NoteConnection v1.8.0

NoteConnection v1.8.0

English

Release Scope

• Compare baseline: v1.7.0..v1.8.0
• Release date: 2026-07-07
• Scope summary:
  • 343 implementation commits since the previous formal release, plus release metadata.
  • Major focus: Knowledge Workspace / Agent Workspace reliability, RSE document-augmented graph RAG, richer one-message answers, runtime observability, desktop sidecar hardening, CI governance, and bilingual documentation.

Breaking Changes

• No intentional breaking API change is introduced in this release.
• The Knowledge Workspace response model remains additive and forward-compatible: new RAG, trace, claim-citation, sufficiency, source-decision, and export fields are optional.
• Runtime caches, graph caches, and probe fixtures may be regenerated after upgrade; this is expected and does not require a manual data migration.

Highlights

• Upgraded Knowledge Workspace answers from narrow scoped summaries to a bounded RSE + document-augmented graph RAG pipeline.
• Added full-document-aware source augmentation for selected scoped documents while keeping model-visible context bounded by RagContextPack.
• Added graph-conditioned evidence assembly so predecessor/successor nodes contribute source-backed content, relation metadata, and intent-ranked graph context rather than title-only hints.
• Preserved the one public chat message contract while moving orchestration, scoring, recovery, source decisions, evidence roles, and claim-citation mapping into backend trace/status/export surfaces.
• Added profile-aware public answer behavior for definitions, comparisons, how-to answers, causal explanations, and generic queries.
• Hardened release review so public answers are checked against citation-backed RAG fragments, profile completeness signals, graph intent, structured contradictions, temporal qualifiers, and diagnostic leakage rules.
• Expanded runtime and CI probes to cover full-document scans, cross-document conflicts, condition-scoped evidence, graph-neighbor source loss, provider judge failure, and answer profile budgets.

Knowledge Workspace RAG and Answer Quality

• Introduced typed RAG evidence contracts:
  • RagEvidenceFragment
  • RagContextPack
  • RagContextBudget
  • RagSufficiencyReview
  • source decisions and failure-stage classifications
• Added source-window and document augmentation assembly:
  • starts from precise evidence spans;
  • reads complete selected source documents when provenance allows;
  • emits direct support, parent context, adjacent context, graph-neighbor support, conflict, background, and degraded source-window states;
  • deduplicates repeated/overlapping windows and preserves line/offset provenance.
• Added budgeted context packing:
  • role-prioritized fragment selection;
  • per-fragment and total-character caps;
  • traceable include/truncate/drop decisions;
  • head/tail-preserving truncation for oversized evidence.
• Added deterministic and optional provider-backed sufficiency judging:
  • deterministic gates run first;
  • optional LLM judge reuses the existing NoteMD provider boundary;
  • malformed JSON, timeout, and provider failure fall back deterministically;
  • recovery is bounded to one assembly/review pass.
• Added richer public answer composition:
  • direct/document/graph evidence budgets are selected by answer profile;
  • public output strips source-authored scaffolding labels such as Prerequisite:, Mechanism:, and Failure mode: while preserving meaningful step labels;
  • compare answers rank evidence by operands and graph intent, and do not promote procedure-like evidence unless the query is about procedures;
  • how-to answers preserve steps, prerequisites, downstream checks, and failure handling;
  • causal answers preserve mechanism, consequence, and boundary evidence.
• Added release-review completeness improvements:
  • rag_answer_completeness now checks required RAG roles and profile-specific signals;
  • drafts missing pack-backed prerequisites, failure handling, mechanisms, consequences, or boundaries are revised from the RAG pack;
  • rag_claim_citation_support contracts unsupported or weakly supported public claims.

Conflict, Qualifier, and Graph Guardrails

• Expanded controlled conflict detection across:
  • measurements and tolerances;
  • dates and release dates;
  • state/status values;
  • quantity and retry limits;
  • location facts;
  • ownership identity facts;
  • endpoint, dependency, format, protocol, semantic-version, service-port, and response-status-code facts.
• Added false-positive guards for scoped qualifiers:
  • temporal current-vs-historical facts;
  • environment-scoped facts;
  • version-scoped facts;
  • platform-scoped facts;
  • cross-document condition-scoped facts.
• Added graph-intent ranking and filtering:
  • comparison queries prefer analogy/contrast neighbors over procedural sequence edges;
  • graph neighbors outside the resolved conversation scope are filtered before evidence assembly;
  • in-scope graph-neighbor source-window failures degrade explicitly instead of silently becoming complete graph support.
• Added runtime probe fixtures under Knowledge_Base/ for:
  • full-document remote-scan hard negatives;
  • cross-document conflicts;
  • temporal/environment/version/platform qualifiers;
  • graph intent;
  • context overflow;
  • repeated snippet anchoring;
  • causal answer profile budgets.

Agent Workspace and Frontend UX

• Added and hardened Agent Workspace frontend runtime surfaces:
  • agent_workspace.js
  • agent_workspace_runtime.js
  • workspace_panes.js
  • localized en / zh status strings
  • modular runtime bridge files and state modules
• Improved Knowledge Workspace visibility for API/RAG state:
  • compact evidence status;
  • provider/runtime readiness indicators;
  • RAG sufficiency and degradation summaries;
  • trace-friendly answer/evidence panes without adding extra chat messages.
• Preserved knowledge-point-oriented evidence display:
  • same knowledge point hits are grouped instead of duplicated;
  • public answers stay natural while trace/export surfaces retain raw evidence details.
• Added frontend contract and runtime behavior tests covering Agent Workspace API parity, runtime behavior, localization, and UI contract surfaces.

Learning Platform, Storage, and Export

• Expanded the learning platform into typed domain boundaries:
  • KnowledgeIngestor
  • KnowledgeQuerier
  • ConversationManager
  • MasteryEngine
  • QualityEvaluator
  • TutorRouter
  • MemoryPolicyManager
• Added or strengthened:
  • knowledge store and persistence tests;
  • query backend tests and external HTTP integration coverage;
  • vector acceleration adapter contracts;
  • runtime capability detection;
  • memory governance;
  • workflow artifact store;
  • resource and workspace registries;
  • indexing lifecycle and segment/unit builders.
• Added answer grounding export support:
  • final public answer claims map to citation ids, RAG fragment ids, source paths, and support status;
  • workspace export bundles deep-clone trace mappings so replay/export consumers can audit grounding without mutable runtime references.

NoteMD, Markdown, and Reader Runtime

• Expanded NoteMD operation surfaces:
  • CLI parser/dispatcher/commands;
  • operation registry and capability manifest;
  • config profile commands;
  • provider diagnostics, provider templates, and provider profiles;
  • Tavily and DuckDuckGo search provider adapters.
• Added diagram-generation planning and response parsing for NoteMD workflows.
• Hardened Mermaid and Markdown handling:
  • Mermaid fence guardrails;
  • browser Mermaid runtime embedding;
  • Resvg runtime embedding;
  • reader renderer coverage;
  • reader and Markdown runtime frontend updates.
• Added regression and contract tests for NoteMD API, CLI, workflow, batch, diagram, app-config, provider templates, and server integration behavior.

Desktop, Tauri, Godot, Path Mode, and Multi-Platform Runtime

• Hardened Tauri sidecar handling:
  • sidecar cleanup;
  • sidecar supply readiness;
  • sidecar validation;
  • sidecar relaunch contracts;
  • sidecar signatures and replacement boundaries;
  • Godot sidecar bootstrap and mirror checks.
• Added Agent Workspace Tauri verification scripts:
  • browser runtime verification;
  • Tauri runtime verification;
  • Tauri window evidence capture;
  • Tauri Rust strict checks;
  • evidence index, manifest, summary, and release-fragment rendering.
• Strengthened PathBridge and runtime transport boundaries:
  • strict schema verification;
  • handshake contracts;
  • runtime port fallback contracts;
  • runtime heap/spool/transport policies;
  • request trace middleware and modular route registration.
• Updated Godot / Path Mode surfaces:
  • reader render client;
  • path renderer;
  • settings panel;
  • WebSocket client;
  • path mode UI and shader updates.
• Expanded Android / Capacitor / mobile governance:
  • Android path mode contracts and smoke checks;
  • Detox pipeline verification;
  • privacy manifest verification;
  • Capacitor runtime, device, bridge, storage, and evidence contracts.

WASM, Performance, and Foundation Runtime

• Added or strengthened WASM parity runtime gates:
  • artifact provisioning;
  • runtime functional checks;
  • output equivalence;
  • benchmark contracts;
  • benchmark guard contracts;
  • history gates.
• Added foundation runtime verification:
  • SQLite runtime contracts and matrix/soak scripts;
  • ANN runtime contracts and release evidence scripts;
  • release evidence verification and multi-host readiness checks.
• Added platform utilities:
  • platform capability detection;
  • render materializer;
  • export profiles;
  • runtime path resilience.

CI, Release Governance, and Repository Operations

• Hardened GitHub Actions workflows:
  • Docs Diataxis Site;
  • Docs GitHub Pages Publish;
  • Fixrisk Operational Readiness;
  • Migration Gates;
  • Mobile E2E Detox Contracts;
  • npm publish;
  • desktop multi-OS release;
  • version check;
  • WASM parity benchmark snapshots.
• Added release and compliance guardrails:
  • license policy contracts;
  • SBOM policy and attestation contracts;
  • LFS asset policy checks;
  • sidecar signature contracts;
  • core real-machine verification helpers.
• Added Trellis task/workflow scaffolding and repository process documentation for structured work tracking.
• Updated AGENTS.md, CI status docs, task/walkthrough docs, and bilingual governance references.

Documentation

• Added and expanded bilingual Diataxis documentation:
  • development progress dashboards;
  • architecture and migration;
  • release and governance;
  • multi-platform build flows;
  • Git LFS asset migration;
  • sidecar supply feasibility;
  • Android release build;
  • Godot sidecar bootstrap;
  • GitHub Pages publish flow;
  • Agent conversation focus mode plan;
  • DeepTutor and memo reuse assessments.
• Added long-form implementation and solution plans:
  • RSE document-augmented graph RAG answer pipeline;
  • agent final-reply review robustness;
  • agent knowledge DAG answer contract;
  • graph preview and review closure;
  • architecture progress alignment;
  • knowledge workspace DAG alignment.
• Expanded bilingual task, implementation plan, walkthrough, README, test report, interface, sidecar, LFS, and multi-platform build docs.

Version Metadata

• Aligned release metadata to 1.8.0 in:
  • package.json
  • package-lock.json
  • src-tauri/tauri.conf.json

Verification

Local verification before release:

• npm test -- --runInBand
• npm run build
• npm run docs:diataxis:check
• npm run docs:site:build
• node scripts/verify-knowledge-workspace-runtime.js --full --case causal_answer_profile_budget_en --case graphintent_compare_neighbor_selection_en

Remote main verification for the pre-release head:

• Docs Diataxis Site: success
• Docs GitHub Pages Publish: success
• Mobile E2E Detox Contracts: success
• Fixrisk Operational Readiness: success
• Migration Gates: success

Known Follow-Up Work

• Broader profile classes and corpus-calibrated answer budgets.
• Larger replay/runtime corpora for weakly supported but lexically similar claims.
• Deeper natural-language synthesis beyond deterministic source-label stripping.
• Continued calibration of graph-confidence thresholds and semantic conflict classes.

GitHub Release Body Template

Use this section as the GitHub Release body:

1. Summary:
   • NoteConnection v1.8.0 upgrades Knowledge Workspace into a bounded RSE + document-augmented graph RAG system with richer one-message answers, stronger release review, and much deeper runtime observability.
2. Major user-facing changes:
   • Source-backed answers now use precise spans, full selected source documents, graph-neighbor evidence, sufficiency review, and profile-aware answer composition.
   • The public chat still returns one answer message; evidence roles, source decisions, recovery, and claim-citation mappings stay in trace/status/export surfaces.
   • Compare, how-to, causal, definition, and generic answers now use different deterministic budgets and completeness gates.
3. Runtime and platform:
   • Agent Workspace frontend, Tauri sidecar handling, PathBridge contracts, Godot/Path Mode integration, mobile contracts, and WASM/foundation gates are substantially hardened.
4. Documentation and governance:
   • Bilingual Diataxis docs, implementation plans, release governance, CI gates, SBOM/license/LFS checks, and operational scripts are synchronized for the new runtime architecture.
5. Compatibility:
   • No intentional breaking API change; new response, trace, RAG, and export fields are additive.

---

中文

发布范围

• 对比基线：v1.7.0..v1.8.0
• 发布日期：2026-07-07
• 范围摘要：
  • 上一个正式版本之后累计 343 个实现提交，外加本次发布元数据。
  • 本次重点：Knowledge Workspace / Agent Workspace 可靠性、RSE document-augmented graph RAG、更充分的单消息回答、运行时可观测性、桌面 sidecar 加固、CI 治理与双语文档。

破坏性变更

• 本版本没有引入有意的破坏性 API 变更。
• Knowledge Workspace 响应模型保持增量兼容：新增 RAG、trace、claim-citation、sufficiency、source-decision 与 export 字段均为 optional。
• 升级后运行时 cache、graph cache 与探针 fixture 可能会重新生成，这是预期行为，不需要手工数据迁移。

版本亮点

• 将 Knowledge Workspace 回答从窄 scoped summary 升级为有界的 RSE + document-augmented graph RAG 链路。
• 对被选 scoped 文档启用完整文档感知的 source augmentation，同时用 RagContextPack 限制模型可见上下文。
• 引入 graph-conditioned evidence assembly，使前驱/后继节点贡献可追溯源内容、关系元数据与按意图排序的图上下文，而不是只展示标题。
• 保持“聊天区只显示一条公开回答”的产品契约，把编排、评分、恢复、source decisions、evidence roles 与 claim-citation mapping 放入后端 trace/status/export surface。
• 为 definition、compare、how-to、causal 与 generic 查询加入 profile-aware 公开回答策略。
• 强化 release review，使公开回答接受 citation-backed RAG fragment、profile completeness signals、graph intent、结构化矛盾、时序限定与诊断泄漏规则的检查。
• 扩展 runtime 与 CI 探针，覆盖 full-document scan、跨文档冲突、条件限定证据、图邻居 source loss、provider judge 失败与 answer profile budget。

Knowledge Workspace RAG 与回答质量

• 引入类型化 RAG 证据契约：
  • RagEvidenceFragment
  • RagContextPack
  • RagContextBudget
  • RagSufficiencyReview
  • source decisions 与 failure-stage classifications
• 新增 source-window / document augmentation assembly：
  • 从精确 evidence span 出发；
  • provenance 允许时读取被选源文档完整内容；
  • 产出 direct support、parent context、adjacent context、graph-neighbor support、conflict、background 与 degraded source-window state；
  • 去重重复/重叠窗口，并保留 line / offset provenance。
• 新增有界 context packing：
  • 按 evidence role 优先级选 fragment；
  • 片段级和总字符数上限；
  • 可追踪 include / truncate / drop decision；
  • 对超大 evidence 采用保留首尾的截断策略。
• 新增确定性与可选 provider-backed sufficiency judging：
  • deterministic gate 优先运行；
  • 可选 LLM judge 复用现有 NoteMD provider 边界；
  • malformed JSON、timeout 和 provider failure 都会回退到确定性路径；
  • recovery 被限制为一次 assembly / review pass。
• 新增更充分的公开回答组织：
  • direct / document / graph evidence budget 按 answer profile 选择；
  • 公开输出会剥离 Prerequisite:、Mechanism:、Failure mode: 等源文档结构标签，同时保留有意义的步骤标签；
  • compare 回答按 operand 与 graph intent 排序，除非 query 本身讨论 procedure，否则不会把 procedure-like evidence 升到公开回答；
  • how-to 回答会保留 steps、prerequisites、downstream checks 与 failure handling；
  • causal 回答会保留 mechanism、consequence 与 boundary evidence。
• 强化 release-review completeness：
  • rag_answer_completeness 现在同时检查 required RAG roles 和 profile-specific signals；
  • draft 如果遗漏 pack 中已有的 prerequisites、failure handling、mechanisms、consequences 或 boundaries，会从 RAG pack 修订；
  • rag_claim_citation_support 会收缩 unsupported 或 weakly supported 的公开 claim。

冲突、限定条件与图护栏

• 扩展受控 conflict detection，覆盖：
  • measurement / tolerance；
  • date / release date；
  • state / status；
  • quantity / retry limit；
  • location；
  • ownership identity；
  • endpoint、dependency、format、protocol、semantic-version、service-port 与 response-status-code。
• 新增限定条件 false-positive guard：
  • temporal current-vs-historical facts；
  • environment-scoped facts；
  • version-scoped facts；
  • platform-scoped facts；
  • cross-document condition-scoped facts。
• 新增 graph-intent ranking 与 filtering：
  • compare 查询会优先 analogy / contrast 邻居，而不是 procedural sequence edge；
  • resolved conversation scope 之外的 graph neighbor 会在 evidence assembly 前过滤；
  • scope 内 graph-neighbor source-window 缺失会显式降级，不会被静默当作完整 graph support。
• 在 Knowledge_Base/ 中新增 runtime probe fixture：
  • full-document remote-scan hard negative；
  • cross-document conflict；
  • temporal / environment / version / platform qualifier；
  • graph intent；
  • context overflow；
  • repeated snippet anchoring；
  • causal answer profile budget。

Agent Workspace 与前端体验

• 新增并加固 Agent Workspace 前端运行时 surface：
  • agent_workspace.js
  • agent_workspace_runtime.js
  • workspace_panes.js
  • 本地化 en / zh 状态文案
  • 模块化 runtime bridge 与 state module
• 增强 Knowledge Workspace 对 API / RAG 状态的可见性：
  • compact evidence status；
  • provider / runtime readiness indicator；
  • RAG sufficiency 与 degradation summary；
  • trace-friendly answer / evidence pane，且不会追加额外聊天消息。
• 保持 knowledge-point-oriented evidence display：
  • 同一知识点的命中会聚合而不是重复展示；
  • 公开回答保持自然语言，trace / export surface 保留原始 evidence 细节。
• 新增前端 contract 与 runtime behavior tests，覆盖 Agent Workspace API parity、运行时行为、本地化与 UI contract surface。

Learning Platform、存储与导出

• 将 learning platform 拆分到更清晰的 typed domain boundary：
  • KnowledgeIngestor
  • KnowledgeQuerier
  • ConversationManager
  • MasteryEngine
  • QualityEvaluator
  • TutorRouter
  • MemoryPolicyManager
• 新增或强化：
  • knowledge store 与 persistence tests；
  • query backend tests 与 external HTTP integration coverage；
  • vector acceleration adapter contracts；
  • runtime capability detection；
  • memory governance；
  • workflow artifact store；
  • resource / workspace registry；
  • indexing lifecycle 与 segment / unit builders。
• 新增 answer grounding export：
  • 最终公开回答 claim 会映射到 citation id、RAG fragment id、source path 与 support status；
  • workspace export bundle 会 deep-clone trace mapping，方便 replay / export 消费者审计 grounding，避免共享可变运行时引用。

NoteMD、Markdown 与 Reader Runtime

• 扩展 NoteMD 操作面：
  • CLI parser / dispatcher / commands；
  • operation registry 与 capability manifest；
  • config profile commands；
  • provider diagnostics、provider templates 与 provider profiles；
  • Tavily / DuckDuckGo search provider adapters。
• 为 NoteMD workflow 新增 diagram-generation planning 与 response parsing。
• 加固 Mermaid 与 Markdown：
  • Mermaid fence guardrails；
  • browser Mermaid runtime embedding；
  • Resvg runtime embedding；
  • reader renderer coverage；
  • reader 与 Markdown runtime 前端更新。
• 新增 NoteMD API、CLI、workflow、batch、diagram、app-config、provider templates 与 server integration 回归/契约测试。

Desktop、Tauri、Godot、Path Mode 与多平台运行时

• 加固 Tauri sidecar：
  • sidecar cleanup；
  • sidecar supply readiness；
  • sidecar validation；
  • sidecar relaunch contracts；
  • sidecar signatures 与 replacement boundaries；
  • Godot sidecar bootstrap 与 mirror checks。
• 新增 Agent Workspace Tauri 验证脚本：
  • browser runtime verification；
  • Tauri runtime verification；
  • Tauri window evidence capture；
  • Tauri Rust strict checks；
  • evidence index / manifest / summary / release-fragment rendering。
• 强化 PathBridge 与 runtime transport 边界：
  • strict schema verification；
  • handshake contracts；
  • runtime port fallback contracts；
  • runtime heap / spool / transport policies；
  • request trace middleware 与 modular route registration。
• 更新 Godot / Path Mode：
  • reader render client；
  • path renderer；
  • settings panel；
  • WebSocket client；
  • path mode UI 与 shader。
• 扩展 Android / Capacitor / mobile 治理：
  • Android path mode contracts 与 smoke checks；
  • Detox pipeline verification；
  • privacy manifest verification；
  • Capacitor runtime、device、bridge、storage 与 evidence contracts。

WASM、性能与 Foundation Runtime

• 新增或强化 WASM parity runtime gates：
  • artifact provisioning；
  • runtime functional checks；
  • output equivalence；
  • benchmark contracts；
  • benchmark guard contracts；
  • history gates。
• 新增 foundation runtime verification：
  • SQLite runtime contracts 与 matrix / soak scripts；
  • ANN runtime contracts 与 release evidence scripts；
  • release evidence verification 与 multi-host readiness checks。
• 新增平台工具：
  • platform capability detection；
  • render materializer；
  • export profiles；
  • runtime path resilience。

CI、发布治理与仓库运营

• 加固 GitHub Actions workflows：
  • Docs Diataxis Site；
  • Docs GitHub Pages Publish；
  • Fixrisk Operational Readiness；
  • Migration Gates；
  • Mobile E2E Detox Contracts；
  • npm publish；
  • desktop multi-OS release；
  • version check；
  • WASM parity benchmark snapshots。
• 新增 release / compliance guardrails：
  • license policy contracts；
  • SBOM policy 与 attestation contracts；
  • LFS asset policy checks；
  • sidecar signature contracts；
  • core real-machine verification helpers。
• 新增 Trellis task / workflow scaffolding 与仓库流程文档，用于结构化工作跟踪。
• 更新 AGENTS.md、CI status、task / walkthrough 与双语治理文档。

文档

• 新增并扩展双语 Diataxis 文档：
  • development progress dashboards；
  • architecture and migration；
  • release and governance；
  • multi-platform build flows；
  • Git LFS asset migration；
  • sidecar supply feasibility；
  • Android release build；
  • Godot sidecar bootstrap；
  • GitHub Pages publish flow；
  • Agent conversation focus mode plan；
  • DeepTutor / memo reuse assessments。
• 新增长篇 implementation / solution plans：
  • RSE document-augmented graph RAG answer pipeline；
  • agent final-reply review robustness；
  • agent knowledge DAG answer contract；
  • graph preview and review closure；
  • architecture progress alignment；
  • knowledge workspace DAG alignment。
• 扩展双语 task、implementation plan、walkthrough、README、test report、interface、sidecar、LFS 与 multi-platform build docs。

版本元数据

• 以下文件版本已统一到 1.8.0：
  • package.json
  • package-lock.json
  • src-tauri/tauri.conf.json

验证

本地发布前验证：

• npm test -- --runInBand
• npm run build
• npm run docs:diataxis:check
• npm run docs:site:build
• node scripts/verify-knowledge-workspace-runtime.js --full --case causal_answer_profile_budget_en --case graphintent_compare_neighbor_selection_en

发布前远端 main 验证：

• Docs Diataxis Site: success
• Docs GitHub Pages Publish: success
• Mobile E2E Detox Contracts: success
• Fixrisk Operational Readiness: success
• Migration Gates: success

后续工作

• 更广 answer profile 类和基于语料校准的回答预算。
• 更大 replay / runtime corpus，用于覆盖“词面相近但支撑不足”的 claim。
• 超出确定性 source-label stripping 的更深自然语言 synthesis。
• 继续校准 graph-confidence threshold 与 semantic conflict class。

GitHub Release 总结

1. 概要：
   • NoteConnection v1.8.0 将 Knowledge Workspace 升级为有界 RSE + document-augmented graph RAG 系统，提供更充分的单消息回答、更强 release review 和更完整运行时可观测性。
2. 主要用户可见变化：
   • 回答现在基于 precise spans、完整被选源文档、graph-neighbor evidence、sufficiency review 与 profile-aware answer composition。
   • 聊天区仍只返回一条公开回答；evidence roles、source decisions、recovery 与 claim-citation mappings 保留在 trace/status/export surface。
   • compare、how-to、causal、definition 与 generic 回答现在使用不同确定性预算与 completeness gates。
3. 运行时与平台：
   • Agent Workspace 前端、Tauri sidecar、PathBridge contract、Godot / Path Mode、mobile contracts 与 WASM / foundation gates 均显著加固。
4. 文档与治理：
   • 双语 Diataxis 文档、实现方案、发布治理、CI gates、SBOM / license / LFS checks 与运维脚本已同步到新的运行时架构。
5. 兼容性：
   • 无有意破坏性 API 变更；新增 response、trace、RAG 与 export 字段均为增量 optional。