dsh-github-copilot 0.4.2-alpha.6 → 0.4.2-alpha.7

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/AGENTS.md CHANGED
@@ -126,7 +126,7 @@ Alpha.25 admits native `subagent/descriptor` v3, already v3 in the retained rc.1
126
126
  - `src/responses-reasoning.ts`, `src/responses-reasoning-text.ts`: selected-model effort mapping and public summary assembly.
127
127
  - `src/tool-schema-compat.ts`: Copilot-only prompt-assembly filter for unusable escalation arguments and action-specific Goal update schemas.
128
128
  - `src/plan.ts`: Copilot-only, fail-closed hosted-search candidate lifecycle.
129
- - `src/auto-assessment-evidence.ts`: bounded request-local semantic classifier milestones copied into strict ephemeral turn evidence; no prompts, replay, credentials or error bodies. Total includes preparation; adapter start and first text are not HTTP dispatch/byte timings. Keep one 8s monotonic deadline checked at preparation/native request/result boundaries and 128-token compact output; delayed timer settlement is not a real-time guarantee. Prefer eligible supplier-advertised reasoning-off classifiers, but request off only when supplier and public prepared-model effort agree; unsupported native controls stay unchanged. No guessed latency ranking, adaptive timeout, cooldown or cross-turn cache without reviewed observations.
129
+ - `src/auto-assessment-evidence.ts`: bounded request-local semantic classifier milestones copied into strict ephemeral turn evidence; no prompts, replay, credentials or error bodies. Total includes preparation; adapter start and first text are not HTTP dispatch/byte timings. Keep one configured monotonic deadline (30s default, 1000–120000ms) checked at preparation/native request/result boundaries and 128-token compact output; delayed timer settlement is not a real-time guarantee. The user-approved budget trial is not calibrated latency or proof of past timeout causes. Empty classifier setting retains metadata-driven automatic selection; a nonempty exact account ID preserves proof/exclusion/text/headroom guards, never a virtual Auto ID or silent alternate-model fallback. Both native volatile Config leaves affect future assessments and preserve admitted turns. Prefer eligible supplier-advertised reasoning-off classifiers, but request off only when supplier and public prepared-model effort agree; unsupported native controls stay unchanged. No guessed optimal/latency ranking, benchmark requests, adaptive timeout, cooldown or cross-turn cache without reviewed observations.
130
130
  - `src/probe.ts`: bounded native-search capability proof.
131
131
  - `src/wire.ts`, `src/wire-anthropic.ts`: inline hosted-search streaming.
132
132
  - `src/traditional-search.ts`: `github-copilot-hosted` `ctx.web` provider.
package/CHANGELOG.md CHANGED
@@ -1,5 +1,23 @@
1
1
  # Changelog
2
2
 
3
+ ## 0.4.2-alpha.7 (prepared)
4
+
5
+ - Raise the auxiliary Auto task-assessment default total budget from eight to
6
+ thirty seconds as a user-approved trial, with a native editable
7
+ `autoSemanticAssessmentTimeoutMs` setting bounded to 1000–120000 ms.
8
+ - Expose `autoSemanticAssessmentModel`: empty keeps the existing metadata-driven
9
+ automatic classifier; an exact current-account ID can select another eligible
10
+ concrete model. Missing, excluded or unsuitable explicit choices diagnose
11
+ unavailable without silently calling another classifier.
12
+ - Persist both volatile native settings across restart while preserving admitted
13
+ steps/retries. Keep cancellation, account proof, monotonic deadline checks,
14
+ 128-token output, context conservatism and one-call/no-retry ownership.
15
+ No empirical optimal-model search or claim that past unknowns were timeouts.
16
+ - Isolate auxiliary classifier reasoning from the route's configured Chat effort:
17
+ a high-effort profile no longer adds high reasoning to classification. Preserve
18
+ ordinary Chat behavior, no tools and native assembly; omitted reasoning or
19
+ native `off` is not proof of supplier-side disablement or a hard credit cap.
20
+
3
21
  ## 0.4.2-alpha.6 (prepared)
4
22
 
5
23
  - Retain fixed assessment outcomes in existing opted-in daily Auto allocation
package/README.md CHANGED
@@ -8,7 +8,7 @@
8
8
 
9
9
  GitHub Copilot account discovery, contextual Auto routing and hosted search for DSH. The plugin reuses DSH's published native adapter, retaining a profile-wide account default with independent Session next-turn choices; it does not patch Core or maintain a second model catalog.
10
10
 
11
- **Stable release: [`0.4.1`](https://github.com/cloga/dsh-github-copilot/releases/tag/v0.4.1); prepared prerelease: `0.4.2-alpha.6`.** `0.4.2-alpha.1` was not published: its release preflight failed, and its immutable tag will not be reused. Supported host: official DSH / Windows Desktop `0.2.0-rc.2`. The stable version is also published on npm's `latest` dist-tag; prereleases remain on their channel tags. Stability-channel promotion does not expand the qualified host baseline or prove live account, model, search or Desktop health. Earlier DSH pins are historical evidence, not supported installation targets. Published, installed in a profile and loaded by a running Host are separate states.
11
+ **Stable release: [`0.4.1`](https://github.com/cloga/dsh-github-copilot/releases/tag/v0.4.1); prepared prerelease: `0.4.2-alpha.7`.** `0.4.2-alpha.1` was not published: its release preflight failed, and its immutable tag will not be reused. Supported host: official DSH / Windows Desktop `0.2.0-rc.2`. The stable version is also published on npm's `latest` dist-tag; prereleases remain on their channel tags. Stability-channel promotion does not expand the qualified host baseline or prove live account, model, search or Desktop health. Earlier DSH pins are historical evidence, not supported installation targets. Published, installed in a profile and loaded by a running Host are separate states.
12
12
 
13
13
  **Upgrade from 0.4.0:** that first stable release can interrupt concurrent managed requests with `COPILOT_PREVIEW_CREDENTIAL_CHANGED` during normal same-account native OAuth renewal. In 0.4.1, a proven plugin-owned refresh with unchanged account, endpoint and entitlements preserves already-dispatched HTTP streams; subsequent requests use freshly validated auth and metadata. Unknown/external credential notifications, revocation, cancellation and provider rejections remain explicit—not a promise of invisible recovery or automatic retry. See [the refresh boundary](./docs/copilot-accounts.md#native-oauth-refresh-and-dispatched-requests-385).
14
14
 
@@ -120,10 +120,10 @@ Supply any launcher patches with repeated `--patch /absolute/file`. Require `sup
120
120
  For a **standalone named profile**:
121
121
 
122
122
  ```sh
123
- dsh plugin --profile web add https://github.com/cloga/dsh-github-copilot/releases/download/v0.4.2-alpha.6/dsh-github-copilot-0.4.2-alpha.6.tgz
123
+ dsh plugin --profile web add https://github.com/cloga/dsh-github-copilot/releases/download/v0.4.2-alpha.7/dsh-github-copilot-0.4.2-alpha.7.tgz
124
124
  ```
125
125
 
126
- For **Desktop**, its native package manager accepts an exact npm spec, for example `dsh-github-copilot@0.4.2-alpha.6`; stable versions publish on npm's `latest` dist-tag and prereleases on their channel tags. The official rc.2 **Desktop-bundled CLI** also supports reserved-profile plugin management; a global/generic `dsh` shim is not equivalent. Qualify the installed entry before use. See [Desktop CLI qualification](./docs/npm-distribution.md#desktop-bundled-cli-on-official-rc2) and the distinct [standalone offline procedure](./docs/npm-distribution.md#controlled-offline-cli-maintenance-for-standalone-profiles). Neither path authorizes registry-policy bypass, peer patches or configuration deletion.
126
+ For **Desktop**, its native package manager accepts an exact npm spec, for example `dsh-github-copilot@0.4.2-alpha.7`; stable versions publish on npm's `latest` dist-tag and prereleases on their channel tags. The official rc.2 **Desktop-bundled CLI** also supports reserved-profile plugin management; a global/generic `dsh` shim is not equivalent. Qualify the installed entry before use. See [Desktop CLI qualification](./docs/npm-distribution.md#desktop-bundled-cli-on-official-rc2) and the distinct [standalone offline procedure](./docs/npm-distribution.md#controlled-offline-cli-maintenance-for-standalone-profiles). Neither path authorizes registry-policy bypass, peer patches or configuration deletion.
127
127
 
128
128
  After an approved reload/restart:
129
129
 
@@ -158,10 +158,14 @@ All three Auto preferences use the same account-verified, non-excluded eligible
158
158
 
159
159
  Categories come from authenticated supplier metadata, not model names, context capacity or measured quality rankings. Within the first fitting category, positive cost weights and the finite continuity bonus described above choose the model; a suitable previous model is not unconditionally retained. Category fallback is explained explicitly. One real answer model stays fixed throughout each admitted turn's steps and retries; compaction preserves its independently resolved native summary route.
160
160
 
161
- Semantic assessment is **enabled by default only for locally unknown tasks**. It makes at most one bounded auxiliary call with an **8-second end-to-end deadline** and **128-token output budget**. Set `github-copilot.autoSemanticAssessment: false` to opt out. Timeout/invalid output leaves demand unknown and uses the preference fallback; it never means simple or triggers another classifier. Caller cancellation and account invalidation remain terminal. Auxiliary calls add latency and supplier charges outside native answer Usage.
161
+ Semantic assessment is **enabled by default only for locally unknown tasks**. It makes at most one bounded auxiliary call with a **30-second default end-to-end deadline** and **128-token output budget**. Set `github-copilot.autoSemanticAssessment: false` to opt out. The native plugin configuration exposes `autoSemanticAssessmentTimeoutMs` (1000–120000 ms; `8000` restores the old budget) and `autoSemanticAssessmentModel` (empty for automatic selection, or an exact current-account model ID). They apply to subsequent assessments, not admitted retries. An unavailable/excluded/unsuitable explicit classifier reports unavailable, never silently switches. Automatic selection is metadata-driven, not measured optimal; the new default is a trial, not proof of past timeout causes. See [assessment settings](./docs/auto-task-routing.md#using-auto).
162
+
163
+ Timeout/invalid output leaves demand unknown and uses the preference fallback; it never means simple or triggers another classifier. Caller cancellation and account invalidation remain terminal. Auxiliary calls add latency and supplier charges outside native answer Usage.
162
164
 
163
165
  The deadline is checked against a monotonic clock at preparation, native request and result boundaries, even if the event loop delays the timer. Such delays can postpone settlement; they do not authorize late dispatch or late-result acceptance. Eligible Lightweight classifiers advertising reasoning `off` take precedence, with deterministic ID tie-breaking; the request uses `off` only if the public native model also supports it. This is not a measured speed ranking. Explanations distinguish the fitting category pool and previous-model continuity from semantic merit.
164
166
 
167
+ The auxiliary classifier does not inherit configured Chat reasoning effort and receives no tools. Native `off` may only omit a reasoning option, so supplier-side thinking is not guaranteed disabled. One call, bounded text input and 128 requested output tokens constrain overhead without promising a hard credit cap; ordinary Chat reasoning/tools remain unchanged.
168
+
165
169
  The reply's selection disclosure shows captured reasons and optional content-free auxiliary milestones. `uncertain` and `insufficient-evidence` do not mean the conversation is empty. **Selection unknown** means no retained record; **Selection unavailable → Retry** means the read failed. Retry rereads that turn, not inference. Evidence is bounded and Host-local, so restart/eviction can lose it.
166
170
 
167
171
  When a failed/incomplete turn has no native Usage model row, the existing ⓘ opens **Turn model and selection evidence**: successful same-turn sources show **Recorded models** with incomplete failed-attempt attribution; reliable same-turn recorded request configuration shows **Requested model**, not dispatch, execution or billing proof. Missing evidence stays **Unknown**, never inferred from today's picker/default or adjacent turns. Selection reasons remain in a disclosure underneath. Complete native routes avoid repeated model UI; turns without a closing message have no existing assistant-actions entry. Native Usage and retries are unchanged. [Evidence boundaries](./docs/automatic-model-routing.md#attribution-and-explanation).
@@ -299,14 +303,14 @@ Commit attribution uses `Assisted-by` with the actual tool, never the model prov
299
303
  GitHub Releases and npm distribute the same original verified tarball. Pin a version; verify Release SHA-256 or npm `dist.integrity`. Never repack an immutable release or move/reuse its tag.
300
304
 
301
305
  ```sh
302
- curl -LO https://github.com/cloga/dsh-github-copilot/releases/download/v0.4.2-alpha.6/dsh-github-copilot-0.4.2-alpha.6.tgz
303
- curl -LO https://github.com/cloga/dsh-github-copilot/releases/download/v0.4.2-alpha.6/SHA256SUMS
306
+ curl -LO https://github.com/cloga/dsh-github-copilot/releases/download/v0.4.2-alpha.7/dsh-github-copilot-0.4.2-alpha.7.tgz
307
+ curl -LO https://github.com/cloga/dsh-github-copilot/releases/download/v0.4.2-alpha.7/SHA256SUMS
304
308
  sha256sum --check SHA256SUMS
305
309
  ```
306
310
 
307
311
  ```powershell
308
312
  $expected = (Get-Content .\SHA256SUMS).Split()[0]
309
- $actual = (Get-FileHash .\dsh-github-copilot-0.4.2-alpha.6.tgz -Algorithm SHA256).Hash.ToLowerInvariant()
313
+ $actual = (Get-FileHash .\dsh-github-copilot-0.4.2-alpha.7.tgz -Algorithm SHA256).Hash.ToLowerInvariant()
310
314
  if ($actual -cne $expected) { throw 'Release checksum mismatch' }
311
315
  ```
312
316
 
package/README.zh.md CHANGED
@@ -8,7 +8,7 @@
8
8
 
9
9
  为 DSH 提供 GitHub Copilot 账号模型发现、上下文 Auto 选模和 hosted search。插件复用 DSH 公开的原生适配器,保留 profile 全局默认账号,并支持 Session 后续 turn 指定独立账号;不修改 Core,也不维护第二套模型目录。
10
10
 
11
- **稳定版:[`0.4.1`](https://github.com/cloga/dsh-github-copilot/releases/tag/v0.4.1);待发布预览版:`0.4.2-alpha.6`。** `0.4.2-alpha.1` 的发布预检失败,因此未发布;其不可变 tag 不会复用。支持宿主:官方 DSH / Windows Desktop `0.2.0-rc.2`。稳定版也发布到 npm 的 `latest` dist-tag;预发布版本仍使用各自渠道 tag。稳定渠道不扩大已验证的宿主基线,也不证明实时账号、模型、搜索或 Desktop 健康状态。较早 DSH pin 仅是历史证据,不是当前安装目标。已发布、已安装到 profile、已被运行中的 Host 加载,是三个不同状态。
11
+ **稳定版:[`0.4.1`](https://github.com/cloga/dsh-github-copilot/releases/tag/v0.4.1);待发布预览版:`0.4.2-alpha.7`。** `0.4.2-alpha.1` 的发布预检失败,因此未发布;其不可变 tag 不会复用。支持宿主:官方 DSH / Windows Desktop `0.2.0-rc.2`。稳定版也发布到 npm 的 `latest` dist-tag;预发布版本仍使用各自渠道 tag。稳定渠道不扩大已验证的宿主基线,也不证明实时账号、模型、搜索或 Desktop 健康状态。较早 DSH pin 仅是历史证据,不是当前安装目标。已发布、已安装到 profile、已被运行中的 Host 加载,是三个不同状态。
12
12
 
13
13
  **从 0.4.0 升级:**首个正式版在同账号正常原生 OAuth 刷新时,可能以 `COPILOT_PREVIEW_CREDENTIAL_CHANGED` 中止并行托管请求。0.4.1 仅在插件拥有的原生刷新被精确验证,且账号、端点与权限连续时,保留已发出的 HTTP 流;后续请求重新验证认证与模型元数据。未知/外部凭据通知、撤销、取消与供应商拒绝仍明确报错,不承诺所有刷新无感,也不自动重试。参见[刷新边界](./docs/copilot-accounts.md#native-oauth-refresh-and-dispatched-requests-385)。
14
14
 
@@ -102,10 +102,10 @@ node package/scripts/check-search-composition.mjs --profile-dir /absolute/profil
102
102
  **独立具名 profile**:
103
103
 
104
104
  ```sh
105
- dsh plugin --profile web add https://github.com/cloga/dsh-github-copilot/releases/download/v0.4.2-alpha.6/dsh-github-copilot-0.4.2-alpha.6.tgz
105
+ dsh plugin --profile web add https://github.com/cloga/dsh-github-copilot/releases/download/v0.4.2-alpha.7/dsh-github-copilot-0.4.2-alpha.7.tgz
106
106
  ```
107
107
 
108
- **Desktop** 原生包管理器接受精确 npm spec,例如 `dsh-github-copilot@0.4.2-alpha.6`;稳定版发布到 npm `latest` dist-tag,预发布版使用对应渠道 tag。官方 rc.2 **Desktop 随附的专属 CLI**也支持管理保留 profile;全局/普通 `dsh` shim 不等价。使用前核验实际安装入口,参见[Desktop CLI 核验](./docs/npm-distribution.md#desktop-bundled-cli-on-official-rc2)及独立的[具名 profile 离线流程](./docs/npm-distribution.md#controlled-offline-cli-maintenance-for-standalone-profiles)。两种路径都不授权绕过包源政策、修改 peer 或删除配置。
108
+ **Desktop** 原生包管理器接受精确 npm spec,例如 `dsh-github-copilot@0.4.2-alpha.7`;稳定版发布到 npm `latest` dist-tag,预发布版使用对应渠道 tag。官方 rc.2 **Desktop 随附的专属 CLI**也支持管理保留 profile;全局/普通 `dsh` shim 不等价。使用前核验实际安装入口,参见[Desktop CLI 核验](./docs/npm-distribution.md#desktop-bundled-cli-on-official-rc2)及独立的[具名 profile 离线流程](./docs/npm-distribution.md#controlled-offline-cli-maintenance-for-standalone-profiles)。两种路径都不授权绕过包源政策、修改 peer 或删除配置。
109
109
 
110
110
  经批准 reload/restart 后:
111
111
 
@@ -140,10 +140,14 @@ dsh plugin --profile web add https://github.com/cloga/dsh-github-copilot/release
140
140
 
141
141
  分类来自认证后的供应方元数据,不按模型名、上下文容量或臆测质量排名。在第一个可容纳输入的分类内,按前述正成本权重和有限连续性加权选择;上一模型合适也不会无条件保留。分类回退明确说明。同一已准入 turn 的工具步骤和重试保持同一真实回答模型;压缩保留独立解析的原生摘要路由。
142
142
 
143
- 语义判断**默认开启,仅用于本地需求未知的任务**。最多一次辅助请求,**8 秒端到端期限、128 tokens 输出预算**。可用 `github-copilot.autoSemanticAssessment: false` 关闭。超时/无效结果保留未知,再按偏好兜底,不解释为简单,也不换分类模型重试。用户取消和账号失效仍终止请求。辅助调用增加延迟与供应方费用,不计入原生回答 Usage。
143
+ 语义判断**默认开启,仅用于本地需求未知的任务**。最多一次辅助请求,**默认 30 秒端到端期限、128 tokens 输出预算**。可用 `github-copilot.autoSemanticAssessment: false` 关闭。原生插件配置提供 `autoSemanticAssessmentTimeoutMs`(1000–120000 毫秒,`8000` 恢复旧预算)和 `autoSemanticAssessmentModel`(空字符串自动选择,或当前账号的精确模型 ID)。只影响后续评估,不改变已准入 turn 的重试。指定分类模型不存在/被排除/不合适时明确报告 unavailable,不偷偷换模型。自动选择依据元数据,不代表实测最优;30 秒是试验默认值,不证明过去主要因超时。参见[评估配置](./docs/auto-task-routing.md#using-auto)。
144
+
145
+ 超时/无效结果保留未知,再按偏好兜底,不解释为简单,也不换分类模型重试。用户取消和账号失效仍终止请求。辅助调用增加延迟与供应方费用,不计入原生回答 Usage。
144
146
 
145
147
  准备、原生请求和结果边界均用单调时钟检查期限,不依赖定时器及时执行。事件循环延迟可能推迟结束,但不授权超期发起请求或接受结果。合格 Lightweight 分类模型优先考虑声明 reasoning `off` 的候选,同级按 ID 确定顺序;只有公开原生模型也支持时才发送 `off`。这不是实测速率排名。解释分别说明该分类的可容纳候选数和上一模型连续性,不将保留上一模型说成语义优选。
146
148
 
149
+ 辅助分类器不再继承配置的 Chat reasoning effort,也不传工具。原生 `off` 可能只是省略 reasoning 参数,因此不保证供应商侧关闭思考。单次调用、有界文本输入和 128 tokens 请求输出预算限制额外开销,但不承诺额度硬上限;普通 Chat 的思考与工具保持不变。
150
+
147
151
  回复的选模说明展示捕获的理由和可选的不含内容的辅助里程碑。`uncertain`、`insufficient-evidence` 不等于会话没有上下文。**Selection unknown**表示记录未保留,**Selection unavailable → Retry**表示读取失败;Retry 只重读该 turn,不重新推理。证据有界且仅限 Host 生命周期,重启/淘汰可能丢失。
148
152
 
149
153
  失败/不完整轮次的原生 Usage 没有模型行时,沿用现有 ⓘ 打开**本轮模型与选择记录**:本轮成功消息来源显示**已记录模型**并提示失败尝试归属不完整;可靠的本轮已记录请求配置仅显示**请求模型**,不证明发送、执行或计费。缺失证据显示**未知**,不从当前选择、全局默认或相邻轮次倒推。原选模说明折叠在下方;原生完整模型记录不重复展示,没有关闭消息的轮次没有现有入口。原生 Usage 与重试保持不变。[证据边界](./docs/automatic-model-routing.md#attribution-and-explanation)。
@@ -281,14 +285,14 @@ Agent 从 `node scripts/agent.mjs describe --json`、`doctor --json`及`plan <ta
281
285
  GitHub Releases 和 npm 分发同一原始已校验 tarball。固定版本并核验 Release SHA-256 或 npm `dist.integrity`;不重打包不可变 Release、不移动/复用 tag。
282
286
 
283
287
  ```sh
284
- curl -LO https://github.com/cloga/dsh-github-copilot/releases/download/v0.4.2-alpha.6/dsh-github-copilot-0.4.2-alpha.6.tgz
285
- curl -LO https://github.com/cloga/dsh-github-copilot/releases/download/v0.4.2-alpha.6/SHA256SUMS
288
+ curl -LO https://github.com/cloga/dsh-github-copilot/releases/download/v0.4.2-alpha.7/dsh-github-copilot-0.4.2-alpha.7.tgz
289
+ curl -LO https://github.com/cloga/dsh-github-copilot/releases/download/v0.4.2-alpha.7/SHA256SUMS
286
290
  sha256sum --check SHA256SUMS
287
291
  ```
288
292
 
289
293
  ```powershell
290
294
  $expected = (Get-Content .\SHA256SUMS).Split()[0]
291
- $actual = (Get-FileHash .\dsh-github-copilot-0.4.2-alpha.6.tgz -Algorithm SHA256).Hash.ToLowerInvariant()
295
+ $actual = (Get-FileHash .\dsh-github-copilot-0.4.2-alpha.7.tgz -Algorithm SHA256).Hash.ToLowerInvariant()
292
296
  if ($actual -cne $expected) { throw 'Release checksum mismatch' }
293
297
  ```
294
298
 
@@ -8,7 +8,7 @@
8
8
  },
9
9
  "package": {
10
10
  "name": "dsh-github-copilot",
11
- "version": "0.4.2-alpha.6"
11
+ "version": "0.4.2-alpha.7"
12
12
  },
13
13
  "evidence": {
14
14
  "kind": "source-and-synthetic-test-inventory",
@@ -741,7 +741,7 @@
741
741
  {"file": "tests/request-body-timeout.spec.ts", "name": "reports a skipped opt-in compression reason without changing identity diagnostics"},
742
742
  {"file": "tests/auto-task-assessment.spec.ts", "name": "rejects late results even before the scheduled timeout callback runs"},
743
743
  {"file": "tests/auto-task-assessment.spec.ts", "name": "aborts preparation at the monotonic deadline before an adapter can start"},
744
- {"file": "tests/preview-route.spec.ts", "name": "does not dispatch an auxiliary request when native preparation outlasts a delayed deadline timer"},
744
+ {"file": "tests/preview-route.spec.ts", "name": "does not dispatch an auxiliary request when native preparation outlasts a delayed deadline timer (budget %s)"},
745
745
  {"file": "tests/preview-provider.spec.ts", "name": "checks the auxiliary deadline immediately before native %s Fetch and releases the lease"},
746
746
  {"file": "tests/auto-task-classifier.spec.ts", "name": "prefers advertised reasoning-off candidates without inferring latency or model-name capabilities"},
747
747
  {"file": "tests/auto-selection-explanation.spec.ts", "name": "distinguishes continuity and category pool from semantic model merit (%s)"},
@@ -1377,6 +1377,8 @@
1377
1377
  {"file": "src/account-model-catalog.ts", "marker": "model_picker_category"},
1378
1378
  {"file": "src/auto-task-assessment.ts", "marker": "COPILOT_AUTO_ASSESSMENT_CONTEXT_OMITTED"},
1379
1379
  {"file": "src/auto-task-classifier.ts", "marker": "classifyTaskWithAdapter"},
1380
+ {"file": "src/config.ts", "marker": "autoSemanticAssessmentTimeoutMs:"},
1381
+ {"file": "src/config.ts", "marker": "autoSemanticAssessmentModel:"},
1380
1382
  {"file": "src/auto-model-routing.ts", "marker": "targetCategory"},
1381
1383
  {"file": "src/config.ts", "marker": "autoSemanticAssessment"},
1382
1384
  {"file": "src/auto-selection-explanation.ts", "marker": "No candidate fits"},
@@ -1416,6 +1418,10 @@
1416
1418
  {"file": "tests/auto-selection-explanation.spec.ts", "name": "distinguishes local rule coverage from absent conversation context in both languages"},
1417
1419
  {"file": "tests/auto-task-assessment.spec.ts", "name": "preserves caller cancellation instead of using a success-shaped fallback"},
1418
1420
  {"file": "tests/config.spec.ts", "name": "defaults semantic Auto assessment on while preserving explicit opt-out"},
1421
+ {"file": "tests/config.spec.ts", "name": "exposes a bounded assessment budget and optional exact classifier without changing opt-out"},
1422
+ {"file": "tests/auto-task-assessment.spec.ts", "name": "allows a response after the old eight-second budget under the new bounded default"},
1423
+ {"file": "tests/auto-task-assessment.spec.ts", "name": "still rejects a result at the new default deadline without relying on timer delivery"},
1424
+ {"file": "tests/auto-task-classifier.spec.ts", "name": "honors an explicit exact account classifier without category guessing or automatic fallback"},
1419
1425
  {"file": "tests/preview-route.spec.ts", "name": "assesses Auto with setting %s and freezes the captured reason across retries"},
1420
1426
  {"file": "tests/turn-selection-card.spec.ts", "name": "explains a captured task/category decision with progressive details (%s)"},
1421
1427
  {"file": "tests/auto-model-routing.spec.ts", "name": "does not discard a complete Core assistant message when it contains reasoning"},
@@ -9,14 +9,14 @@
9
9
  "sampleCount": 63,
10
10
  "limitations": "Read-only strict profile-local aggregate review, daily UTC buckets rather than exact event times. Producer 0.4.2-alpha.2; not current loaded-version proof. No Session/turn IDs, sequence, lower-category pool fit or classifier diagnostic retained. completeHistory=false, restarts=1, dropped/expired/saturated=0 in this snapshot. Cannot recover failure causes, transition matrix, execution, quality, billing or savings; no raw export uploaded."
11
11
  },
12
- "previousReview": "Issues #332/#333 prior allocation review and post-delivery review #334; diagnostic-only follow-up #398.",
12
+ "previousReview": "Issues #332/#333 prior allocation review and post-delivery review #334; fixed-outcome diagnostics #398; user-directed budget/config trial #400.",
13
13
  "findings": "63 decisions: 44 local/unknown, 4 semantic/unknown and 15 semantic/complex, all powerful target/selection. Sol selected 57 and Astra 6; 47 marked previous-model retention, not a transition matrix. Local/unknown may include disabled/unavailable/timeout/failed/invalid-result fallbacks. Semantic/unknown may include omitted-context conservatism. Persisted aggregates discarded those diagnostics, so historical causes cannot be established.",
14
- "decision": "Retain classification, category routing, positive weights, continuity, eight-second deadline, context safety and existing consent. Add only fixed content-free assessment outcomes to daily candidate/no-fit strata and their existing table. Keep legacy absent reasons explicitly unrecorded. Do not force downgrades, increase the timeout or infer classifier/network failure from unknown demand.",
15
- "validation": "Regression-first collector tests reproduced diagnostic cohort collapse and merging fresh rows with legacy unrecorded rows. 70 focused assessment/classifier/routing/collector/UI tests, source/test types and four public gateway/storage fixture tests including reopen passed. The complete 130-file low-concurrency suite passed 2724 tests with two expected skips; the default-concurrency full gate hit an existing 5000ms allocation stress-test timeout. No timeout or assertion weakened. Tooling/package and required Windows/Linux delivery gates are recorded separately in the implementation PR; this document does not claim live activation.",
14
+ "decision": "User explicitly approves a bounded budget/config experiment despite unproven timeout prevalence: default 30000ms, native editable integer 1000-120000ms with 8000 restoring the old budget; empty classifier selects automatically, exact account ID chooses only that eligible concrete model. Following requested overhead research, isolate classifier from configured Chat reasoning effort: reproduce inherited high in synthetic native wire, omit that auxiliary profile default and preserve ordinary Chat high. Native off/omission is not supplier disable proof; do not force low/minimal or invent a hard credit cap. Retain classification, category routing, weights, continuity, 128-token output, omitted-context safety, account/exclusion gates, cancellation and one-call/no-retry ownership. Keep fixed diagnostic cohorts and legacy unrecorded reasons. Defer benchmark/adaptive optimal-model selection until representative accuracy/latency/cost evidence and separate consent; do not infer classifier/network failure from unknown demand.",
15
+ "validation": "Regression-first tests failed for the hard-coded deadline, missing configuration and inherited high-effort classifier wire. Complete local pnpm verify passed: 2735 Vitest tests with 2 expected skips, 340 tooling tests, source/test types, Agent/baseline checks, clean build and package smoke. A synthetic valid nine-second result is accepted, thirty-second deadline results rejected, explicit 8s/45s budgets reach preparation fences, and fixed/missing/excluded/disabled classifiers plus same-turn settings changes retain ownership. The native wire regression verifies omitted classifier reasoning/tools and 128 output tokens while ordinary Chat preserves high effort. Exact local pack/tarball checks passed; native SettingsForms restart/fiber fixture and required Windows/Linux results are recorded in the implementation PR when executed. Synthetic evidence does not establish live timeout prevalence, classification improvement, supplier thinking disablement or optimal model choice.",
16
16
  "followUp": {
17
17
  "owner": "Maintainer or Agent assigned to post-delivery review issue #334",
18
18
  "trigger": "Review assessment outcomes after 200 new decisions with recorded reasons, or before the next assessment policy change. Retain the prior allocation trigger of 200 opportunities per model across 20 Sessions only if independently evidenced; aggregates cannot count Sessions.",
19
- "metric": "Selected plus no-fit counts by fixed assessment outcome, producer version, source and demand; explicit unrecorded/truncated/dropped/expired/saturated/restart coverage. Keep candidate opportunities and actual versus expected selections separate from task counts.",
20
- "action": "Use local sanitized aggregates to fix an evidenced dominant assessment failure or retain the conservative fallback; unknown or insufficient evidence defers tuning. No automatic upload, transition reconstruction, quality inference or credit inference. Track implementation in #398 and the open post-delivery review in #334."
19
+ "metric": "Selected plus no-fit counts by fixed assessment outcome, producer version, source and demand; explicit unrecorded/truncated/dropped/expired/saturated/restart coverage. Review captured classifier/budget/total/first-text latency only when independently available, not inferred from aggregate counts. Keep 8s and 30s/version/custom-model cohorts separate; session-normalized rates require independently evidenced Sessions.",
20
+ "action": "Compare unknown/timeout coverage and answer-start latency without assuming causal improvement; retain, lower, raise within bounds or revert the trial based on reviewed evidence. Fix an evidenced dominant failure or defer when uncertain. No automatic upload, optimal-model benchmark, transition reconstruction, quality or billing inference. Implementation #400 and open post-delivery review #334."
21
21
  }
22
22
  }
@@ -3,7 +3,7 @@
3
3
  Tracking: [#258](https://github.com/cloga/dsh-github-copilot/issues/258) and
4
4
  [evidence refinement #262](https://github.com/cloga/dsh-github-copilot/issues/262).
5
5
 
6
- **Current contract:** source candidate `0.4.0-alpha.99`, targeting official DSH/Windows
6
+ **Current contract:** source candidate `0.4.2-alpha.7`, targeting official DSH/Windows
7
7
  Desktop `0.2.0-rc.2`. Semantic assessment is enabled by default with bounded
8
8
  auxiliary timing/output evidence. Neither publication nor
9
9
  these explanations prove the version loaded in a particular Host.
@@ -22,6 +22,38 @@ namespace; it is not an extra switch on the Web search card. This changes
22
22
  subsequent assessments, not an already-admitted turn. Local known-task rules
23
23
  still work when auxiliary inference is disabled.
24
24
 
25
+ The native plugin configuration exposes these profile-wide settings under
26
+ `github-copilot` (not the Web search routing card):
27
+
28
+ | Setting | Default | Meaning |
29
+ |---|---|---|
30
+ | `autoSemanticAssessmentTimeoutMs` | `30000` | Integer total preparation/response budget from 1000 to 120000 ms. `8000` restores the earlier budget; zero/unlimited is not supported. |
31
+ | `autoSemanticAssessmentModel` | `''` | Empty uses metadata-driven automatic classifier selection. A nonempty value is an exact current-account model ID, not a display name or virtual Auto choice. |
32
+
33
+ For example, keep automatic selection while allowing the new default budget:
34
+
35
+ ```yaml
36
+ github-copilot:
37
+ autoSemanticAssessment: true
38
+ autoSemanticAssessmentTimeoutMs: 30000
39
+ autoSemanticAssessmentModel: ''
40
+ ```
41
+
42
+ These fields require the new plugin release; adding them to an older version
43
+ does not implement the behavior. Changes apply to subsequent assessments,
44
+ not an admitted turn's steps/retries. The 30-second default is a user-approved
45
+ trial, not calibrated latency or proof that past unknown results were timeouts.
46
+ Larger budgets can increase time before the answer begins and auxiliary charges.
47
+ Native Chat/search deadlines and user cancellation are independent.
48
+
49
+ An explicit classifier may use any account-advertised category, but still must
50
+ be an entitled, unexcluded concrete model with text capability, validated protocol
51
+ and the existing input/output headroom. A missing, excluded or unsuitable choice
52
+ produces `COPILOT_AUTO_CLASSIFIER_CONFIGURED_MODEL_UNAVAILABLE`, then the existing
53
+ unknown-demand/unavailable diagnostic; no silent alternate classifier is called.
54
+ Answer selection remains a separate Auto decision. The classifier result does
55
+ not choose its own answer model.
56
+
25
57
  Open the reply's selection explanation for the captured decision. Evidence is
26
58
  Host-lifetime and bounded; after restart or eviction it can be unavailable.
27
59
  Do not infer an old decision from the current picker.
@@ -30,7 +62,7 @@ Do not infer an old decision from the current picker.
30
62
  |---|---|
31
63
  | `uncertain` / unknown demand | The available evidence does not establish task demand; it is not a failed answer or a claim that the conversation is empty. |
32
64
  | `insufficient-evidence` | A local rule or classifier could not justify a demand category. Longer context alone does not guarantee classification. |
33
- | Semantic timeout | No validated result within the 8-second end-to-end budget. Demand remains unknown; Efficiency/Balance target Versatile, Intelligence targets Powerful, subject to eligibility and category fallback. |
65
+ | Semantic timeout | No validated result within the captured configured end-to-end budget (30 seconds by default). Demand remains unknown; Efficiency/Balance target Versatile, Intelligence targets Powerful, subject to eligibility and category fallback. |
34
66
  | Missing auxiliary milestone | That milestone was not observed before assessment settled, not proof the supplier never sent data. |
35
67
  | Selection unknown/unavailable | Captured selection evidence is absent or could not be read, distinct from unknown task demand. |
36
68
 
@@ -115,7 +147,7 @@ charges and latency, separate from native chat Usage and without inferred cost.
115
147
 
116
148
  ## Auxiliary latency evidence and bounded output (#272)
117
149
 
118
- The assessment still has one **8-second end-to-end deadline**, including account
150
+ The assessment has one **configured end-to-end deadline, 30 seconds by default**, including account
119
151
  discovery/preparation and the native call. It does not retry on timeout or try a
120
152
  second classifier. The response budget is now 128 output tokens, with compact
121
153
  demand JSON and at most three fixed signals requested. The strict decoder
@@ -131,17 +163,44 @@ this is not a hard real-time guarantee or evidence of a supplier latency cause.
131
163
  The owned timer and abort listener are disposed when assessment settles.
132
164
 
133
165
  Among eligible supplier Lightweight text candidates with the existing input
134
- headroom, prefer advertised reasoning `off`, then deterministic ID ordering.
166
+ headroom, automatic selection prefers unmarked cost, advertised reasoning `off`,
167
+ then deterministic ID ordering. An explicit exact-ID setting selects only that
168
+ eligible model, without category or cost re-ranking.
135
169
  This does not infer speed, health or quality from effort or model names, and
136
170
  does not launch a second candidate if native support is absent.
137
171
 
172
+ Automatic selection is metadata-driven, not an empirical fastest/best model
173
+ search. No additional benchmark requests, adaptive deadlines, per-model health
174
+ ranking, cooldown or cross-turn cache are introduced. Those require separately
175
+ reviewed representative latency/accuracy/cost observations and consent, not a
176
+ guess from model names or a timeout suspicion. Review outcomes under #334 before
177
+ claiming the increased default improves live classification.
178
+
138
179
  Request `off` reasoning only when both current supplier metadata advertises it
139
180
  and the public prepared model lists an `off` effort. Supplier metadata alone
140
181
  does not establish native support: current managed reasoning maps can decline
141
- `off` even when advertised. In that case retain native policy, do not change
182
+ `off` even when advertised. The auxiliary adapter does not inherit the route's
183
+ configured Chat reasoning effort: a synthetic high-effort profile previously
184
+ sent `reasoning.effort: high` on the classifier request. Its isolated public
185
+ profile now omits that default while ordinary Chat retains it. This prevents an
186
+ explicit high-effort instruction, not supplier-side thinking.
187
+
188
+ Even native `off` can mean omitting the reasoning option rather than sending a
189
+ supplier disable instruction. Do not force `low` or `minimal`: on some protocols
190
+ that enables thinking and enlarges the SDK's requested output budget. Retain
191
+ native assembly, do not change
142
192
  shared model maps, guess a wire value or patch Core. No claim that reasoning is
143
193
  disabled for all classifiers is made.
144
194
 
195
+ Tools and an execution loop are absent. Only locally unknown tasks use one
196
+ classifier call per admitted turn; subsequent steps/retries reuse its result.
197
+ Text projection stays bounded to twelve rows, 1600 characters per row and
198
+ 8000 serialized characters; files, images and replay bodies are not included.
199
+ The requested output budget remains 128 tokens. Neither those bounds nor the
200
+ waiting deadline establishes a hard supplier reasoning-token, credit or dollar
201
+ cap. Measure accuracy, elapsed time and auxiliary usage separately before
202
+ reducing context/output or claiming a cheapest/fastest classifier.
203
+
145
204
  Captured Host-lifetime selection evidence now includes a bounded public
146
205
  classifier ID, configured waiting budget, total monotonic assessment elapsed
147
206
  time, adapter-invocation start, first nonempty text time, native stop/non-stop
@@ -53,8 +53,10 @@ never reconstructed from demand/source. Count decisions using Selected plus
53
53
  No-fit, not candidate opportunities (one decision may have many candidates).
54
54
  These outcomes explain classifier settlement, not network root cause or model
55
55
  execution; no error bodies, signals, prompts or Session/turn IDs are added.
56
- Classification, its eight-second budget and conservative omitted-context guard
57
- are unchanged. Older strict plugin versions may reject new snapshot fields on
56
+ Classification and the conservative omitted-context guard stay unchanged.
57
+ The separately configured assessment budget defaults to 30 seconds as of alpha.7;
58
+ producer versions separate this trial from the earlier eight-second budget.
59
+ Older strict plugin versions may reject new snapshot fields on
58
60
  rollback; do not delete or silently rewrite retained evidence to make them parse.
59
61
  The separate **Enable request observations** control uses the hidden volatile
60
62
  `github-copilot.requestDiagnosticsEnabled` leaf with path-level CAS and exact
package/lib/client.js CHANGED
@@ -7190,7 +7190,7 @@ Set the \`cycles\` parameter to \`"ref"\` to resolve cyclical schemas with defs.
7190
7190
  };
7191
7191
  //#endregion
7192
7192
  //#region package.json
7193
- var version = "0.4.2-alpha.6";
7193
+ var version = "0.4.2-alpha.7";
7194
7194
  //#endregion
7195
7195
  //#region lib/types/diagnostics-client.js
7196
7196
  let owner;