 多跳长解码与请求级诊断。**
6
6
 
7
- - **当前版本**:`2.2.4`([CHANGELOG.md](<CHANGELOG.md>))
7
+ - **当前版本**:`2.2.6`([CHANGELOG.md](<CHANGELOG.md>))
8
8
  - **npm 包名**:`dsh-vertex-argon`(保留既有包名、官方通道路由与本地存储域 ID,官方通道用户可直接原地升级;`2.0.0` 已移除 `1.6.x` 非官方反代通道)
9
9
  - **运行要求**:DSH `>=0.2.0-rc.2`,Node.js `^22.19.0 || >=24.0.0`(见 [package.json](<package.json>))
10
10
 
package/adapter.js CHANGED
@@ -423,6 +423,10 @@ const EXTRA_BODY_STRIPPED_KEYS = [
423
423
  'thinking_config',
424
424
  'thinkingBudget',
425
425
  'thinking_budget',
426
+ 'thinkingLevel',
427
+ 'thinking_level',
428
+ 'reasoningEffort',
429
+ 'reasoning_effort',
426
430
  'includeThoughts',
427
431
  'include_thoughts',
428
432
  'temperature',
@@ -439,7 +443,7 @@ const EXTRA_BODY_STRIPPED_KEYS = [
439
443
  ]
440
444
 
441
445
  function sanitizeGemini4ExtraBody(extraBody, stripExplicitCache) {
442
- if (extraBody === undefined || extraBody === null) return extraBody
446
+ if (extraBody === undefined) return extraBody
443
447
  if (!isPlainObjectRecord(extraBody)) {
444
448
  throw new LlmError('Gemini 4 httpOptions.extraBody must be a plain object.', 'INVALID_REQUEST')
445
449
  }
@@ -450,7 +454,7 @@ function sanitizeGemini4ExtraBody(extraBody, stripExplicitCache) {
450
454
  delete cleanExtraBody.cached_content
451
455
  }
452
456
  for (const cfgKey of ['generationConfig', 'generation_config']) {
453
- if (cleanExtraBody[cfgKey] === undefined || cleanExtraBody[cfgKey] === null) continue
457
+ if (!Object.hasOwn(cleanExtraBody, cfgKey) || cleanExtraBody[cfgKey] === undefined) continue
454
458
  if (!isPlainObjectRecord(cleanExtraBody[cfgKey])) {
455
459
  throw new LlmError(`Gemini 4 httpOptions.extraBody.${cfgKey} must be a plain object.`, 'INVALID_REQUEST')
456
460
  }
@@ -580,7 +584,9 @@ export function gemini4StreamSimple(
580
584
  delete body.cachedContent
581
585
  delete body.cached_content
582
586
  }
583
- const { temperature: _t, topP: _tp, topK: _tk, candidateCount: _cc, frequencyPenalty: _fp, presencePenalty: _pp, ...cleanConfig } = body.config
587
+ delete body.reasoningEffort
588
+ delete body.reasoning_effort
589
+ const { temperature: _t, topP: _tp, topK: _tk, candidateCount: _cc, frequencyPenalty: _fp, presencePenalty: _pp, reasoningEffort: _reBody, reasoning_effort: _reSnake, ...cleanConfig } = body.config
584
590
  if (supportsXhigh) {
585
591
  delete cleanConfig.cachedContent
586
592
  delete cleanConfig.cached_content
@@ -846,14 +852,33 @@ export function resolveConfigValues(raw = {}) {
846
852
  }
847
853
  }
848
854
 
855
+ function freezeDescriptor(descriptor) {
856
+ if (!descriptor || typeof descriptor !== 'object') return descriptor
857
+ const copy = {
858
+ ...descriptor,
859
+ ...(Array.isArray(descriptor.input) ? { input: Object.freeze([...descriptor.input]) } : {}),
860
+ ...(descriptor.thinkingLevelMap && typeof descriptor.thinkingLevelMap === 'object'
861
+ ? { thinkingLevelMap: Object.freeze({ ...descriptor.thinkingLevelMap }) }
862
+ : {}),
863
+ ...(descriptor.cost && typeof descriptor.cost === 'object'
864
+ ? { cost: Object.freeze({ ...descriptor.cost }) }
865
+ : {}),
866
+ }
867
+ for (const sym of Object.getOwnPropertySymbols(descriptor)) {
868
+ copy[sym] = descriptor[sym]
869
+ }
870
+ return Object.freeze(copy)
871
+ }
872
+
849
873
  /** Serve the official Vertex AI and Gemini API models in one plugin group. */
850
874
  export function argonProvider(nativeVertex, models, defaultLevel = undefined, nativeGoogle = googleProvider(), longDecode = true, recordDiagnostic, serviceConfig = {}, responseHub, structuredOutput) {
851
- const modelMap = new Map((Array.isArray(models) ? models : [models]).map(m => [m.id, m]))
875
+ const frozenModels = (Array.isArray(models) ? models : [models]).map(freezeDescriptor)
876
+ const modelMap = new Map(frozenModels.map(m => [m.id, m]))
852
877
  return {
853
878
  id: PROVIDER,
854
879
  name: 'Gemini 4 Adapter',
855
880
  auth: nativeVertex.auth,
856
- getModels: () => [...modelMap.values()],
881
+ getModels: () => [...frozenModels],
857
882
  streamSimple(requestModel, context, options = {}) {
858
883
  const configured = modelMap.get(requestModel?.id)
859
884
  if (!configured || requestModel?.provider !== PROVIDER) {
@@ -868,6 +893,8 @@ export function argonProvider(nativeVertex, models, defaultLevel = undefined, na
868
893
  id: configured.id,
869
894
  provider: PROVIDER,
870
895
  api: configured.api,
896
+ baseUrl: configured.baseUrl,
897
+ contextWindow: configured.contextWindow,
871
898
  maxTokens: configured.maxTokens,
872
899
  thinkingLevelMap: configured.thinkingLevelMap,
873
900
  [MAX_TOKENS_CEILING]: configured[MAX_TOKENS_CEILING] ?? (Number.isSafeInteger(configured.maxTokens) ? configured.maxTokens : undefined),
@@ -1187,8 +1214,9 @@ export function createArgonAdapter(ctx, configInput, nativeVertex = googleVertex
1187
1214
  cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 },
1188
1215
  })
1189
1216
  }
1190
- for (const modelDescriptor of models) {
1191
- modelDescriptor[MAX_TOKENS_CEILING] = cfg.explicitMaxTokensCap
1217
+ for (let i = 0; i < models.length; i++) {
1218
+ models[i][MAX_TOKENS_CEILING] = cfg.explicitMaxTokensCap
1219
+ models[i] = freezeDescriptor(models[i])
1192
1220
  }
1193
1221
  const configuredMaxTokens = new Map(models.map(m => [m.id, cfg.maxTokens]))
1194
1222
  const configuredReasoning = cfg.hasExplicitReasoning ? cfg.reasoning : undefined
@@ -1,5 +1,7 @@
1
1
  import { createHash, randomUUID } from 'node:crypto'
2
2
 
3
+ const ALLOWED_PART_KEYS = new Set(['text', 'thought', 'thoughtSignature', 'thought_signature', 'functionCall', 'function_call'])
4
+
3
5
  const PREFIX = '{"dshGoogleSignatureOnly":'
4
6
  const base64 = value => typeof value === 'string' && value.length > 0 && value.length % 4 === 0 && /^[A-Za-z0-9+/]+={0,2}$/.test(value)
5
7
  const digest = content => createHash('sha256').update(JSON.stringify(content.map(block => block.type === 'text' ? ['text', block.text] : block.type === 'thinking' ? ['thinking', block.thinking] : ['toolCall', block.id, block.name, block.arguments]))).digest('hex')
@@ -115,7 +117,7 @@ export function createGoogleSignatureBridge(context, model) {
115
117
  const historyDigest = history.digest('hex')
116
118
  const content = [], pending = [], standalone = new Map(), usedToolIds = new Set()
117
119
  const partial = { role: 'assistant', api: model.api, provider: model.provider, model: model.id, content }
118
- let current, currentSignedInThisChunk = false, seen = false, bytes = 0
120
+ let current, currentFromThisChunk = false, seen = false, bytes = 0
119
121
  const emit = event => {
120
122
  if (pending.length >= 32768) throw failure('Google response exceeded the bounded pending event count.')
121
123
  pending.push({ ...event, partial })
@@ -136,7 +138,7 @@ export function createGoogleSignatureBridge(context, model) {
136
138
  ...(sig ? { contentSignature: sig } : {}),
137
139
  })
138
140
  current = undefined
139
- currentSignedInThisChunk = false
141
+ currentFromThisChunk = false
140
142
  }
141
143
  const finishMessage = message => {
142
144
  close()
@@ -160,15 +162,56 @@ export function createGoogleSignatureBridge(context, model) {
160
162
  },
161
163
  observe(chunk) {
162
164
  seen = true
163
- currentSignedInThisChunk = false
165
+ currentFromThisChunk = false
164
166
  for (const part of chunk.candidates?.[0]?.content?.parts ?? []) {
167
+ if (!part || typeof part !== 'object' || Array.isArray(part)) {
168
+ close()
169
+ throw failure('Google returned a malformed candidate part.')
170
+ }
171
+ const definedKeys = Object.keys(part).filter(key => part[key] !== undefined)
172
+ if (definedKeys.length === 0) continue
173
+ const unsupportedKeys = definedKeys.filter(key => !ALLOWED_PART_KEYS.has(key))
174
+ if (unsupportedKeys.length > 0) {
175
+ close()
176
+ throw failure('Google returned an unsupported candidate part (' + unsupportedKeys.join(', ') + ') that cannot be replayed.')
177
+ }
178
+ const call = part.functionCall ?? part.function_call
179
+ const hasText = part.text !== undefined
180
+ const hasCall = call !== undefined
181
+ if (part.thought !== undefined && typeof part.thought !== 'boolean') {
182
+ close()
183
+ throw failure('Google returned a candidate part with an invalid thought flag.')
184
+ }
185
+ if (hasText && typeof part.text !== 'string') {
186
+ close()
187
+ throw failure('Google returned a candidate part with an invalid text field.')
188
+ }
189
+ if (hasCall && (!call || typeof call !== 'object' || Array.isArray(call))) {
190
+ close()
191
+ throw failure('Google returned a candidate part with an invalid functionCall payload.')
192
+ }
193
+ if (hasText && hasCall) {
194
+ close()
195
+ throw failure('Google returned a mixed text/functionCall part that cannot be represented losslessly.')
196
+ }
197
+ if (hasCall && part.thought === true) {
198
+ close()
199
+ throw failure('Google returned a thought-flagged functionCall part that cannot be represented in replay history.')
200
+ }
165
201
  bytes += Buffer.byteLength(JSON.stringify(part))
166
202
  if (bytes > 64 * 1024 * 1024) throw failure('Google response exceeded the bounded signature replay buffer.')
167
203
  // A Part may serialize the signature in either wire spelling.
168
204
  const sig = part.thoughtSignature ?? part.thought_signature
169
- const onlySignature = typeof sig === 'string' && Object.keys(part).filter(key => part[key] !== undefined).every(key => key === 'thoughtSignature' || key === 'thought_signature' || key === 'thought')
205
+ if (sig !== undefined && typeof sig !== 'string') {
206
+ close()
207
+ throw failure('Google returned a candidate part with an invalid thoughtSignature.')
208
+ }
209
+ const onlySignature = typeof sig === 'string' && !hasText && !hasCall
170
210
  if (onlySignature) {
171
- if (!sig || (part.thought !== undefined && typeof part.thought !== 'boolean')) throw failure('Google returned an invalid standalone signature.')
211
+ if (!sig) {
212
+ close()
213
+ throw failure('Google returned an invalid standalone signature.')
214
+ }
172
215
  bytes += 320 // Include the local versioned codec and content binding in the retention bound.
173
216
  if (bytes > 64 * 1024 * 1024) throw failure('Google signature replay metadata exceeded the bounded response buffer.')
174
217
  close()
@@ -177,13 +220,17 @@ export function createGoogleSignatureBridge(context, model) {
177
220
  emit({ type: 'text_start', contentIndex: index }); emit({ type: 'text_end', contentIndex: index, content: '' })
178
221
  continue
179
222
  }
180
- if (part.text !== undefined) {
223
+ if (hasText) {
181
224
  const type = part.thought === true ? 'thinking' : 'text'
182
225
  const sigKey = type === 'text' ? 'textSignature' : 'thinkingSignature'
183
- const splitSigned = type === 'text'
184
- ? (Boolean(current?.textSignature) || (Boolean(sig) && part.text === ''))
185
- : ((Boolean(sig) && (Boolean(current?.thinkingSignature) || part.text === '')) ||
186
- (Boolean(current?.thinkingSignature) && (currentSignedInThisChunk || current.thinking === '')))
226
+ const curText = type === 'text' ? current?.text : current?.thinking
227
+ const curSig = current?.[sigKey]
228
+ const sameChunkSigBoundary = currentFromThisChunk && (Boolean(curSig) || Boolean(sig))
229
+ const crossChunkSigConflict =
230
+ (Boolean(curSig) && Boolean(sig)) ||
231
+ (Boolean(curSig) && curText === '') ||
232
+ (Boolean(sig) && part.text === '')
233
+ const splitSigned = sameChunkSigBoundary || crossChunkSigConflict
187
234
  if (!current || current.type !== type || splitSigned) {
188
235
  close(); current = type === 'text' ? { type, text: '' } : { type, thinking: '' }
189
236
  const index = start(current)
@@ -191,23 +238,25 @@ export function createGoogleSignatureBridge(context, model) {
191
238
  }
192
239
  const key = type === 'text' ? 'text' : 'thinking'
193
240
  current[key] += part.text
194
- if (sig) {
195
- current[sigKey] = sig
196
- if (type === 'thinking') currentSignedInThisChunk = true
197
- }
241
+ currentFromThisChunk = true
242
+ if (sig) current[sigKey] = sig
198
243
  emit({ type: type === 'text' ? 'text_delta' : 'thinking_delta', contentIndex: content.length - 1, delta: part.text })
244
+ continue
199
245
  }
200
- if (part.functionCall) {
246
+ if (hasCall) {
201
247
  close()
202
- const call = part.functionCall
203
- const id = call.id && !usedToolIds.has(call.id) ? call.id : 'argon_' + randomUUID().replaceAll('-', '')
248
+ const rawId = call.id ?? call.call_id
249
+ const id = rawId && !usedToolIds.has(rawId) ? rawId : 'argon_' + randomUUID().replaceAll('-', '')
204
250
  usedToolIds.add(id)
205
251
  const block = { type: 'toolCall', id, name: call.name || '', arguments: call.args ?? {}, ...(sig ? { thoughtSignature: sig } : {}) }
206
252
  const index = start(block)
207
253
  emit({ type: 'toolcall_start', contentIndex: index })
208
254
  emit({ type: 'toolcall_delta', contentIndex: index, delta: JSON.stringify(block.arguments) })
209
255
  emit({ type: 'toolcall_end', contentIndex: index, toolCall: block })
256
+ continue
210
257
  }
258
+ close()
259
+ throw failure('Google returned an unrepresentable candidate part.')
211
260
  }
212
261
  },
213
262
  stream(source) {
package/index.js CHANGED
@@ -123,11 +123,18 @@ export function apply(ctx, config) {
123
123
  try {
124
124
  registration = ctx.llm.registerAdapter([PROVIDER], adapter)
125
125
  ctx.effect(() => () => {
126
- unpatchPiAi()
127
- registration?.()
126
+ try {
127
+ registration?.()
128
+ } finally {
129
+ unpatchPiAi()
130
+ }
128
131
  })
129
132
  } catch (error) {
130
- unpatchPiAi()
133
+ try {
134
+ registration?.()
135
+ } finally {
136
+ unpatchPiAi()
137
+ }
131
138
  throw error
132
139
  }
133
140
  if (typeof ctx.on === 'function') {
package/package.json CHANGED
@@ -1,6 +1,6 @@
1
1
  {
2
2
  "name": "dsh-vertex-argon",
3
- "version": "2.2.4",
3
+ "version": "2.2.6",
4
4
  "description": "Gemini 4 Adapter for DSH: official Vertex AI & Gemini API channels (Argon, Carbon, Hydrogen & Neon), resilient long decode, explicit service tiers, and request diagnostics.",
5
5
  "license": "MIT",
6
6
  "repository": {