obsidian-plugin/jacobinwwey_obsidian-NotEMD
1
0
Fork 0
mirror of https://github.com/jacobinwwey/obsidian-NotEMD.git synced 2026-07-22 05:48:27 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions
jacobinwwey_obsidian-NotEMD/change.md
aliyun1121003339 a996688ed5 fix(release): make validation portable 
Keep CircuitikZ filesystem fixtures host-native, preserve Node in CLI fallback PATH tests, and own the website frontmatter parser at the root test boundary so Linux release runners match local verification.
2026-07-21 23:03:06 +08:00

78 KiB
Raw Permalink Blame History

Notemd Change Log

This document summarizes the major functional and architectural changes implemented.

1.9.4 - 2026-07-21

English

Highlights

  • Expanded diagrams into a source-aware, multi-format platform with explicit CircuitikZ, Draw.io, Drawnix, SVG, PNG, PDF, HTML, and editable artifact contracts.
  • Added a managed CircuitikZ desktop environment with trusted compiler discovery, pinned Tectonic installation, integrity verification, native compile diagnostics, and dependency-free fallback previews.
  • Added Vault-persistent diagram history with search, filtering, pagination, artifact actions, direct navigation, and an adaptive right-side drawer.
  • Added settings fuzzy search, stable favorites, and section navigation, plus safer batch-folder confirmation and creation behavior.
  • Completed published multilingual documentation routes and hardened Slidev standalone export and GitHub Pages delivery.

Changes

  • Added first-class experimental diagram UI support for circuit intent and the circuitikz render target.
  • Added CircuitikzRenderer for constrained DiagramSpec.circuitSpec payloads. It writes deterministic circuitikz .tex source and attaches a white-background SVG preview companion for Obsidian viewing.
  • Extended the offline diagram artifact CLI so --target circuitikz can write TeX plus SVG/PNG/PDF preview companions. Direct svg, png, and pdf targets can also render validated circuit specs through the same companion SVG path.
  • Hardened the artifact CLI for npm 11 argument rewriting on Windows by accepting the ordered positional fallback shape, and made the circuitikz SVG canvas background inline white so standalone viewers do not show transparent SVGs as black.
  • Kept the circuit path topology-first: the prompt asks for structured CircuitSpec, validation rejects free-form/non-circuit payloads, and the preview companion is documented as review evidence rather than a LaTeX/TikZJax compile result.
  • Updated the GitHub Pages diagram documentation and maintainer runbooks. The MDX synchronization policy is explicit: do not blindly commit every generated locale file for small feature increments; commit localized MDX only when published route/frontmatter/heading or visible behavior parity requires it.
  • Added an optional managed CircuitikZ desktop runtime. Notemd can discover trusted system Tectonic/pdfLaTeX executables, install or repair a pinned Tectonic build outside the Vault with integrity checks, remove only the managed installation, and report smoke-test capabilities while dependency-free previews remain available on mobile.
  • Fixed long diagram-history drawers so the flex container can shrink, the history body scrolls within the preview shell, and overscroll stays contained.
  • Updated the maintainer CLI bridge to prefer obsidian-cli native eval when that wrapper is installed and fall back to the official obsidian eval command only when the wrapper command is unavailable.
  • Synchronized the npm 10 lockfile used by GitHub Actions and made CircuitikZ, CLI fallback, and website documentation contract fixtures portable across Windows and Linux release runners.

中文

亮点

  • 将图表能力扩展为识别源类型的多格式平台，为 CircuitikZ、Draw.io、Drawnix、SVG、PNG、PDF、HTML 与可编辑产物建立明确契约。
  • 新增 CircuitikZ 桌面端托管环境，支持可信编译器发现、固定版本 Tectonic 安装、完整性校验、原生编译诊断与零依赖预览回退。
  • 新增按 Vault 持久化的图形历史，支持搜索、筛选、分页、产物操作、直达入口与自适应右侧抽屉。
  • 新增设置模糊搜索、稳定收藏与大项导览，并完善批处理目标文件夹的确认及创建逻辑。
  • 补齐已发布的多语言文档路由，并加固 Slidev standalone 导出与 GitHub Pages 交付。

变更

  • 为实验性图表 UI 增加一等 circuit intent 与 circuitikz render target 支持。
  • 新增 CircuitikzRenderer，用于受约束的 DiagramSpec.circuitSpec payload。它会写出确定性 circuitikz .tex 源文件，并附加白底 SVG preview companion 供 Obsidian 查看。
  • 扩展离线 diagram artifact CLI，使 --target circuitikz 可以写出 TeX 以及 SVG/PNG/PDF 预览 companion。直接 svg、png 与 pdf target 也可以通过同一条 companion SVG 路径渲染已验证电路 spec。
  • 加固 artifact CLI，使其兼容 Windows 上 npm 11 参数重写后的有序位置参数形态；同时将 circuitikz SVG 画布背景改为 inline 白色，避免 standalone viewer 把透明 SVG 显示成黑底。
  • 电路路径继续保持 topology-first：prompt 要求结构化 CircuitSpec，校验拒绝自由形式/非电路 payload，并且文档明确 preview companion 是审查证据，不是 LaTeX/TikZJax 编译结果。
  • 更新 GitHub Pages 图表文档与维护者 runbook。MDX 同步策略已明确：小功能增量不要盲目提交所有 generated locale 文件；只有公开 route/frontmatter/heading 或可见行为需要对齐时，才提交对应 localized MDX。
  • 新增可选的 CircuitikZ 桌面端托管运行时。Notemd 可以发现受信任的系统 Tectonic/pdfLaTeX 可执行文件，在 Vault 之外通过完整性校验安装或修复固定版本的 Tectonic，仅删除托管安装，并报告 smoke-test 能力；移动端仍可使用零依赖预览。
  • 修复较长图形历史抽屉的滚动：flex 容器现在可以收缩，历史正文在预览 shell 内滚动，并限制 overscroll 外溢。
  • 更新维护者 CLI bridge：安装了 obsidian-cli wrapper 时优先使用 obsidian-cli native eval；仅当 wrapper 命令不可用时，回退到官方 obsidian eval 命令。
  • 同步 GitHub Actions 使用的 npm 10 lockfile，并使 CircuitikZ、CLI fallback 与网站文档契约 fixture 可在 Windows 和 Linux 发布 runner 上稳定执行。

1.9.3

Highlights

  • Slidev export is now a report-gated multi-format workflow covering HTML, standalone HTML, PDF, PNG, and PPTX from the same prepared deck source.
  • PPTX export now emits visible native editable text and native table-cell text instead of transparent selectable overlays.
  • The reported PPTX page 17-29 occlusion issue is fixed by deterministic background, shape, table, and text layer ordering.

New Features

  • Added environment and output reporting for the real Slidev export workflow, including local Slidev fork detection, full Slidev skill root loading, and per-format capability checks.
  • Added standalone HTML bundle support with offline file:// behavior, relative asset handling, and Vue component transformation fixes.
  • Added editable PPTX export contracts for visible native text, rich text runs, font policy, paragraph spacing, native tables, table-cell rich text, code backgrounds, and rendered-HTML visual reference comparison.
  • Added verifier metrics for fallback-only element kinds, transparent overlay removal, table-owned decoration ownership, and PPTX XML layer-order regressions.

Fixes

  • Fixed PPTX code-background rectangles and decorative native shapes so they render below visible text instead of covering editable text.
  • Removed transparent text from the default PPTX editability path; the displayed text is now the editable text.
  • Fixed table rich-text preservation and coverage reporting so native table output is independently auditable.
  • Fixed table overlay leak reporting so suppressed table-cell overlays do not produce false-positive leak counts.

Verification

  • Revalidated with real docs/architecture.zh-CN.md Slidev PPTX export through the local jacob-user Playwright/Slidev environment, followed by a direct PPTX XML scan proving zero code-background-after-text, native-shape-after-text, transparent-alpha, or table-overlay-text matches on slides 17-29.
  • Release gates: npm run chronicle:sync-repo-saga, npm run chronicle:update -- --tag 1.9.3, npm run build, npm test -- --runInBand, npm run audit:i18n-ui, npm run audit:render-host, obsidian help, obsidian-cli help, and git diff --check.

1.9.2

Highlights

  • Sidebar footer scrolling and API observability styling are restored, so request activity no longer crowds out log output in the shipped sidebar layout.
  • Local knowledge inspect now exposes bounded query diagnostics, including low-signal navigation-note cautions for index.* and related basename-derived queries.
  • Maintainer CLI docs and examples now use the correct vault-relative path contract, and Chapter Split + TOC now has dedicated showcase documentation.

New Features

  • Added dedicated Chapter Split + TOC spotlight docs, docs-home entry points, and showcase artwork so the feature is now part of the repository’s visible shipped surface instead of only maintainer-facing references.
  • Extended local-knowledge.inspect explainability with queryDiagnostics, including the derived basename, bounded stripped-source length used for diagram-source derivation, and cautions for generic navigation-like source names.
  • Expanded maintainer CLI help and capability-matrix guidance so --vault docs examples consistently demonstrate vault-relative sourcePath and knowledgePaths inputs.
  • Added maintainer inspect examples for broader docs-vault local knowledge checks, covering explicit research queries, cross-folder batch-title source paths, and diagram-source retrieval with bounded topK / slidingWindowSize overrides.

Fixes

  • Fixed the sidebar footer container so the log/API region can scroll vertically again instead of getting pinned behind oversized activity content.
  • Fixed the maintainer docs-vault examples that previously used repo-relative docs/... paths, which could mislead local knowledge inspection and chapter-split invocations in a real --vault docs run.
  • Fixed Stage-C documentation drift by keeping local knowledge fixture coverage, maintainer helper output, and bilingual CLI capability docs aligned around the same vault-relative example payloads.
  • Fixed chronicle refresh helper authorship so release-driven chronicle updates preserve maintainer identity instead of falling back to a bot identity.

Verification

  • Revalidated with serial npm run chronicle:sync-repo-saga / npm run chronicle:update -- --tag 1.9.2, npm run build, npm test -- --runInBand, npm run audit:i18n-ui, npm run audit:render-host, obsidian help, obsidian-cli help, and git diff --check before cutting the release.

1.9.1

Highlights

  • Provider settings now converge on a cleaner required-fields-first surface while preserving saved advanced overrides and discovery metadata behavior.
  • Fetch-model-list now covers a broader bounded set of provider catalog shapes and keeps discovered output-token ceilings scoped to provider overrides instead of global Max tokens.
  • Release-facing truth is synchronized for the full 1.9.0..1.9.1 delta, including bilingual release notes, welcome digest, version markers, and chronicle references.

New Features

  • Expanded bounded provider model discovery across more OpenAI-compatible endpoint families, additional registry envelopes, continuation tokens, and hosted-provider-specific catalog shapes.
  • Added clearer provider settings guidance so model-aware token notices distinguish provider output-token overrides from the global runtime Max tokens budget.
  • Preserved provider-scoped maxOutputTokens autofill as the discovery lane while keeping manual typed-model edits responsible for global model-aware token guidance.
  • Extended Stage-C local knowledge-base routing and inspect/evaluation semantics so task-scoped retrieval sources and failure-state visibility stay explicit.
  • Deepened the diagram semantic-verification lane and kept packaging/runtime boundary checks fail-closed on current main.

Fixes

  • Fixed discovered-model apply so it no longer silently rewrites the global Max tokens field when the real change belongs to the active provider override.
  • Fixed unresolved discovered-model ceilings so an existing valid provider override is preserved instead of being cleared or degraded.
  • Fixed release-modal recent update content so the latest two entries now match the shipped version boundary and localized text bundle.
  • Fixed advanced provider-panel persistence so saved advanced overrides still surface after reopening settings.
  • Realigned progress/planning docs with the actual shipped provider-settings architecture rather than the older global token-sync wording.

Verification

  • Revalidated with npm run build, npm test -- --runInBand, npm run audit:i18n-ui, npm run audit:render-host, obsidian help, obsidian-cli help, local Obsidian hot reload/runtime inspection, and git diff --check before cutting the release.

1.9.0

Highlights

  • Provider settings now converge on a metadata-driven core/advanced split, and supported providers can fetch transient model suggestions without creating a second persisted provider catalog.
  • Developer-gated advanced batch folder selection now supports saved profiles, one-run overrides, preset chips, and inline file preview before execution.
  • Developer-gated relaxed input file restrictions now land as a shared runtime contract for selected source-preserving tasks, including PDF text extraction through Obsidian's PDF runtime.
  • Release-facing version truth, welcome digest, release notes, and chronicle refresh are synchronized at the 1.9.0 boundary.

New Features

  • Added provider-field taxonomy metadata (core, contextual, advanced, developer) so the provider settings panel can render required/default-visible controls separately from secondary tuning controls.
  • Added lightweight in-settings model discovery for a broader bounded set of official providers: selected OpenAI-compatible /models presets (including LM Studio plus Qwen/Doubao/Moonshot/GLM/SiliconFlow/Groq/Fireworks/Nebius/Cerebras/OpenRouter/Requesty families), Together's dedicated /models response shape, Anthropic GET /models, Ollama tags, and Google Gemini model listing. Discovery results are transient suggestions only; persisted provider state still uses the existing manual model string.
  • Extended the generic OpenAI Compatible preset so known official hosts can also reuse upstream provider token-cap metadata for bare model IDs. That keeps Fetch model list, provider output-token autofill, and model-aware guidance aligned even when users choose plain IDs such as gpt-4o, qwen3-coder-plus, or mimo-v2.5-pro instead of provider-prefixed gateway IDs.
  • Expanded the bounded OpenAI-compatible preset set with LiteLLM, Nebius, Cerebras, Hugging Face, and Vercel AI Gateway, while keeping discovery support honest per endpoint family rather than over-claiming generic /models parity. Hugging Face now participates in the shared bounded /models discovery path, while Azure OpenAI remains manual-first because deployment names are required.
  • Upgraded Vercel AI Gateway model discovery from a single-source registry lookup to a bounded dual-source merge of the official /v1/models catalog plus v3/ai/config, so fetch-model-list stays usable even when one gateway catalog surface is temporarily incomplete or unavailable.
  • Extended fetch-model-list coverage with OpenRouter's bounded chat + embedding catalog merge, LiteLLM's bounded /models + /model/info merge, Huawei Cloud MaaS's dedicated v2/models registry endpoint, bounded paginated traversal for Anthropic and Google model catalogs, and broader OpenAI-compatible schema parsing for models[], list/items, object-shaped catalogs, model_id, and endpoint-type-style responses.
  • Extended the transient model-discovery parser again so fetched rows can preserve display labels plus max-output-token hints when providers expose them, and so broader registries can be filtered using capability/modality/status metadata instead of only brittle model-name heuristics. This reduces audio/image/unavailable entries leaking into text-generation model suggestions while preserving manual model entry as the fallback path.
  • Expanded that bounded parser once more so fetch model list can also consume hosted types classifiers, nested data/result catalogs, and LiteLLM-style litellm_params + model_info payloads, which keeps real-world gateway registries and proxy metadata aligned with the in-settings model picker and token guidance.
  • Expanded fetch model list again through the shared discovery core instead of more provider-specific branches: bounded OpenAI-compatible registry traversal now also follows continuation signals such as next_url, links.next, nextPageToken, and next_cursor, preserves already-fetched valid pages when a later page fails, and accepts wider payload keys such as rows/records, value/values, uid, identifier, modelId, and provider_model_id.
  • Expanded that shared discovery core once more so fetch model list also tolerates publisherModels-style registries and resource-shaped model names such as models/<id> or publishers/<owner>/models/<id>, normalizing them back to usable model IDs without breaking existing path-style provider IDs such as Fireworks model names.
  • Extended that same shared discovery path again so wrapped registries such as data.provider_models, result.data.providerModels, and result.data.publisherModels are treated like first-class inputs instead of being silently missed when providers add one more response envelope around the catalog.
  • Extended the shared discovery collector once more so wrapped registry, registries, and services catalogs now enter the same bounded parser path instead of being ignored unless they happened to sit under a narrower already-supported key.
  • Model-aware global token guidance now persists an explicit auto-managed tracking marker (globalModelAwareMaxTokensTracking) instead of relying only on value-equality heuristics. That keeps manual model changes, chunk-size auto-fill, partial/complete reset, and runtime max-token ceiling behavior aligned across settings reloads and later sessions, while discovered-model apply now follows its own provider-scoped override lane.
  • Expanded generic OpenAI Compatible fetch-model-list behavior so known hosts such as OpenRouter, Requesty, Together, Huawei Cloud MaaS, Vercel AI Gateway, and local LiteLLM-style proxies now auto-resolve into the matching bounded discovery family instead of always falling back to a plain generic /models request.
  • Extended that generic bounded-host upgrade so OVMS-style local /v3 endpoints are recognized separately from LiteLLM-style local proxies, preserving OVMS's /v3/models plus legacy /v1/config fallback path even when users stay on the generic OpenAI Compatible preset.
  • Unified OpenAI-compatible endpoint normalization across runtime and fetch-model-list so users can paste /responses, /chat/completions, or /models endpoint forms directly without breaking model discovery or request routing.
  • Expanded fetch-model-list again with xAI's dedicated /v1/language-models registry plus bounded fallback to /v1/models, so Grok-facing setups get a narrower language-model list without introducing a second persisted catalog subsystem.
  • Tightened the shared parser so alias-only model entries can still recover a usable identifier when upstream payloads omit the primary id field, while avoiding the bad tradeoff of expanding every alias into separate picker rows.
  • Hardened shared endpoint normalization one more time so pasted OpenAI-compatible model endpoints remain usable even when users include query strings or hash fragments after /models, /responses, or /chat/completions.
  • Added bounded AIHubMix model discovery through its hosted /api/v1/models registry, and wired generic OpenAI Compatible endpoints targeting aihubmix.com into that same family-aware discovery path.
  • Added a first-class GitHub Models preset plus bounded discovery through its hosted catalog/models + /v1/models merge, and wired generic OpenAI Compatible endpoints targeting models.github.ai into that same family-aware discovery path.
  • Added a first-class PPIO preset plus bounded discovery through its chat + embedding + reranker registry merge, and wired generic OpenAI Compatible endpoints targeting api.ppinfra.com into that same family-aware discovery path.
  • Added a first-class New API preset that reuses the shared bounded OpenAI-compatible /models flow instead of requiring a separate heavy catalog subsystem.
  • Added a first-class OVMS preset that prefers the current local /v3/models endpoint and only falls back to legacy /v1/config when needed, preserving bounded local discovery without broad localhost auto-detection claims.
  • Added the Relax input file restrictions developer setting plus the shared supported-input reader layer.
  • Extended advanced folder-task selection with profile-aware folder locking, temporary overrides, preview panels, and example/preset chips.

Fixes

  • Provider advanced settings now auto-expand when a provider already has saved advanced overrides, preserving visibility for existing data.json behavior instead of silently collapsing non-default live config.
  • Added discovery prerequisite validation and endpoint normalization so /chat/completions, /models, and /tags style base URLs do not cause malformed model-list requests.
  • Added canonical provider-name normalization for legacy persisted/imported aliases such as Xiaomi -> Xiaomi MiMo, preventing duplicate provider rows across settings load and profile import.
  • Added model-aware token guidance in provider settings so the current model's known output-token cap is surfaced beside Model, provider override, and global Max tokens, with immediate re-render after edits.
  • Added model-aware auto-sync for global token defaults: when Max tokens and Chunk word count are still following the default/auto-managed baseline, changing the typed selected model now updates them to the model's known ceiling and the matching recommended chunk size. Once the user manually diverges, later typed model changes no longer overwrite those values.
  • Extended model-aware token guidance so explicit gateway/provider-prefixed model IDs from presets such as AIHubMix, GitHub Models, and other bounded gateway families can still drive Max tokens / chunk-size guidance when the upstream model owner is clear, while keeping generic bare-model guessing intentionally conservative for custom OpenAI Compatible endpoints.
  • Unified task-aware input validation across sidebar actions, command entrypoints, batch adapters, and path-based diagram generation.
  • Corrected relaxed-mode batch empty-folder notices so they describe supported inputs rather than hard-coded .md / .txt assumptions.
  • Kept source-mutating and verbatim-sensitive flows intentionally restricted while widening only the safe allowlist.

Verification

  • Revalidated locally with serial repo-saga refresh, build, full Jest, i18n audit, render-host audit, repo-local Obsidian build/test shims, CLI help checks, and whitespace verification before release.

2026-05-25

Bounded Product-Surface Recovery On Current Main

  • Recovered bounded backup-branch product slices onto the rewritten current main without reviving the disproven dedicated-runtime topology.
  • Current main now again includes settings reset, concept-note prerequisite guidance, concept synonym suppression, saved file-selection profiles for folder tasks, local knowledge retrieval, chapter split, and saved-artifact-aware diagram preview follow-through.
  • Kept the recovery architecture-compatible with the current single-entry main.js + inline srcdoc runtime truth rather than replaying older detached runtime assumptions.
  • Re-synchronized the shipped 1.8.9 truth across package metadata, the welcome-modal release digest, and the README family without reintroducing stale render-host.mjs packaging claims.

Documentation And Archive Recovery

  • Restored backup-branch release/doc artifacts for 1.8.6 through 1.8.9 under docs/releases/ and re-aligned the current package-facing version surfaces to the actual latest recovered release boundary.
  • Re-aligned README and progress documents so they no longer describe recovered current-main features as “missing on current main.”

Verification Status

  • Recovery acceptance was re-validated with build, Jest, UI-string audit, render-host audit, and whitespace checks before the final commit batch.

2026-05-24

Bounded CLI/Public-Surface Restoration On Current Main

  • Restored a bounded backup-branch slice that is truthful on the rewritten current main, instead of reviving the wider maintainer bridge that depends on unrecovered product features.
  • Added provider.profile.export-redacted, cli.public-surface.export, and a derived src/operations/publicCliSurface.ts so the public-safe export surface now comes from the same registry/capability/contract stack as the rest of the automation metadata.
  • Added handling tags for secret-bearing provider profile import/export paths and excluded raw provider export from the public-safe surface.
  • Added a repo-local maintainer helper surface: npm run cli:help and npm run cli:invoke -- --vault <vault> --operation <operation-id> [--pretty], intentionally limited to export-only operations with no input payload.
  • Added localized UI strings, notices, tests, and maintainer documentation updates for the redacted export / public-surface slice.

Repo-Saga Serial Execution Guardrail

  • Added a repo-saga execution lock helper so chronicle:update and chronicle:sync-repo-saga fail fast instead of racing on shared clone/cache state.
  • Documented the serial-execution rule in maintainer workflow docs and covered lock-release behavior in Jest.

1.8.4

Highlights

  • Diagram intent controls added across settings and sidebar/workbench flows
  • Batch progress persistence and per-setting sync locality controls added
  • Mermaid/preview/export reliability significantly hardened
  • Release workflow and multilingual docs aligned with the real codebase delta
  • Version metadata synchronized for the 1.8.4 release

New Features

  • Diagram Intent Control: Added preferred diagram intent selectors in settings and the Notemd Workbench/sidebar path.
  • Batch Progress Store: Long-running batch flows now persist progress for better interrupt/resume behavior.
  • Sync Locality Control: Added per-setting sync toggles so secret-like provider values can remain local.
  • LLM Response Cache: Added in-memory caching for repeated identical LLM calls in the same session.
  • Batch Extract Specific Original Text: Added a folder-scoped batch action for the configured original-text extraction workflow so the same question set can run across all eligible files in a selected folder.
  • Xiaomi MiMo Preset: Added a first-class Xiaomi MiMo OpenAI-compatible preset with MiMo defaults and shared-runtime coverage.
  • Welcome Modal Release Digest: The first-install welcome modal now embeds the latest two release summaries in a scrollable panel so new users can see what changed without leaving the plugin.

Fixes

  • Diagram Enforcement: Explicit diagram intent now blocks HTML fallback, retries with the requested type, and backfills missing node labels from ids.
  • Mermaid Stability: Sanitized trailing whitespace, added parse-error retry, and aligned boundary handling with Obsidian Mermaid 11.4.1 expectations.
  • Artifact Saving: Mermaid retry failures now still save best-effort output; Vega-Lite exports now save readable .md artifacts instead of raw .json.
  • Release Workflow Hardening: .github/workflows/release.yml now pins supported actions/checkout and actions/setup-node majors.
  • README Contract Coverage: Added dedicated tests to keep localized README version/sponsor/badge alignment from drifting again.
  • Concept-Note Path Guardrails: Add-links / concept-extraction flows now warn when concept-note path prerequisites are not configured and can deep-link users into the correct settings section.
  • OpenAI-Compatible Endpoint Normalization: Base URLs that already include /chat/completions or /models are now normalized before request assembly.
  • Settings Input Focus Stability: Remaining conditional settings inputs now commit on blur/Enter instead of redrawing mid-typing, fixing the one-digit blur regression.
  • Safer Defaults And Processing Guidance: Customize Concept Note Path and Enable Mermaid Error Detection now default to enabled; Max Tokens is ordered before Chunk Word Count, and the chunk field now documents and auto-fills the recommended one-third ratio when users have not customized it.

Chores

  • Added docs/releases/1.8.4.md and docs/releases/1.8.4.zh-CN.md.
  • Refreshed architecture, progress, Drawnix feasibility, and maintainer release-workflow docs.
  • Added docs/brainstorms/2026-05-03-mainline-stabilization-and-ci-hardening-requirements*.md.
  • Added .trellis/ to .gitignore and concise local-workflow-state protection rules to AGENTS.md.
  • Kept the existing 1.8.3 historical entry below for traceability.

1.8.5

Highlights

  • Diagram command, sidebar action, workflow-builder, and workbench-visible copy now converge on canonical Generate diagram / Preview diagram wording while preserving legacy compatibility IDs.
  • The 1.8.5 release truth is now synchronized across packaged assets, welcome-modal release digest, README metadata, and bilingual release notes.
  • The next CLI implementation wave now includes the first checked-in packaging / semantic-verification convergence slice instead of leaving that phase only in brainstorm notes and task scaffolding.

New Features

  • Canonical Diagram Labels: Sidebar buttons, workflow-builder action help, command labels, localized tooltips, and workbench-visible notices now consistently present the canonical diagram action names.
  • Release Digest Refresh: The first-install welcome modal now advertises 1.8.5 and 1.8.4 as the latest two updates, keeping the in-plugin onboarding summary aligned with the actual shipped release.
  • Maintainer Semantic Verification Helper: The repo now ships npm run verify:diagram-semantics, which generates secret-free checklist templates with repo gates, vault-aware CLI checks, packaging-boundary reminders, and per-surface evidence sections.
  • Next-Phase Task Seeded: Added a dedicated Trellis task shell for packaging / semantic-verification convergence so the next implementation batch can land with persisted context instead of ad-hoc chat memory.

Fixes

  • Legacy Experimental Copy Cleanup: Removed the remaining user-visible experimental diagram wording from diagram action labels, workbench logs, and completion/error notices where the runtime is already canonical internally.
  • Release Truth Alignment: Version markers in package.json, manifest.json, versions.json, README.md, README_zh.md, and release-note references are now synchronized for 1.8.5.
  • Welcome Modal Freshness: The built-in recent-release digest no longer lags one patch behind the shipped plugin version.
  • Packaging-Boundary Truth Lock: Maintainer docs and generated verification templates now explicitly state that audit:render-host proves the current single-entry main.js + inline srcdoc contract, not finished heavy-runtime isolation.

Chores

  • Added docs/releases/1.8.5.md and docs/releases/1.8.5.zh-CN.md.
  • Kept the packaging / semantic-verification convergence direction anchored in current maintainer docs, generated helper output, and active Trellis task scaffolding.

1.8.3

Highlights

  • First-install Welcome Modal with i18n across all 22 locales
  • Sponsor support: Give a Star on GitHub + Buy Me a Coffee (ko-fi)
  • Cline-aligned output token resolution for unknown models
  • Diagram edge field normalization for LLM JSON output compatibility
  • Hard constraints documented for MermaidProcessor decomposition and legacy prompt retirement
  • notebook-navigator cross-reference analysis with 5 improvement angles identified

New Features

  • Welcome Modal: Shown on first install with plugin capabilities overview, LLM setup hint, closing poetic message, and sponsor buttons. Fully i18n-backed (22 locales).
  • Sponsor Support: Settings tab sponsor section with GitHub Star + ko-fi Coffee buttons. manifest.json fundingUrl for native Obsidian community browser display.

Fixes

  • Cline-aligned token resolution: Unknown models with default maxTokens now defer to the API provider instead of being capped at 8192.
  • Diagram edge normalization: LLM outputs using source/target edge fields are now normalized to from/to, fixing validation failures with real API responses.

Chores

  • Hard constraints codified for MermaidProcessor decomposition: each sub-task requires real Obsidian verification with saved image validation.
  • Legacy prompt retirement gated: original scenario usability must be fully preserved.
  • notebook-navigator design reference analysis added to audit docs and roadmap.
  • ko-fi link: https://ko-fi.com/jacobinwwey

1.8.3

亮点

  • 首次安装欢迎弹窗，覆盖全部 22 种语言
  • 赞助支持：给 GitHub 加星 + 请喝杯咖啡（ko-fi）
  • Cline 对齐的未知模型输出令牌处理
  • 图表边缘字段规范化，兼容 LLM JSON 输出
  • MermaidProcessor 分解和旧版提示退役的硬性约束已文档化
  • notebook-navigator 交叉参考分析，识别 5 项改进方向

新功能

  • 欢迎弹窗：首次安装时展示，包含插件能力概述、LLM 配置提示、结语诗句和赞助按钮。22 种语言完整 i18n。
  • 赞助支持：设置页赞助板块，含 GitHub 加星 + ko-fi 咖啡按钮。manifest.json fundingUrl 支持 Obsidian 社区插件浏览器原生展示。

修复

  • Cline 对齐令牌解析：未知模型使用默认 maxTokens 时现在由 API 提供商自行决定，不再被限制在 8192。
  • 图表边缘规范化：LLM 输出使用 source/target 边缘字段现被规范化为 from/to，修复真实 API 响应的验证失败。

杂项

  • MermaidProcessor 分解硬性约束：每个子任务需真实 Obsidian 验证及图像保存核验。
  • 旧版提示退役门槛：必须完全保留原场景可用性。
  • notebook-navigator 设计参考分析已添加至审计文档和路线图。
  • ko-fi 链接：https://ko-fi.com/jacobinwwey

[Unreleased]

English

  • Mainline Force-Rewrite Audit: Added a fresh current-main audit plus a new unified follow-through matrix that explicitly separates rewritten origin/main shipped truth from backup-branch-only evidence, so progress docs no longer overclaim features missing from the live mainline.
  • Clean-State Guardrails For Local Verification: .gitignore now excludes root-level render-host.mjs plus local vault-generated artifacts under docs/ such as error logs, provider diagnostics, preview exports, and saved diagram artifacts, preventing local Obsidian verification from dirtying the repository.
  • Repo-Saga Serial Execution Guardrail: npm run chronicle:sync-repo-saga and npm run chronicle:update now take a shared execution lock under .cache/.repo-saga-execution.lock, so overlapping local or CI chronicle refreshes fail fast instead of racing on the same repo-saga cache tree.
  • Model-Aware Output Token Caps: Added Cline-aligned known-model max-output-token metadata and now resolve the effective output cap from the actual selected model instead of hardcoded provider defaults. When the global max-token setting is still at the stock default, supported models automatically use their true known ceiling; when users set a custom global or provider-specific cap, requests are clamped to the model’s known maximum. This now applies across OpenAI-compatible providers plus Anthropic, Google, Azure OpenAI, and Ollama runtimes.
  • Developer-Mode Diagnostics Panel: Added a dedicated Settings developer panel gated by a new Developer mode switch, so normal users do not see developer-only controls.
  • Selectable Diagnostic Call Modes + Stability Runs: Developer diagnostics can now run with selectable call modes (including OpenAI-compatible forced transport modes) and can execute repeated stability runs with aggregated reporting.
  • OpenAI-Compatible Non-Stream Stage In Stable Mode: After a primary direct streaming failure, stable-mode OpenAI-compatible calls now attempt direct non-stream transport before falling back to requestUrl. This captures providers that return valid buffered responses but are unstable on streaming sockets.
  • Protocol-Aware Streaming Fallback Across All LLM Paths: Extended long-request fallback parsing beyond OpenAI-compatible providers so Anthropic Messages, Google Gemini, Azure OpenAI, and Ollama now also use protocol-aware streamed fallback handling on desktop http/https and non-desktop fetch. Legacy exported OpenAI-style provider wrappers now delegate to the same shared streamed fallback path instead of keeping buffered-only logic.
  • Cross-Transport Partial Stream Debugging: Shared debug output now preserves parsed partial stream text for Anthropic, Google/Azure-style SSE flows, and Ollama NDJSON fallbacks, not just OpenAI-compatible calls.
  • Stable-Mode Primary Transport Preference (OpenAI-Compatible): In stable mode, OpenAI-compatible runtime calls now prefer direct desktop/web streaming transport as the primary long-request path before trying requestUrl. This reduces false-fail chains where requestUrl disconnects even though upstream providers eventually return successful non-stream responses.
  • Regression Coverage: Added focused runtime tests for Anthropic, Google, Azure OpenAI, Ollama, and the remaining direct OpenAI-style provider wrappers on streamed fallback success and interruption scenarios.
  • Real Endpoint Diagnostic CLI: Added scripts/diagnose-llm-provider.js (and npm run diagnose:llm) to run reproducible buffered/streaming long-request diagnostics against real provider endpoints with sanitized per-attempt timing, headers, partial bodies, and parsed stream fragments.
  • Settings Developer Diagnostic Button: Added Developer provider diagnostic (long request) in settings to run an in-plugin runtime probe for the active provider and persist a full report (Notemd_Provider_Diagnostic_*.txt) in vault root.
  • Diagnostic Runtime Helper + Tests: Added src/providerDiagnostics.ts and trigger-chain coverage in src/tests/providerDiagnostics.test.ts to lock report generation behavior for both success and provider-failure debug paths.
  • Docs and Agent Guide Alignment: Updated both READMEs and AGENTS.md so in-plugin diagnostics and CLI diagnostics are documented and required to stay semantically aligned.

Chinese (中文)

  • 主线 Force Rewrite 审计: 新增一份针对当前主线的重新审计文档与统一推进矩阵，明确区分被重写后的 origin/main 发货真值与仅存在于备份分支中的证据，避免进度文档继续高估 live mainline 已具备的能力。
  • 本机验证的 Clean-State 守护: .gitignore 现在会忽略根目录 render-host.mjs 以及 docs/ 下本地 vault 生成的错误日志、provider 诊断文件、预览导出产物和已保存图表工件，避免每次本机 Obsidian 验证后都把仓库弄脏。
  • Repo-Saga 串行执行守护: npm run chronicle:sync-repo-saga 与 npm run chronicle:update 现在会共享 .cache/.repo-saga-execution.lock 执行锁，避免本地或 CI 并发刷新同一套 repo-saga 缓存目录时互相踩状态。
  • 模型感知的输出 Token 上限: 新增与 Cline 对齐的已知模型最大输出 Token 元数据，并且运行时现在会根据用户实际选择的模型来解析有效输出上限，而不是依赖硬编码的 provider 默认值。当全局最大 Token 仍保持插件默认值时，已支持模型会自动使用其真实已知上限；当用户手动设置全局或 provider 级上限时，请求会被钳制到该模型的已知最大值。此行为现已覆盖 OpenAI-compatible 路径以及 Anthropic、Google、Azure OpenAI、Ollama 运行时。
  • 开发者模式诊断面板: 设置页新增由“Developer mode”开关控制的独立开发者面板，默认对普通用户隐藏开发者专用控件。
  • 可选诊断调用方式与稳定性多轮测试: 开发者诊断现在可选择调用方式（含 OpenAI-compatible 的强制传输模式），并支持按指定轮次执行稳定性测试并输出聚合报告。
  • OpenAI-compatible 稳定模式新增非流式阶段: 当主直连流式传输失败后，稳定模式下的 OpenAI-compatible 调用现在会先尝试直连非流式传输，再回退到 requestUrl。这能覆盖“流式链路不稳定但非流式可正常返回”的 Provider 场景。
  • 全 LLM 路径的协议感知流式回退: 长请求回退解析能力已从 OpenAI-compatible 扩展到 Anthropic Messages、Google Gemini、Azure OpenAI 和 Ollama，使这些 transport 在桌面 http/https 与非桌面 fetch 下都能走协议感知的流式回退链路。遗留的 OpenAI 风格导出 Provider 包装函数也已收敛到同一套共享流式回退路径，不再保留 buffered-only 旧逻辑。
  • 跨 Transport 的部分流式调试: 共享调试输出现在不仅覆盖 OpenAI-compatible，也会为 Anthropic、Google/Azure 风格 SSE，以及 Ollama NDJSON 回退保留“已解析的部分流式文本”。
  • 稳定模式下的主传输优先策略（OpenAI-Compatible）: 在稳定 API 调用模式下，OpenAI-compatible 运行时链路会优先使用桌面/网页流式传输作为主长请求路径，再回退到 requestUrl。这可以减少“requestUrl 先断连，但上游最终成功返回”导致的误失败链路。
  • 回归测试覆盖: 新增 Anthropic、Google、Azure OpenAI、Ollama，以及剩余直连 OpenAI 风格 Provider 包装函数在流式回退成功与中途中断场景下的定向运行时测试。
  • 真实 Endpoint 诊断 CLI: 新增 scripts/diagnose-llm-provider.js（以及 npm run diagnose:llm），可直接对真实 Provider 执行 buffered/streaming 长请求对照诊断，并输出脱敏后的每次尝试耗时、响应头、部分响应体与已解析流式片段。
  • 设置页开发者诊断按钮: 在设置页新增 Developer provider diagnostic (long request) 按钮，可直接对当前活动 Provider 发起插件内运行时探针，并在仓库根目录保存完整报告（Notemd_Provider_Diagnostic_*.txt）。
  • 诊断运行时模块与测试: 新增 src/providerDiagnostics.ts，并在 src/tests/providerDiagnostics.test.ts 中补齐按钮触发链路对应的成功/失败路径测试，确保报告输出与调试信息稳定可回归。
  • 文档与代理指南对齐: 已同步更新中英文 README 与 AGENTS.md，明确插件内诊断与 CLI 诊断需要保持协议语义一致。

[1.8.1] - 2026-04-13

English

  • Complete Supported UI Locale Coverage: Finished localizing the remaining visible settings, workflow builder, sidebar tooltip, extracted-text settings, advanced settings, and output-path surfaces for every advertised UI locale.
  • Stricter Regression Coverage: Added focused tests so supported locales no longer silently fall back to English on those visible UI surfaces.
  • Documentation Reality Check: Updated the root READMEs and Language Hub to distinguish maintainer source docs, published README translations, and the exact in-app UI locale catalog.
  • Version Alignment: Synchronized release metadata and packaged documentation for 1.8.1.

Chinese (中文)

  • 补齐受支持 UI Locale 的可见覆盖面: 完成了所有已宣称支持 locale 在剩余设置项、工作流构建器、侧边栏提示、原文提取设置、高级设置以及输出路径配置等可见界面上的本地化补齐。
  • 强化回归测试覆盖: 新增定向测试，确保这些已支持 locale 不再在上述可见 UI 上静默回落到英文。
  • 文档现实校准: 更新仓库根 README 与语言中心，明确区分维护源文档、已发布 README 译文以及应用内 UI locale 的精确目录。
  • 版本元数据对齐: 已将 1.8.1 的版本信息与随包文档同步完成。

[1.8.0] - 2026-04-09

English

  • First-Principles Language Support Architecture: Added a unified language domain model that separates UI locale from task output language policy.
  • UI i18n Foundation: Introduced centralized locale catalogs for all supported UI locales (en, ar, de, es, fa, fr, id, it, ja, ko, nl, pl, pt, pt-BR, ru, th, tr, uk, vi, zh-CN, zh-TW) with fallback chaining and interpolation support.
  • Runtime String Migration: Migrated high-impact settings/sidebar/error-modal/runtime notices to shared i18n strings.
  • Policy Unification Across Task Paths: Consolidated task-language decisions in one policy layer and wired it into prompt, processing, research, and translation flows.
  • Locale/RTL Hardening: Added locale-aware timestamp formatting and RTL safety styling guards.
  • Regression Harness: Added reproducible baseline/compare scripts and preserved phase evidence for before/after robustness checks.
  • Build Stability with Reference Repo: Excluded ref/** from TypeScript/Jest/ESLint scopes so reference projects can coexist without breaking plugin build/test gates.

Chinese (中文)

  • 第一性原理语言支持架构: 新增统一语言域模型，明确分离 UI 语言与任务输出语言策略。
  • UI i18n 基础设施: 为全部受支持的 UI locale（en、ar、de、es、fa、fr、id、it、ja、ko、nl、pl、pt、pt-BR、ru、th、tr、uk、vi、zh-CN、zh-TW）引入集中式语言目录，并支持回退链与变量插值。
  • 运行时文案迁移: 将设置页、侧边栏、错误弹窗和主流程关键提示迁移到共享 i18n 字符串体系。
  • 任务语言策略收口: 将任务语言决策统一收敛到策略层，并接入 prompt、处理、研究与翻译路径。
  • Locale/RTL 鲁棒性增强: 新增本地化时间格式化与 RTL 布局安全样式保护。
  • 回归脚本化能力: 增加可复现 baseline/compare 脚本，并保留阶段性前后对比证据以保障改造过程鲁棒性。
  • 参考仓共存构建稳定性: 将 ref/** 排除出 TypeScript/Jest/ESLint 编译与测试范围，使参考仓引入不再破坏插件主工程门禁。

[1.7.14] - 2026-03-30

English

  • Single-File CTA Visual Mapping: Sidebar colorful CTA styling is now strictly mapped to single-file actions. Batch/folder-level actions no longer use CTA color treatment, reducing scope confusion and accidental batch triggers.
  • Workflow CTA Rule Alignment: Custom workflow buttons now inherit CTA styling only when every step is single-file. Mixed workflows that include batch steps are rendered as non-CTA to keep visual semantics consistent.
  • Regression Coverage: Added a sidebar DOM test to lock CTA mapping behavior for single-file actions vs batch/folder actions and default mixed workflow buttons.

Chinese (中文)

  • 单文件 CTA 视觉映射: 侧边栏彩色 CTA 样式现在严格对应“单文件处理”动作；批量/文件夹级动作不再使用 CTA 色彩，以减少范围误判和误触发批处理的风险。
  • 工作流 CTA 规则对齐: 自定义工作流按钮仅在“所有步骤均为单文件动作”时才应用 CTA 样式；只要包含批量步骤即回落为非 CTA 样式，保持视觉语义一致。
  • 回归测试覆盖: 新增侧边栏 DOM 测试，锁定单文件动作、批量/文件夹动作以及默认混合工作流按钮的 CTA 映射行为。

[1.7.13] - 2026-03-30

English

  • Sidebar Interaction Polish: Refined sidebar button interactions with clearer hover, pressed, and focus-visible states so action feedback is obvious during rapid workflows.
  • CTA Readability Fix: Improved text contrast on colorful CTA buttons (including One-Click Extract and Batch generate from titles) to avoid low-contrast white-on-light combinations across different Obsidian themes.
  • Docked Footer Visibility Tuning: Increased persistent footer sizing constraints and log minimum height so progress + log telemetry stays visible when multiple panels are expanded.

Chinese (中文)

  • 侧边栏交互打磨: 细化了侧边栏按钮的悬停、按下与键盘焦点态，让高频工作流中的操作反馈更明确。
  • CTA 可读性修复: 提升了彩色 CTA 按钮（包括 One-Click Extract、Batch generate from titles）的文字对比度，避免不同主题下出现浅底白字可见性不足的问题。
  • 底部停靠区可见性优化: 调整了固定底部区域尺寸约束与日志最小高度，在展开多个分组时仍能稳定看到进度与日志信息。

[1.7.8] - 2026-03-29

English

  • OpenAI-Compatible Streaming Fallback: Long-running OpenAI-compatible task calls now upgrade their desktop http/https and non-desktop fetch fallback attempts to streaming response parsing after an initial transient requestUrl disconnect. This helps slow gateways and reverse proxies return body chunks earlier instead of failing the whole request while waiting for one large buffered response.
  • Parsed Partial Stream Debugging: Shared deep-debug output now records both raw partial bodies and parsed partial stream text when a streamed fallback attempt is interrupted. This gives developers enough evidence to distinguish between transport resets and upstream provider-side error payloads.
  • Regression Coverage: Added focused tests for desktop and web streaming fallback assembly, plus interrupted streaming fallback diagnostics.

Chinese (中文)

  • OpenAI-compatible 流式回退: 长耗时的 OpenAI-compatible 任务调用在首次 requestUrl 瞬时断连后，现在会把桌面端 http/https 与非桌面环境 fetch 的回退尝试升级为流式响应解析。这样慢速网关和反向代理可以更早返回 body 分片，而不是一直等待整块缓冲响应后直接判定整次请求失败。
  • 已解析流式片段调试: 共享深度调试输出现在会在流式回退被中途打断时，同时记录原始部分响应体和“已解析的部分流式文本”。开发者可以更清楚地区分到底是传输层重置，还是上游 Provider 已经返回了可用于定位问题的错误/内容片段。
  • 回归测试覆盖: 新增桌面端与 Web 流式回退拼装测试，以及流式回退中断时的诊断信息测试。

[1.7.7] - 2026-03-26

English

  • Cross-Provider Deep Debugging: The shared LLM transport/error path now records per-attempt debug metadata for every provider, including transport name, sanitized request URL and headers, elapsed duration, response headers, partial response bodies, and stack traces. This makes slow upstream gateway resets and proxy disconnects visible without relying on provider-specific logging.
  • Connection Test Debug Consistency: Provider connection probes now use the same instrumented request path as runtime calls, so debug mode exposes the same transport-level evidence during "Test Connection" failures.
  • Non-Desktop Runtime Fallback: When desktop Node transport is unavailable, transient requestUrl disconnects now retry the same attempt through browser fetch, extending runtime resilience and provider connection testing to mobile/non-desktop environments.
  • Chinese Preset Refresh: Synced selected China-focused preset defaults with current cline model baselines, including Qwen -> qwen3-235b-a22b, Moonshot -> kimi-k2-0905-preview, and MiniMax -> MiniMax-M2.7.
  • Regression Coverage: Added focused tests for shared debug metadata rendering and interrupted desktop fallback responses with partial-body capture.

Chinese (中文)

  • 跨 Provider 深度调试: 共享的 LLM 传输/错误处理链路现在会为所有 Provider 记录按尝试维度展开的调试元数据，包括 transport 名称、脱敏后的请求 URL/请求头、耗时、响应头、部分响应体与堆栈信息。这样即使没有 Provider 特定日志，也能直接看见慢速上游网关重置或代理断连发生在哪一段链路上。
  • 连接测试调试一致性: Provider 的连接探测现在复用与运行时调用相同的带仪表化请求路径，因此“测试连接”失败时，调试模式也会输出同等级别的 transport 证据。
  • 非桌面运行时回退: 当桌面 Node 传输不可用时，requestUrl 的瞬时断连现在会在同一次调用内切换到浏览器 fetch 重试，从而将运行时鲁棒性和 Provider 连接测试能力扩展到移动端/非桌面环境。
  • 中国区预设刷新: 已将部分中国区 Provider 预设默认模型同步到当前 cline 基线，包括 Qwen -> qwen3-235b-a22b、Moonshot -> kimi-k2-0905-preview、MiniMax -> MiniMax-M2.7。
  • 回归测试覆盖: 新增共享调试元数据渲染测试，以及桌面端 fallback 在响应中途断开时对部分响应体的捕获测试。

[1.7.6] - 2026-03-26

English

  • Runtime Transport Robustness: Long-running LLM task calls now switch from Obsidian requestUrl to desktop http/https transport after transient disconnects such as ERR_CONNECTION_CLOSED, then continue into the existing stable retry sequence only if the fallback also fails. This hardens slow translation and generation jobs against proxy or gateway drops.
  • China Provider Expansion Round 2: Added first-class presets for Qwen Code, Z AI, and Huawei Cloud MaaS, with routing coverage, connection-test coverage, and synchronized English/Chinese provider documentation.
  • Sidebar Footer Stability: Refined the sidebar footer into a docked status/log area with a clearer standby progress state so the log panel remains visible and the ready-state progress area is readable even when every section is expanded.
  • Regression Coverage: Added focused runtime fallback tests for every active transport path plus sidebar DOM tests for the docked footer and standby progress state.

Chinese (中文)

  • 运行时传输鲁棒性增强: 长耗时 LLM 任务在遇到 ERR_CONNECTION_CLOSED 等瞬时断连时，现在会先从 Obsidian 的 requestUrl 切换到桌面端 http/https 传输；只有该回退也失败时，才进入原有稳定重试序列。慢速翻译和生成任务对代理/网关断连的容忍度明显提升。
  • 中国区 Provider 第二轮扩展: 新增 Qwen Code、Z AI、Huawei Cloud MaaS 三个一等预设，并补齐运行时路由测试、连接测试覆盖以及中英文 Provider 文档说明。
  • 侧边栏底部区域稳定性提升: 将侧边栏底部进一步整理为停靠式状态/日志区域，并增强 Ready 状态下的待机进度显示；即使所有分组全部展开，日志面板也不会继续被挤压到不可见。
  • 回归测试覆盖: 新增覆盖全部活跃 transport 路径的运行时回退测试，以及锁定 docked footer 与 standby 进度态的 sidebar DOM 测试。

[1.7.5] - 2026-03-26

English

  • Provider Connection Test Robustness: Connection tests now fall back to the stable retry sequence after the first transient network disconnect instead of failing immediately, aligning the "Test Connection" button with the runtime resilience already used by task execution.
  • Full Transport Coverage: The transient-failure fallback now covers every transport used by built-in provider presets, including OpenAI-compatible providers, Anthropic, Google, Azure OpenAI, and Ollama.
  • Regression Coverage: Added focused provider-support tests for both runtime API calls and connection-test probes so transient disconnect handling stays locked in across transports.
  • Documentation: Updated the English and Chinese READMEs plus version metadata to document the new provider-connection retry behavior for 1.7.5.

Chinese (中文)

  • Provider 连接测试鲁棒性增强: 连接测试在首次遇到瞬时网络断连后，不会再立刻失败，而是会回退到稳定重试序列，使“测试连接”按钮与实际任务执行路径的鲁棒性保持一致。
  • 全 Transport 覆盖: 这套瞬时失败回退机制现已覆盖所有内置 Provider 预设所使用的传输链路，包括 OpenAI-compatible、Anthropic、Google、Azure OpenAI 与 Ollama。
  • 回归测试覆盖: 新增针对运行时 API 调用与连接测试探测链路的 Provider 支持测试，确保跨 transport 的瞬时断连处理不会回退。
  • 文档更新: 已同步更新中英文 README 与版本元数据，记录 1.7.5 的 Provider 连接测试重试行为。

[1.7.4] - 2026-03-26

English

  • Sidebar Layout Stability: Reworked the sidebar into a scrollable action area plus a persistent footer so the progress card and log output remain visible even when all action groups are expanded.
  • Clearer Progress Feedback: Moved the percentage text into a dedicated status pill, kept the progress bar visible in the ready state, and improved visual contrast for the sidebar progress area.
  • Regression Coverage: Added focused sidebar DOM tests to lock in the persistent footer layout and the new progress summary behavior.
  • Documentation: Updated the English and Chinese READMEs to reflect the pinned sidebar telemetry layout.

Chinese (中文)

  • 侧边栏布局稳定性: 将侧边栏重构为可滚动的工作区加固定底部区域，即使所有动作分组全部展开，进度卡片和日志输出也不会被挤出视野。
  • 更清晰的进度反馈: 将百分比文本移到独立状态标签中，保留 Ready 状态下可见的进度条，并增强侧边栏进度区域的视觉对比度。
  • 回归测试覆盖: 新增聚焦侧边栏 DOM 的测试，锁定固定底部布局与新的进度摘要展示行为。
  • 文档更新: 同步更新中英文 README，补充固定底部进度/日志布局说明。

[1.6.5] - 2026-01-27

Bug Fixes

English

  • Fix installation Error: Delete the unused js files, Fix the Severe installation difficulties

Chinese (中文)

  • 解决安装问题: 删除无用js文件，解决了严重的安装问题。

[1.6.4] - 2026-01-27

English

  • Search Engine Support: Resolved issues with DuckDuckGo search by implementing a robust, modular architecture. The new DuckDuckGoProvider now uses DOMParser for reliable HTML parsing with a regex fallback, ensuring consistent search results even if the layout changes.
  • Modular Architecture: Refactored the search logic into a modular SearchProvider system, making it easier to add new search engines and maintain existing ones.
  • Code Quality: Integrated comprehensive unit tests for search providers to prevent regression.

Chinese (中文)

  • 搜索引擎支持: 通过实施稳健的模块化架构解决了DuckDuckGo搜索的问题。新的 DuckDuckGoProvider 现在使用 DOMParser 进行可靠的 HTML 解析，并带有正则表达式回退机制，即使布局发生变化也能确保一致的搜索结果。
  • 模块化架构: 将搜索逻辑重构为模块化的 SearchProvider 系统，从而更易于添加新的搜索引擎和维护现有引擎。
  • 代码质量: 集成了针对搜索提供商的全面单元测试，以防止回归。

[1.6.3] - 2026-01-27

English

  • API Error Debugging: Fixed an issue where the "API Error Debugging Mode" would not show the full error response body for certain providers (like LMStudio 429 errors). Now, detailed JSON responses and status codes are correctly captured and logged when this mode is enabled, aiding significantly in troubleshooting connectivity and rate-limit issues.

Chinese (中文)

  • API 错误调试: 修复了“API 错误调试模式”无法显示某些提供商（如 LMStudio 429 错误）的完整错误响应正文的问题。现在，启用此模式后，可以正确捕获并记录详细的 JSON 响应和状态代码，极大地帮助排查连接和速率限制问题。

v1.6.2

English

  • Modularized API Error Handling: Refactored error handling logic to ensure consistency across all tasks.
  • Enhanced Debugging: The "API Error Debugging Mode" now fully supports "Translate", "Search", and "Connection Test" tasks, providing detailed logs (HTTP status codes, raw responses) for deeper troubleshooting.

Chinese (中文)

  • 模块化API错误处理: 重构了错误处理逻辑，确保所有任务的一致性。
  • 增强调试: “API错误调试模式”现在完全支持“翻译”、“搜索”和“连接测试”任务，提供详细的日志（HTTP状态代码、原始响应）以便进行更深入的故障排除。

Bug Fixes

English

  • Fix Mermaid Table Corruption: Resolved a critical issue where the "Batch Mermaid Fix" and "Deep Debug" features would incorrectly apply syntax fixes to Markdown tables (e.g., changing | : --- | to | : -- "- |), ensuring that modifications are strictly limited to code blocks within mermaid tags.
  • Enhanced Table Protection: Implemented robust safeguards in the deep debug processor to specifically ignore lines matching Markdown table separators (e.g., | :--- |) and lines containing :-- :, preserving table integrity even in aggressive fix modes.

Chinese (中文)

  • 修复Mermaid表格损坏: 解决了“批量Mermaid修复”和“深度调试”功能错误地将语法修复应用于Markdown表格（例如，将 | : --- | 更改为 | : -- "- |）的关键问题，确保修改严格限制在 mermaid 标签内的代码块中。
  • 增强的表格保护: 在深度调试处理器中实现了稳健的保护措施，专门忽略匹配Markdown表格分隔符（例如 | :--- |）和包含 :-- : 的行，即使在激进的修复模式下也能保持表格完整性。

Code/Architecture Update

English

  • Modularized API Error Handling: Refactored llmUtils.ts to introduce a centralized handleProviderError function. This function supports both concise error reporting (default) and verbose, "DeepSeek-style" debugging logs (status codes, raw responses), ensuring consistent error handling across all 10+ supported LLM providers.
  • Automatic Error Logging: Implemented a saveErrorLog utility in fileUtils.ts. This function captures the error message, stack trace, and the full session log from the reporter, saving it to a timestamped file (e.g., Notemd_Error_Log_2023-10-27_10-00-00.txt) in the vault root whenever a process fails. It also intelligently prompts users to enable debug mode if it's currently disabled.

Chinese (中文)

  • 模块化API错误处理: 重构了 llmUtils.ts，引入了集中的 handleProviderError 函数。该函数支持简洁的错误报告（默认）和详细的“DeepSeek风格”调试日志（状态代码、原始响应），确保所有10多个受支持的LLM提供商具有一致的错误处理。
  • 自动错误日志记录: 在 fileUtils.ts 中实现了 saveErrorLog 工具。该函数捕获错误消息、堆栈跟踪和来自报告器的完整会话日志，并在进程失败时将其保存到保管库根目录下的带时间戳的文件（例如 Notemd_Error_Log_2023-10-27_10-00-00.txt）。如果当前禁用了调试模式，它还会智能地提示用户启用调试模式。

Implementation of Functionality

English

  • API Error Debugging Mode: Added a new setting in the "Stable API calls" section.
    • Enabled: Activates detailed logging for all API calls, including HTTP status codes and raw response text, aiding in troubleshooting connection issues.
    • Disabled: Maintains the standard, clean user experience with concise error messages.
  • Automatic Log File Generation:
    • When an error occurs during any major operation (processing files, generating content, research, etc.), a detailed log file is now automatically created in the vault root.
    • This file includes the error context and session history, making it easier for users to report bugs and for developers to diagnose issues.

Chinese (中文)

  • API错误调试模式: 在“稳定API调用”部分添加了一个新设置。
    • 启用: 激活所有API调用的详细日志记录，包括HTTP状态代码和原始响应文本，有助于排查连接问题。
    • 禁用: 保持标准的、干净的用户体验，提供简洁的错误消息。
  • 自动日志文件生成:
    • 当任何主要操作（处理文件、生成内容、研究等）期间发生错误时，现在会自动在保管库根目录中创建一个详细的日志文件。
    • 该文件包含错误上下文和会话历史记录，使用户更容易报告错误，也便于开发人员诊断问题。

Implementation of Functionality

English

  • Extract Specific Original Text:
    • Introduced a new feature to extract verbatim text segments from notes based on user-defined questions.
    • Merged Query Mode: Efficiently processes all questions in a single API call.
    • Translation Support: Optionally includes translations of the extracted text.
    • Custom Output: Configurable output paths and filenames.
    • Batch Mermaid Fix Enhancements:
    • Advanced Fix Mode: Added robust handling for unquoted Mermaid node labels containing nested brackets (e.g., Node[Label [Text]]) and special characters.
    • Note Conversion: Added functionality to automatically convert note right/left of comments in Mermaid graphs to edge labels, ensuring better rendering and adherence to standard graph syntax.
    • Malformed Arrow Fix: Implemented deep debugging logic to correct malformed arrow labels where the arrow syntax is incorrectly embedded within quotes (e.g., -->" becomes " -->), ensuring valid Mermaid syntax.
    • Standardize Pipe Labels: Added logic to robustly handle and quote edge labels containing pipes (e.g., |Text| -> |"Text"|), resolving syntax errors in complex diagrams involving mathematical notation or loose pipe usage.
    • Merge Double Labels: Introduced a fix for conflicting double edge labels (e.g., -- "A" -->|"B"|), merging them into a single descriptive label (-- "A<br>(B)" -->) to resolve rendering ambiguities.
    • Error Reporting: Automatically generates a mermaid_error_{foldername}.md report listing files with remaining errors.
    • Move Error Files: Added an option to automatically move files with persistent errors to a specific folder for manual review.

Chinese (中文)

  • 提取特定原始内容:
    • 引入了一项新功能，可根据用户定义的问题从笔记中逐字提取文本段落。
    • 合并查询模式: 在单个API调用中高效处理所有问题。
    • 翻译支持: 可选包含提取文本的翻译。
    • 自定义输出: 可配置输出路径和文件名。
  • 批量Mermaid修复增强:
    • 高级修复模式: 增加了对包含嵌套括号（例如 Node[标签 [文本]]）和特殊字符的未加引号Mermaid节点标签的稳健处理。
    • 注释转换: 增加了自动将 Mermaid 图表中的 note right/left of 注释转换为连接线标签的功能，确保更好的渲染效果并符合标准图表语法。
    • 畸形箭头修复: 实现了深度调试逻辑，用于纠正箭头语法被错误嵌入引号内的畸形箭头标签（例如 -->" 变为 " -->），确保 Mermaid 语法的有效性。
    • 标准化管道标签: 添加了稳健处理和引用包含管道符的连接线标签的逻辑（例如 |文本| -> |"文本"|），解决了涉及数学符号或松散管道符用法的复杂图表中的语法错误。
    • 合并双重标签: 引入了对冲突双重边缘标签（例如 -- "A" -->|"B"|）的修复，将其合并为单个描述性标签（-- "A<br>(B)" -->），以解决渲染歧义。
    • 错误报告: 自动生成 mermaid_error_{foldername}.md 报告，列出仍有错误的文件。 * 移动错误文件: 添加了一个选项，可自动将存在持久错误的文件移动到特定文件夹以供手动审查。

Code/Architecture Update

English

  • Robust Regex for Mermaid: Implemented a sophisticated regex pattern in mermaidProcessor.ts to correctly identify and quote complex node labels without breaking existing valid syntax.
  • Documentation Synchronization: Fully synchronized README.md and README_zh.md to ensure all new features (Extract Text, Mermaid Fixes) are documented in both languages.

Chinese (中文)

  • Mermaid的稳健正则表达式: 在 mermaidProcessor.ts 中实现了复杂的正则表达式模式，以正确识别并引用复杂的节点标签，而不会破坏现有的有效语法。
  • 文档同步: 完全同步了 README.md 和 README_zh.md，确保所有新功能（提取文本，Mermaid修复）都以两种语言记录。

Code/Architecture Update

English

  • Improved Progress Bar Logic: Enhanced the logic for progress bar updates to ensure it accurately reflects the completion status of operations, preventing it from getting stuck at intermediate percentages.

Chinese (中文)

  • 改进的进度条逻辑: 增强了进度条更新逻辑，确保其准确反映操作的完成状态，防止其卡在中间百分比。

Implementation of Functionality

English

  • Accurate UI Feedback for Wiki-Link Command: The "Create Wiki-Link & Generate Note from Selection" command now provides accurate progress updates, reaching 100% completion and automatically closing the progress modal upon successful execution.

Chinese (中文)

  • Wiki-Link命令的准确UI反馈: “从选中文本创建维基链接并生成笔记”命令现在提供准确的进度更新，在成功执行后达到100%完成并自动关闭进度模态框。

Code/Architecture Update

English

  • Refined Concurrent Processing: Implemented a staggered launch mechanism in the createConcurrentProcessor utility. This ensures that API calls within a parallel batch are spaced out by the configured apiCallIntervalMs, preventing request bursts and resolving potential 429 rate-limit errors from LLM providers.
  • Modularized Syntax Fixing: Refactored the batchFixMermaidSyntaxInFolder function by extracting the core logic into a new, reusable fixMermaidSyntaxInFile function. This improves code modularity and maintainability.

Chinese (中文)

  • 优化并发处理: 在 createConcurrentProcessor 工具函数中实现了交错启动机制。这确保了并行批处理中的API调用能根据配置的 apiCallIntervalMs 间隔执行，防止了请求爆发并解决了来自LLM提供商的潜在429速率限制错误。
  • 模块化语法修复: 重构了 batchFixMermaidSyntaxInFolder 函数，将核心逻辑提取到一个新的、可复用的 fixMermaidSyntaxInFile 函数中。这提高了代码的模块化和可维护性。

Implementation of Functionality

English

  • Enabled "Auto Mermaid Fix" Feature: Implemented the core logic for the "Auto Mermaid Fix" feature by adding the fixMermaidSyntaxInFile function. This feature, when enabled in settings, automatically corrects Mermaid and LaTeX syntax in notes generated by various commands, ensuring diagram validity.
  • Updated Documentation: Updated README.md and README_zh.md to reflect the improved reliability of batch processing.

Chinese (中文)

  • 启用“自动Mermaid修复”功能: 通过添加 fixMermaidSyntaxInFile 函数，实现了“自动Mermaid修复”功能的核心逻辑。该功能在设置中启用后，会自动校正由各种命令生成的笔记中的Mermaid和LaTeX语法，确保图表的有效性。
  • 更新文档: 更新了 README.md 和 README_zh.md，以反映批处理可靠性的提升。

Code/Architecture Update

English

  • Parallel Processing Framework: Introduced a parallel processing framework for batch operations. This was achieved by implementing a custom Semaphore for concurrency control and a createConcurrentProcessor factory function in utils.ts.
  • Refactored Batch Commands: Refactored all major batch processing commands (processFolderWithNotemdCommand, batchGenerateContentForTitles, batchExtractConceptsForFolderCommand) in main.ts and fileUtils.ts. These commands now utilize the new concurrent processor, allowing for serial I/O operations while executing LLM API calls in parallel. This significantly improves performance for large folders.
  • Enhanced Progress Reporting: Updated the ProgressReporter interface and its UI implementations (ProgressModal, NotemdSidebarView) to display the number of active parallel tasks, providing users with more detailed feedback during batch operations.
  • Robustness and Error Handling:
    • Resolved various TypeScript build errors, including incorrect import paths for normalizeNameForFilePath and type mismatches for view.file?.basename.
    • Added safeguards to prevent runtime errors, such as ensuring a note is successfully created before attempting to generate content for it.

Chinese (中文)

  • 并行处理框架: 为批量操作引入了并行处理框架。通过在 utils.ts 中实现自定义的 Semaphore (信号量) 和 createConcurrentProcessor 工厂函数来完成此项工作。
  • 重构批量处理命令: 重构了 main.ts 和 fileUtils.ts 中的所有主要批量处理命令 (processFolderWithNotemdCommand, batchGenerateContentForTitles, batchExtractConceptsForFolderCommand)。这些命令现在利用新的并发处理器，允许在并行执行 LLM API 调用的同时进行串行 I/O 操作，从而显著提高了处理大文件夹时的性能。
  • 增强的进度报告: 更新了 ProgressReporter 接口及其 UI 实现 (ProgressModal, NotemdSidebarView)，以显示活动并行任务的数量，在批量操作期间为用户提供更详细的反馈。
  • 健壮性与错误处理:
    • 解决了多个 TypeScript 构建错误，包括 normalizeNameForFilePath 的不正确导入路径以及 view.file?.basename 的类型不匹配问题。
    • 增加了安全措施以防止运行时错误，例如在尝试为笔记生成内容之前，确保笔记已成功创建。

Implementation of Functionality

English

  • New Batch Processing Settings: Added new user settings in the UI to control the parallel processing behavior:
    • Enable Batch Parallelism
    • Batch Concurrency
    • Batch Size
    • Delay Between Batches (ms)
    • API Call Interval (ms)
  • New Command: "Create Wiki-Link & Generate Note from Selection":
    • This new command streamlines the knowledge creation workflow.
    • It allows a user to select text, which is then automatically converted to a [[wiki-link]].
    • A corresponding concept note is created in the designated "Concept Note Folder". If the note already exists, a backlink is added instead.
    • Finally, its content is automatically generated by the LLM using the "Generate Content from Title" functionality.
  • New Setting: "Auto-run Mermaid Syntax Fix": Added a setting to automatically run a syntax-fixing pass on notes after content generation tasks to ensure generated Mermaid diagrams are valid.
  • Updated Documentation: Updated README.md and README_zh.md to comprehensively document all new settings and the new "Create Wiki-Link & Generate Note" command.

Chinese (中文)

  • 新增批量处理设置: 在UI中添加了新的用户设置以控制并行处理行为：
    • 启用批处理并行化
    • 批处理并发数
    • 批处理大小
    • 批处理间隔延迟 (毫秒)
    • API 调用间隔 (毫秒)
  • 新命令：“从选中文本创建维基链接并生成笔记”:
    • 这个新命令简化了知识创建的工作流程。
    • 它允许用户选择文本，然后自动将其转换为 [[维基链接]]。
    • 在指定的“概念笔记文件夹”中创建一个相应的概念笔记。如果笔记已存在，则会添加反向链接。
    • 最后，使用“根据标题生成内容”功能，由 LLM 自动为该笔记生成内容。
  • 新设置：“生成后自动修复Mermaid语法”: 添加了一个设置，用于在内容生成任务后自动对笔记运行语法修复，以确保生成的Mermaid图表有效。
  • 更新文档: 更新了 README.md 和 README_zh.md，以全面记录所有新设置和“从选中文本创建维基链接并生成笔记”的新命令。

2026-05-01 — Cline-Aligned Output Token Handling and Diagram Pipeline Hardening

Cline-Aligned Unknown Model maxOutputTokens

Problem: resolveProviderTokenLimit passed the global maxTokens setting (default 8192) for completely unknown models not in KNOWN_MODEL_MAX_OUTPUT_TOKENS. This was incorrect — forcing a fixed cap on models the system doesn't understand risks silently limiting output on high-capability models, or requesting unsupported max_tokens values on constrained models.

Fix: When a model is not known (no entry in KNOWN_MODEL_MAX_OUTPUT_TOKENS) AND the global maxTokens setting equals the default (untouched), the system now returns undefined — letting the API provider decide. This matches Cline's behavior exactly. User-set custom maxTokens values for unknown models are preserved (backward compatibility).

Scope: resolveProviderTokenLimit in src/llmUtils.ts. Affects all transports (OpenAI-compatible, Anthropic, Google, Azure OpenAI, Ollama).

Tests: Added 2 tests in src/tests/llmUtilsProviderSupport.test.ts:

  • Unknown model + default → max_tokens not sent (API decides)
  • Unknown model + custom value → user's value passed through

Diagram Edge Field Normalization

Problem: Live LLM testing against DeepSeek revealed that LLM outputs use source/target edge field names, but DiagramEdge expects from/to. This caused validation failures when LLMs returned valid specs with the "wrong" field names.

Fix:

  • normalizeSpec in src/diagram/diagramSpecResponseParser.ts now normalizes multiple edge field conventions: from/source/sourceId/start → from, to/target/targetId/end → to
  • buildDiagramSpecPrompt in src/diagram/prompts/diagramSpecPrompt.ts now explicitly instructs LLMs to use from/to field names

Live LLM Chain Tests

New test file src/tests/liveChainTest.test.ts (untracked, for manual execution):

  • Tests real DeepSeek API calls from the test vault configuration
  • Verifies chat completion, diagram spec generation, and full diagram pipeline
  • All 5 tests pass against live DeepSeek API