Skip to content

CHANGELOG.md

Latest commit

 

History

History
273 lines (211 loc) · 17.3 KB

CHANGELOG.md

File metadata and controls

273 lines (211 loc) · 17.3 KB

Changelog

All notable changes to this project will be documented in this file.

This project follows Keep a Changelog and Semantic Versioning.

版本控制流程说明见：documents/version_control_plan.md

Unreleased

Added

  • Added message-sidecar metadata for structured citations and artifact presence.

Changed

  • Unified citation stripping with structured source retention across capture, reader, and export.
  • Hardened Qwen and Yuanbao parser alignment against current live DOM structures.
  • Added a reproducible Playwright auth, storage-state, and DOM sampling bootstrap workflow.

Fixed

  • Fixed ChatGPT thinking UI so it no longer splits a single assistant reply into multiple logical messages.
  • Fixed reader rendering so citation and search noise no longer appears as abrupt tail text after the main reply body.
  • Fixed Qwen and Yuanbao capture drift against current production DOM structures while keeping tables, math, and code paths intact.

Docs

  • Added DOM sampling bootstrap and platform handoff notes for thinking-boundary repair, citation governance, and five-platform parser sampling.

Chore

  • Added Playwright local auth and DOM sampling tooling for repeatable capture verification.

[1.2.0-rc.8] - 2026-03-08

Changed

  • Updated the internal GET_ALL_EDGES contract so Network can request edge computation for its current node set instead of relying on previously opened Library conversations.

Fixed

  • Restored Network graph edges by lazily ensuring vectors for the active node set when opening the Network view, removing the hidden dependency on Library detail opens.

Chore

  • Bumped extension release metadata to 1.2.0-rc.8 for tagging and distribution alignment.

1.2.0-rc.7 - 2026-03-07

Added

  • Added a shared platform normalization helper for runtime-facing Yuanbao identity handling with legacy YUANBAO compatibility.
  • Added a web dashboard Appearance section in the avatar settings drawer with an explicit dark-mode toggle.

Changed

  • Canonicalized platform naming from YUANBAO to Yuanbao across runtime types, storage reads/writes, sidepanel filters, and @vesti/ui platform metadata.
  • Migrated stored conversation platform values to Yuanbao and kept short-term legacy normalization for persisted inputs.
  • Restored web dashboard badges to the solid-fill platform style while keeping dock / Threads visuals on their existing token system.
  • Synced web dashboard theme state with dock appearance via shared vesti_ui_settings.themeMode updates.

Fixed

  • Realigned Kimi capture to the current DOM structure, excluded chat-header pollution, and restored cold-start transient availability.
  • Realigned Yuanbao capture to the current hyc-* DOM and merged CoT + final response into a single archived AI turn without doc-title noise.
  • Hardened ChatGPT capture with selector+anchor extraction, startup warm capture, and content cleaning for noisy toolbars and controls.
  • Fixed reader AST rendering so structured math, tables, lists, and code blocks preserve their intended layout instead of collapsing into plain paragraphs.
  • Hardened Qwen capture for Monaco code blocks and custom markdown paragraph spacing so code and flowing prose render cleanly in Threads Reader.
  • Declared a local @vesti/ui dashboard type shim to unblock rc7 typecheck/build on the current mainline.
  • Fixed Kimi web badge contrast so light mode uses dark text and dark mode uses white text.
  • Fixed web dashboard theme refresh so dock-initiated appearance changes propagate without reopening the page.

Docs

  • Updated Kimi / Yuanbao Phase 3 capture docs and execution notes.
  • Updated the UI refactor spec to document web-vs-dock badge decoupling and dashboard theme sync.
  • Added the rc7 engineering handoff for the Yuanbao web dashboard and capture-engine follow-up context.

Chore

  • Prepared release metadata and mirror packaging for v1.2.0-rc.7.

1.2.0-rc.6 - 2026-03-07

Added

  • Added Kimi and Yuanbao capture entrypoints (frontend/src/contents/kimi.ts, frontend/src/contents/yuanbao.ts) with transient status + force-archive handlers.
  • Added Kimi and Yuanbao parser modules with selector+anchor extraction, strict session-ID gating, parse stats logging, and v1.2 governance compatibility.

Changed

  • Expanded capture host routing for Kimi to include www.kimi.com + kimi.com (with transitional kimi.moonshot.cn compatibility) and Yuanbao yuanbao.tencent.com for GET_ACTIVE_CAPTURE_STATUS and FORCE_ARCHIVE_TRANSIENT.
  • Expanded platform enum/theme mapping to 8 platforms (ChatGPT/Claude/Gemini/DeepSeek/Qwen/Doubao/Kimi/Yuanbao).
  • Added sidepanel + capsule light/dark token mapping for Kimi/Yuanbao while preserving existing Threads layout structure.
  • Updated extension manifest host permissions and web-accessible matches for Kimi/Yuanbao domains.

Fixed

  • None yet

Docs

  • Updated documents/capture_engine/v1_3_platform_expansion_spec.md for Phase 3 (Kimi + Yuanbao).
  • Added documents/capture_engine/v1_3_phase3_manual_sampling_checklist.md.
  • Added documents/capture_engine/v1_3_phase3_execution_log.md.
  • Updated documents/ui_refactor/v1_4_ui_refactor_component_system_spec.md with 8-platform token contract and Yuanbao naming rule.

Chore

  • None yet

1.2.0-rc.3 - 2026-02-27

Added

  • None yet

Changed

  • Updated @vesti/ui build chain to use a stable invocation path without pnpm exec esbuild dependency coupling.
  • Added explicit esbuild dependency in frontend to guarantee CI binary availability during packaging.
  • Refreshed frontend dependency sync after prebuild so file: package snapshots stay aligned in CI packaging runs.

Fixed

  • Stabilized CI release packaging path to avoid intermittent esbuild not found and @vesti/ui resolve failures.

Docs

  • None yet

Chore

  • None yet

1.2.0-rc.2 - 2026-02-25

Added

  • Added v1.5-lite floating capsule shell baseline with Shadow DOM isolation, collapsed/expanded views, runtime status polling, and in-capsule action wiring (Archive now, Open Dock).
  • Added host-scoped capsule settings service persisted to chrome.storage.local under vesti_capsule_settings.

Changed

  • Expanded capsule primary rollout hosts from partial whitelist to all currently supported chat platforms (ChatGPT, Claude, Gemini, DeepSeek, Qwen, Doubao).
  • Kept draggable hotfix behavior as default: collapsed capsule is directly draggable with 5px anti-misfire threshold and post-drag click suppression.
  • Polished expanded capsule card UI to match the approved prototype skin (spacing, hierarchy, status rows, metric cards, and action buttons) while preserving v1.5-lite state semantics.
  • Synced expanded capsule light/dark rendering to vesti_ui_settings.themeMode with runtime storage-change updates and listener cleanup on destroy.
  • Aligned platform badge color tokens to sidepanel light/dark triplets (bg/text/border) for ChatGPT, Claude, Gemini, DeepSeek, Qwen, and Doubao.
  • Aligned expanded-card typography with sidepanel stacks (--font-ui, --font-vesti-serif equivalent), including serif + tabular numerics for Messages/Turns large metrics.
  • Rolled back collapsed capsule theming to fixed light appearance (independent from theme mode) while keeping expanded-card theme switching enabled.

Fixed

  • Fixed capsule drag regression that blocked viewport repositioning on primary hosts.
  • Disabled native image drag on capsule logo to avoid dragging into composer inputs as text/image payload.

Docs

  • Added engineering handoff documents/engineering_handoffs/2026-02-25-v1_5_lite_capsule_rollout_and_ui_closeout.md covering implementation log, release plan, and architecture details.

Chore

  • None yet

1.2.0-rc.1 - 2026-02-25

Added

  • Gemini/DeepSeek Phase1 capture entrypoints (frontend/src/contents/gemini.ts, frontend/src/contents/deepseek.ts) with transient status + force-archive handlers.
  • Formal Gemini/DeepSeek parser modules with selector+anchor strategies, noise cleaning, role inference, parse stats logging, strict session ID extraction, and source_created_at best-effort extraction.
  • Doubao/Qwen Phase2 capture entrypoints (frontend/src/contents/doubao.ts, frontend/src/contents/qwen.ts) with transient status + force-archive handlers.
  • Formal Doubao/Qwen parser modules with selector+anchor strategies, role inference fallbacks, strict session ID extraction, and parse stats logging.
  • v1.6 dual-track AST foundation: strict ast_v1 node contract, shared DOM-to-AST extractor (P0 full coverage, P1 math/table probes for ChatGPT/Claude/Gemini), parser perf fallback controller, and Reader-side AST renderer component with KaTeX support.

Changed

  • Extension host permissions now include https://gemini.google.com/* and https://chat.deepseek.com/*.
  • Extension host permissions now also include https://www.doubao.com/* and https://chat.qwen.ai/*.
  • Background capture host resolver now recognizes Gemini/DeepSeek for GET_ACTIVE_CAPTURE_STATUS and FORCE_ARCHIVE_TRANSIENT.
  • Background capture host resolver now recognizes Doubao/Qwen for GET_ACTIVE_CAPTURE_STATUS and FORCE_ARCHIVE_TRANSIENT.
  • Capture observability tightened: gate and pipeline logs now include platform/session + decision metadata.
  • Settings capture-status guidance now references all supported capture platforms (ChatGPT/Claude/Gemini/DeepSeek/Doubao/Qwen).
  • Added internal turn_count semantics (AI replies) to conversation capture/persistence and upgraded Smart minTurns evaluation to the same AI-turn metric.
  • Timeline/Insights counters now display X messages · Y turns; Reader header now labels count as messages.
  • Platform badge color tokens are now unified to six Metro theme colors (ChatGPT/Claude/Gemini/DeepSeek/Qwen/Doubao).
  • Release-line governance is split into serial tracks: v1.6 data pipeline, v1.7 multi-agent/prompt backend, v1.8 Reader+Insights UI.
  • Insights page now keeps v1.8.1 grouped IA (On-demand, Scheduled, Discovery) and upgrades Weekly Digest to a dynamic state machine (idle/generating/ready/sparse_week/error) with phase-track generation feedback, previous-natural-week Mon-Sun range unification, and local idle-list collapse (N more/Collapse).
  • Thread Summary pipeline is now aligned to the latest skill contract while keeping conversation_summary.v2 naming: parser and adapter support both legacy v2 shape and upgraded v2 shape (thinking_journey[], real_world_anchor, glossary-style key_insights[]), and Insights Thread Summary UI now renders the full structured journey view with generation shell + no-flash regeneration behavior.
  • Capture persistence upgraded to schema v5-compatible writes (content_ast, content_ast_version, degraded_nodes_count) with legacy-safe read defaults; Reader now uses hybrid AST-first rendering with plain-text fallback; JSON export now carries AST fields as optional extensions.
  • Floating capsule now supports in-page drag interaction with a 5px anti-misfire threshold and a 90%-scaled diameter (43.2px).

Fixed

  • Gemini title extraction now prefers [role='heading'], removes You said prefix for title-only parsing, and falls back safely when generic headings are detected.
  • Corrected turns/message mismatch in sidepanel views and active capture status snapshots.
  • DeepSeek parser now supports .ds-message-based DOM (no <main> requirement), hybrid class role inference, and explicit /a/chat/s/<id> session path extraction.
  • Insights accordion header descriptions (Thread Summary, Weekly Digest) now keep compact one-line truncation when closed, expand to two-line readable copy when open, and expose full text via tooltip.

Docs

  • Updated documents/capture_engine/v1_3_platform_expansion_spec.md with strict-ID alignment for Phase1 execution.
  • Added documents/capture_engine/v1_3_phase1_execution_log.md.
  • Updated documents/capture_engine/v1_3_platform_expansion_spec.md with Phase2 execution profile and strict host scope.
  • Added documents/capture_engine/v1_3_phase2_execution_log.md.
  • Added documents/capture_engine/v1_3_phase2_manual_sampling_checklist.md.
  • Added documents/reader_pipeline/v1_6_data_pipeline_dual_track_spec.md.
  • Added documents/reader_pipeline/v1_6_ast_probe_cheat_sheet.md.
  • Added documents/reader_pipeline/v1_6_schema_v5_migration_spec.md.
  • Added documents/reader_pipeline/v1_6_performance_fallback_spec.md.
  • Added documents/reader_pipeline/v1_6_manual_sampling_and_acceptance.md.
  • Added documents/orchestration/v1_7_multi_agent_orchestration_spec.md.
  • Added documents/orchestration/v1_7_runtime_event_contract.md.
  • Added documents/orchestration/v1_7_feature_flag_rollout_spec.md.
  • Added documents/orchestration/v1_7_manual_sampling_and_acceptance.md.
  • Added documents/prompt_engineering/v1_7_prompt_as_code_contract.md.
  • Added documents/prompt_engineering/v1_7_prompt_schema_drift_gate.md.
  • Added canonical v1.7 prompt files: documents/prompt_engineering/thread-summary-skill.md and documents/prompt_engineering/weekly-digest-skill.md.
  • Added temporary alias note: documents/prompt_engineering/synthesis_skill.md.
  • Updated documents/prompt_engineering/compaction-skill.md to Agent A v2 contract (volume rigidity, empirical anchoring veto, subject isolation, sparse degradation rules).
  • Added documents/prompt_engineering/compaction-skill-rubric.md (scoring matrix + veto rules split from runtime prompt).
  • Updated documents/prompt_engineering/v1_7_prompt_as_code_contract.md with Agent A prompt/rubric layering rules.
  • Updated documents/prompt_engineering/v1_7_prompt_schema_drift_gate.md with Agent A specialized drift checks.
  • Aligned v1.7 planning docs to the new schema matrix: defaults conversation_summary.v3 and weekly_lite.v2, with one-cycle legacy coexistence for v2/v1.
  • Added documents/ui_refactor/v1_8_1_insights_ui_refactor_spec.md.
  • Added documents/ui_refactor/v1_8_1_insights_state_machine_contract.md.
  • Added documents/ui_refactor/v1_8_1_insights_manual_sampling_and_acceptance.md.
  • Updated documents/ui_refactor/README.md with v1.8.1 Insights package inventory.
  • Updated documents/prompt_engineering/insights_prompt_ui_engineering.md with v1.8.1 Weekly dynamic state machine, previous-natural-week range contract, and forward-compatible section rendering rules.
  • Added documents/ui_refactor/v1_8_2_thread_summary_ui_refactor_spec.md.
  • Added documents/ui_refactor/v1_8_2_thread_summary_state_machine_contract.md.
  • Added documents/ui_refactor/v1_8_2_thread_summary_manual_sampling_and_acceptance.md.

Chore

  • Added CI workflow /.github/workflows/prompt-schema-drift-pr.yml for strict mock prompt-schema drift gating on PR changes.
  • Added CI workflow /.github/workflows/prompt-live-smoke-nightly.yml for scheduled live smoke evaluation with optional secrets.

1.1.0-rc.4 - 2026-02-15

Added

  • Timeline conversation cards now support inline title rename with persistence (Pencil, Enter/Blur save, Esc cancel).
  • Added dedicated Data tab in Dock with full Data Management panel.
  • Added local telemetry action type rename_title for card action tracking.
  • Added unlimitedStorage permission and storage limit helpers (900MB warning, 1GB write block).
  • Added export serializer layer for json/txt/md with unified payload (content, mime, filename).
  • Added bundled serif assets: frontend/public/fonts/TiemposHeadline-Medium.woff2, frontend/public/fonts/TiemposText-Regular.woff2.

Changed

  • Settings page now keeps model access controls and links to Data Management instead of duplicating full data actions.
  • Sidepanel typography contract tightened (vesti-page-title, vesti-brand-wordmark) with preload/fallback behavior.
  • Export pipeline uses structured EXPORT_DATA format responses across runtime handlers.
  • Release artifacts refreshed from commit a9e1572.

Fixed

  • Settings toggle thumb vertical alignment is center-locked (44x24 track, 20x20 thumb, no Y-axis jitter).
  • Duration utility ambiguity resolved to explicit transition duration syntax in key UI controls.

Docs

  • Added documents/engineering_data_management_v1_2.md.
  • Updated documents/prompt_engineering/insights_prompt_ui_engineering.md to v1.2-ui-pre.6.
  • Added UI guardrail section in archive/frontend_prototypes/frontend-prompting-system.md.

Release Artifact

  • release/Vesti_MVP_v1.1.0-rc.4.zip
  • SHA256: B86BF1B8BC4064504D1CA850A4A80CCD8FEFAFD93E723635FD86E2D2D99F7AF6
  • Built at: 2026-02-15 21:38:32 +08:00

1.0.0 - 2026-02-11

Added

  • MVP 基线发布（Local-First）
  • ChatGPT / Claude 会话捕获、IndexedDB 存储、Sidepanel Timeline/Reader
  • ModelScope 摘要与周报基础能力（MVP 范围）

Notes

  • v1.0.0 作为后续版本治理起点；从该版本之后，统一采用分支 + annotated tag 发布。