dsh-github-copilot 0.4.0-alpha.87 → 0.4.0-alpha.91

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.
@@ -8,7 +8,7 @@
8
8
  },
9
9
  "package": {
10
10
  "name": "dsh-github-copilot",
11
- "version": "0.4.0-alpha.87"
11
+ "version": "0.4.0-alpha.91"
12
12
  },
13
13
  "evidence": {
14
14
  "kind": "source-and-synthetic-test-inventory",
@@ -478,6 +478,8 @@
478
478
  "sourceMarkers": [
479
479
  {"file": "src/request-budget.ts", "marker": "calculateRequestBudget"},
480
480
  {"file": "src/compaction-pressure.ts", "marker": "installCopilotCompactionPressure"},
481
+ {"file": "src/pre-step-pressure.ts", "marker": "installCopilotPreStepPressure"},
482
+ {"file": "src/pre-step-pressure.ts", "marker": "selection.pending !== null"},
481
483
  {"file": "src/agent-compaction.ts", "marker": "presets.serviceFor(agent, 'compaction')"},
482
484
  {"file": "src/preview-route.ts", "marker": "COPILOT_CONTEXT_BUDGET_EXCEEDED"},
483
485
  {"file": "src/preview-provider.ts", "marker": "inspectRequest"}
@@ -485,6 +487,8 @@
485
487
  "tests": [
486
488
  {"file": "tests/preview-route.spec.ts", "name": "rejects independent input overflow before model wire, prepared=%s"},
487
489
  {"file": "tests/compaction-pressure.spec.ts", "name": "does not substitute a global engine when the bound preset has none"},
490
+ {"file": "tests/pre-step-pressure.spec.ts", "name": "skips pending selection instead of compressing against the previous route"},
491
+ {"file": "tests/fixtures/compaction-pressure-core.fixture.ts", "name": "prevents continuing-step pressure attempts while retaining complete native turn usage, preStep=%s"},
488
492
  {"file": "tests/auto-model-host.spec.ts", "name": "uses owning preset recovery availability (%s) for Auto diagnostics"},
489
493
  {"file": "tests/fixtures/scoped-compaction-core.fixture.ts", "name": "global pressure finds isolated preset recovery: %j"},
490
494
  {"file": "tests/preview-route.spec.ts", "name": "reserves the requested output separately from the independent prompt ceiling"},
@@ -578,6 +582,7 @@
578
582
  {"file": "src/model-exclusions.ts", "marker": "class ModelExclusionTurns"},
579
583
  {"file": "src/remote.ts", "marker": "githubCopilot.setModelExcluded"},
580
584
  {"file": "src/preview-provider.ts", "marker": "native.streamSimple"},
585
+ {"file": "src/auto-assessment-evidence.ts", "marker": "SemanticAssessmentEvidenceSchema"},
581
586
  {"file": "src/request-body-timeout-marker.ts", "marker": "COPILOT_REQUEST_BODY_TIMEOUT"},
582
587
  {"file": "src/request-body-evidence.ts", "marker": "REQUEST_BODY_EVIDENCE_LIMITS"},
583
588
  {"file": "src/copilot-stream-liveness.ts", "marker": "COPILOT_STREAM_IDLE_TIMEOUT"},
@@ -599,6 +604,8 @@
599
604
  {"file": "tests/preview-route.spec.ts", "name": "enriches only a verified 408 while preserving native failure classification and retry ownership"},
600
605
  {"file": "tests/preview-provider.spec.ts", "name": "observes verified %s timeout without changing bytes, Response, native retries or credential ownership"},
601
606
  {"file": "tests/request-body-timeout.spec.ts", "name": "bounds stalled response observation and gives cancellation priority"},
607
+ {"file": "tests/auto-task-assessment.spec.ts", "name": "captures bounded phase evidence for a timed-out partial result without treating it as a validated task"},
608
+ {"file": "tests/auto-task-evidence.spec.ts", "name": "captures monotonic auxiliary phases without text, accounting or late mutation"},
602
609
  {"file": "tests/request-body-evidence.spec.ts", "name": "measures a nine MiB native-style inline image without exposing or decoding it"},
603
610
  {"file": "tests/preview-provider.spec.ts", "name": "captures final %s callback bytes and dispatch-local header time before clone observation"},
604
611
  {"file": "tests/copilot-stream-liveness.spec.ts", "name": "does not count consumer think time as provider idle"},
@@ -1,5 +1,10 @@
1
1
  # Agent readiness audit
2
2
 
3
+ > **Historical audit:** dated follow-ups preserve plans and validation blockers
4
+ > at the time, not current setup requirements or release status. Start with
5
+ > [AGENTS.md](../AGENTS.md), current README and the
6
+ > [rc.2 review](./official-first-020-rc2.md).
7
+
3
8
  Tracking: [#73](https://github.com/cloga/dsh-github-copilot/issues/73). Audited baseline: `d9d0954` (2026-09-05). This is an evidence inventory, not a blanket certification of live Copilot behavior.
4
9
 
5
10
  ## Scope and method
@@ -3,6 +3,41 @@
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:** published `0.4.0-alpha.88`, targeting official DSH/Windows
7
+ Desktop `0.2.0-rc.2`. Semantic assessment is enabled by default with bounded
8
+ auxiliary timing/output evidence. Neither publication nor
9
+ these explanations prove the version loaded in a particular Host.
10
+
11
+ ## Using Auto
12
+
13
+ Choose **Auto · Balance**, **Auto · Efficiency**, or **Auto · Intelligence**
14
+ under the managed **GitHub Copilot** provider in DSH's ordinary picker. This is
15
+ the plugin's DSH routing policy, not Copilot App's Auto or GitHub's private
16
+ router. Selecting a real model exits Auto. Every step/retry in an admitted
17
+ turn keeps its chosen model; the next turn can choose differently.
18
+
19
+ Semantic assessment is already enabled for locally unknown tasks. To opt out,
20
+ set `autoSemanticAssessment: false` in the plugin's `github-copilot` settings
21
+ namespace; it is not an extra switch on the Web search card. This changes
22
+ subsequent assessments, not an already-admitted turn. Local known-task rules
23
+ still work when auxiliary inference is disabled.
24
+
25
+ Open the reply's selection explanation for the captured decision. Evidence is
26
+ Host-lifetime and bounded; after restart or eviction it can be unavailable.
27
+ Do not infer an old decision from the current picker.
28
+
29
+ | Reading | Meaning |
30
+ |---|---|
31
+ | `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
+ | `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. |
34
+ | Missing auxiliary milestone | That milestone was not observed before assessment settled, not proof the supplier never sent data. |
35
+ | Selection unknown/unavailable | Captured selection evidence is absent or could not be read, distinct from unknown task demand. |
36
+
37
+ There is no assessment retry or automatic deadline extension. Caller cancellation
38
+ aborts the turn rather than invoking fallback. Auxiliary calls can add latency
39
+ and supplier charges outside native answer Usage.
40
+
6
41
  ## Requirements
7
42
 
8
43
  Auto must explain why the initiating turn chose its actual route, not display a
@@ -53,7 +88,7 @@ Attachments, unfamiliar instructions and genuinely ambiguous continuations can
53
88
  still remain unknown. These refinements do not change category preference,
54
89
  previous-model eligibility or semantic safeguards.
55
90
 
56
- The semantic experiment uses one fixed concrete account candidate through the
91
+ The semantic assessment uses one fixed concrete account candidate through the
57
92
  existing managed native adapter, never an Auto ID and never `ctx.llm` routing
58
93
  recursion. It has no tools, bounded text-only assessment context, bounded JSON
59
94
  output, one deadline, cancellation and no new retry loop. Text from messages is
@@ -62,7 +97,7 @@ are not read; omitted context makes classification uncertain and cannot justify
62
97
  downshifting. Assessment input is a separate projection: the actual chat
63
98
  messages, system prompt, tools, attachments and history stay untouched.
64
99
 
65
- Only locally unknown tasks invoke the enabled experiment. Context packing keeps
100
+ Only locally unknown tasks invoke the enabled assessment. Context packing keeps
66
101
  the current and nearest preceding user requests before recent output, then
67
102
  restores chronological order. At most twelve complete JSON rows, 1600 text
68
103
  characters per row and 8000 serialized characters are retained; JSON escape
@@ -78,6 +113,48 @@ instead of falling back. Credential, metadata and account revocation remain
78
113
  fatal through existing lease/admission guards. Extra calls incur supplier
79
114
  charges and latency, separate from native chat Usage and without inferred cost.
80
115
 
116
+ ## Auxiliary latency evidence and bounded output (#272)
117
+
118
+ The assessment still has one **8-second end-to-end deadline**, including account
119
+ discovery/preparation and the native call. It does not retry on timeout or try a
120
+ second classifier. The response budget is now 128 output tokens, with compact
121
+ demand JSON and at most three fixed signals requested. The strict decoder
122
+ retains its seven-signal compatibility bound; raw explanation, confidence
123
+ percentages and reasoning are not requested. Truncation remains failure, never
124
+ accepted partial JSON.
125
+
126
+ Request `off` reasoning only when both current supplier metadata advertises it
127
+ and the public prepared model lists an `off` effort. Supplier metadata alone
128
+ does not establish native support: current managed reasoning maps can decline
129
+ `off` even when advertised. In that case retain native policy, do not change
130
+ shared model maps, guess a wire value or patch Core. No claim that reasoning is
131
+ disabled for all classifiers is made.
132
+
133
+ Captured Host-lifetime selection evidence now includes a bounded public
134
+ classifier ID, configured waiting budget, total monotonic assessment elapsed
135
+ time, adapter-invocation start, first nonempty text time, native stop/non-stop
136
+ finish observation and strict result-validation status. Preparation is included
137
+ in total time; adapter start is **not HTTP dispatch**, first text is **not first
138
+ network byte**, and these milestones cannot distinguish connection, supplier
139
+ queueing or computation. Absence says no milestone was observed before
140
+ settlement, not that the provider never emitted data. Counts are characters,
141
+ not tokens/credits; no prompt, text, replay, credential, error body or hash is
142
+ captured. Late callbacks cannot rewrite a settled decision.
143
+
144
+ The existing strict explicit-Session/turn Remote carries this optional evidence;
145
+ old explanations remain decodable without it. Numeric/model evidence is copied
146
+ and frozen in the bounded store and routing decision, not durably appended or
147
+ reconstructed after restart. The footer says assessment timed out, demand
148
+ remained unknown, and preference policy selected the category as a fallback.
149
+ Progressive details show milestones separately from native answer Usage.
150
+
151
+ This is instrumentation and bounded output optimization, not a measured latency
152
+ improvement or classifier calibration. No latency-ranked classifier pool,
153
+ cross-turn assessment cache, cooldown, adaptive deadline or extra probe is
154
+ introduced without reviewing real observations. Existing local evidence,
155
+ omitted-context downshift protection, cancellation, account/proof revocation,
156
+ same-turn single-flight, eligibility and category policy remain unchanged.
157
+
81
158
  ## Category policy and continuity
82
159
 
83
160
  | Task demand | Efficiency | Balance | Intelligence |