Changelog

[0.6.0] - 2026-09-22

• Add the obsidian-literature-base Skill for building seven Obsidian Base views over MinerU publications and Zotero reading records.
• Generate per-paper catalogue notes with prefixed or AI-assisted topics, cover images, and foldable galleries containing every parsed image.
• Add an optional topic Wiki with a generated literature hub, shared-topic pages, bidirectional Wikilinks, preserved manual synthesis sections, and stale-page reporting.
• Keep MinerU and Zotero managed artifacts read-only, refuse unsafe overwrites, and cover Base, gallery, topic, Wiki, and refresh behavior with automated tests.

[0.5.1] - 2026-09-22

• Rewrite literature-reading as one concise Agent workflow that defaults to complete paper interpretation and keeps focused evidence retrieval within the same Skill.
• Require figure interpretation to combine visual inspection, captions, and the authors' surrounding discussion, with linked MinerU asset previews in optional Markdown output.
• Keep paper-reading responsible only for PDF selection and MinerU publication, and remove the redundant literature-retrieval Skill.
• Align parsing guidance with the runtime order: MinerU cache/upload occurs before publication conflict, exclusion, or recovery checks.

[0.5.0] - 2026-09-22

• Add shared Obsidian-ready bibliographic frontmatter to MinerU Markdown and Zotero reading records, including Zotero deep links.
• Normalize creator names and Zotero tags into Obsidian-compatible list properties while retaining complete source metadata in metadata.json.
• Prefix every parsed image filename and rewritten Markdown image link with the source Zotero key.
• Document the published frontmatter and image naming contract and cover it with unit and live Zotero-to-MinerU publication checks.

[0.4.2] - 2026-09-21

• Pin the release-gate Ruff version and align test helpers with that rule set so validated releases cannot diverge between local and CI environments.

[0.4.1] - 2026-09-21

• Require explicit user confirmation of attachmentKey before parsing or synchronizing Zotero items with multiple PDFs.
• Fix arXiv requests by sending the required Atom and contact headers, and keep Crossref's first search page relevant by omitting the wildcard cursor.
• Mark omnischolar_status as network-capable while preserving live image-model discovery.
• Honor configured sync conflict behavior and conflict directories, reject unsafe conflict paths, and wire cache/recovery controls to their runtime behavior.
• Remove obsolete managed assets during republish, preserve unmanaged user files, and prevent distinct Zotero publications with identical stems from sharing a directory.
• Align the Skills, bilingual documentation, tool counts, latest-package commands, runtime user agents, and release validation workflow.

[0.4.0] - 2026-09-21

• Remove the dedicated literature-reading MCP tools and keep paper interpretation in Skills.
• Keep Zotero access, MinerU parsing, and conservative synchronization as MCP capabilities.
• Rewrite paper-reading, literature-reading, and literature-retrieval Skills around local MinerU Markdown and assets.
• Save interpretation Markdown beside the MinerU source with task-defined filenames and structure.
• Remove fixed structured-analysis output directories, filename templates, and analysis frontmatter generation.
• Keep Zotero reading records under each paper's source/ directory.

[0.3.5] - 2026-09-21

• Fix targeted-reading figure/table normalization to be idempotent.
• Prevent duplicate generation of the four figure-analysis fields.
• Remove image width aliases and keep Markdown figure tables at exactly two columns.
• Preserve user-provided figure interpretations and author statements.
• Add regression coverage for Chinese figure titles and complex Markdown table rows.

[0.3.4] - 2026-09-21

• Fix structured analysis frontmatter so generated documents contain one canonical frontmatter block.
• Render each targeted-reading figure/panel as a two-column Markdown table row using the original MinerU asset embed.
• Keep figure analysis as four ordered fields: title, source location, author statement, and interpretation.
• Remove generated thumbnail asset support and retain only original MinerU images.
• Add regression coverage for repeated frontmatter, table image parsing, original asset links, and ordered figure analysis fields.

[0.3.3] - 2026-09-20

• Add configurable source PDF and styled Zotero reading-record publication for notes and PDF annotations.
• Add source-fingerprinted omnischolar_analysis outputs for single-paper full-read/targeted-reading and multi-paper compare/review workflows.
• Improve Obsidian analysis links, canonical Sources sections, figure embeds, Markdown table layouts, and block-math formula output.
• Fix Zotero local API annotation discovery and filter annotations by their actual parent attachment.
• Update SCI literature Skills and release documentation for language-aware structured reading workflows.

[0.3.2] - 2026-09-20

• Fix Zotero PDF annotation discovery for the local API by querying annotation items by parentItem and filtering returned annotations by their actual parent attachment.
• Add regression coverage for annotation colors, pages, comments, tags, and cross-attachment filtering.

[0.3.1] - 2026-09-20

• Add configurable Zotero source output with copied PDFs and a styled zotero-reading-record.md that separates notes from PDF annotations.
• Add omnischolar_analysis for source-fingerprinted full-read, targeted-reading, compare, and review Markdown outputs under configurable Analysis/Single and Analysis/Multi directories.
• Update literature Skills with SCI-focused single-paper templates, compact multi-paper comparison/review workflows, language-aware output, and source-conflict rules.

[0.3.0] - 2026-09-20

• Add bounded local focused retrieval with BM25 + TF-IDF ranking, paragraph locators, section filters, and explicit empty-result behavior.
• Add persistent bounded reading contexts for multi-turn single-paper and multi-paper workflows.
• Extend literature reading with stable paragraph anchors, figure/table analysis context, multi-paper comparison, review evidence, and cache-aware workflows.
• Add the literature-retrieval Skill and align all Skill frontmatter with the current validator.
• Keep MCP results protocol-compatible without duplicating full payloads in structuredContent; add regression coverage for standard MCP clients.
• Validate real Zotero workflows across three papers, including cache hits, forced reparse/overwrite, full-text pagination, figures, formulas, comparison, review, and fail-closed errors.

[0.2.0] - 2026-09-18

• Keep full parsed Markdown in the local publication while returning bounded parse metadata to agents.
• Add omnischolar_read for cursor-based full-text reading, figures/tables, formulas, paragraph lookup, paper comparison, and literature-review evidence packets.
• Stop duplicating every MCP result in both text and structured content; Pi adapters now keep one canonical result representation.
• Add the literature-reading Skill and make metadata-only Zotero item reads the default.

[0.1.24] - 2026-09-17

• Add named custom provider profiles with independent API keys, endpoints, model capabilities, and image parameter contracts.
• Support custom model-level controls for size, resolution, background, output format, quality, count, and image editing workflows.

[0.1.23] - 2026-09-17

• Extend OpenAI-compatible image request timeouts to 180 seconds by default, with a bounded custom-provider override.

[0.1.22] - 2026-09-17

• Move current GPT Image and Gemini image model presets to the custom provider example; official providers continue to use their own model discovery or curated catalogs.

[0.1.21] - 2026-09-17

• Auto-probe custom OpenAI-compatible providers at baseUrl/models when no catalog endpoint is configured, with safe fallback to explicit model contracts.
• Preset the current official GPT Image and Gemini image model IDs in the example configuration.
• Document custom catalog discovery and capability metadata requirements.

[0.1.20] - 2026-09-17

• Default Fal image generation to official sync_mode=true with an explicit opt-out for hosted-URL queue results.
• Add regression coverage and update scientific-figure routing guidance for CDN-independent Fal artifacts.

[0.1.19] - 2026-09-17

• Distinguished Qwen AI Platform public DashScope endpoints from Bailian workspace endpoints.
• Inferred the Vertex project from credentialsFile.project_id, included Vertex auth dependencies by default, and added a long image-generation timeout.
• Exposed verified provider parameter contracts through omnischolar_image_models and documented provider-specific image controls.
• Mapped Qwen resolution/aspectRatio to the native size parameter and added regression coverage.

[0.1.18] - 2026-09-17

• Added Semantic Scholar author search/detail/papers operations, recommendation enrichment, bounded retries, and Retry-After handling.
• Added image-provider model discovery, explicit model selection, normalized image options, and verified fallback catalogs for providers without discovery APIs.
• Added native Qwen/DashScope image routing, workspace/region configuration, Fal queue URL normalization, Vertex service-account JSON authentication, global Gemini image generation, and Vertex image extraction.
• Updated image-provider documentation, example configuration, scientific-figure routing guidance, and regression tests.

[0.1.12] - 2026-09-16

• Switched all documented and installer-generated MCP launchers to the latest PyPI package form.
• Added complete latest-version update and uninstall commands for Python, MCP, Skills, Codex, Claude Code, and Pi.
• Kept Codex TOML and stdio-host MCP configuration shapes separate and valid.

[0.1.11] - 2026-09-16

• Switched MCP launchers from fixed PyPI versions to @latest.
• Added explicit refresh commands for forcing uv cache updates.
• Added complete install, update, and uninstall guidance for MCP, Skills, Codex, Claude Code, and Pi.
• Updated the host installer to write the latest uvx MCP command with host-correct config shapes.

[0.1.10] - 2026-09-16

• Added _ as a supported shared filename separator alongside - and +.
• Added a separate folder-name template for per-publication directories.
• Applied the shared separator to paper, folder, and parsed-image asset naming.

[0.1.9] - 2026-09-16

• Added title-only and author/title filename template combinations.
• Added configurable attachment image filenames with original-stem and extension variables.
• Added collision and unsafe-name checks for configured attachment filenames.

[0.1.8] - 2026-09-16

• Added configurable Obsidian output subfolders.
• Added configurable filename templates and -/+ separators for new publications.
• Preserved existing manifest paths when output naming settings change.

[0.1.7] - 2026-09-16

• Included the complete configuration-template loader in the published Python package.
• Ensured first-run initialization uses the full editable template on installed releases.

[0.1.6] - 2026-09-16

• Generate the complete editable configuration template on first initialization.
• Reserve direct apiKey and optional apiKeyEnv fields for every credentialed service.
• Include Vertex project, location, and access-token fields plus all image-provider sections.

[0.1.5] - 2026-09-16

• Documented and tested direct apiKey entries in the user configuration.
• Kept apiKeyEnv as an optional environment-variable alternative.

[0.1.4] - 2026-09-16

• Fixed the user configuration directory to use one cross-platform app directory level.
• Stopped reading legacy nested configuration paths; existing legacy files are left untouched.
• Documented environment-variable-based API key setup and platform-specific user paths.

[0.1.3] - 2026-09-16

• Fixed release-version drift between Python, MCP launchers, Pi, and plugin manifests.
• Added automatic creation of one user-level configuration file when no config is discoverable.
• Made the Pi Extension launch the matching PyPI MCP through uvx instead of requiring a separate global CLI.

[0.1.2] - 2026-09-16

Added

• Added the npm-published Pi MCP bridge Extension and Pi Extension + Skills installation flow.
• Added token-free npm trusted publishing through GitHub Actions OIDC.
• Added verified Hermes and WorkBuddy/CodeBuddy MCP templates.
• Included all README-linked documentation in the Pi npm package from version 0.1.1.
• Added a Git-backed Codex marketplace and a one-command plugin MCP path through pinned uvx.
• Added a Claude Code marketplace that installs the same eight Skills and pinned local MCP server.

Fixed

• Removed duplicate paper titles when MinerU already returns a matching top-level heading.
• Renamed parsed figures to deterministic image-N files in first-reference order and rewrote Markdown links before publication.
• Flattened nested MinerU asset paths before the sync publisher's filename safety check.
• Bounded long paper stems with deterministic hash suffixes to avoid Windows path-length failures and truncation collisions.
• Made sync repair, restore, and apply operations reuse valid MinerU cache entries instead of forcing a paid re-upload; only explicit reparse bypasses the cache.
• Reported the stored parser cache key in sync plans for existing publications.
• Added workspace-scoped DashScope and Qwen Cloud endpoints to the example configuration and image-provider documentation.

Changed

• Unified Python, Pi npm, MCP registry, and agent-plugin package versions at 0.1.2.

[0.1.0] - 2026-09-10

Added

• Local Python stdio MCP server with 38 research tools and eight shared Skills.
• Literature search, read-only Zotero, MinerU parsing and Markdown sync.
• Ai4Scholar citation and research tools, Materials Project queries and exports, CAS interface checks, and scientific image generation.
• Host installer with dry-run, backup, rollback, update, uninstall, checksum checks, and MCP handshake validation.
• English and Simplified Chinese user documentation, live validation records, and plugin listing material.

Security

• Separate approval for paid calls and external uploads.
• Limits for responses, files, paths, pagination, polling, and archive extraction.
• Secret removal from status, errors, logs, and tool results.

Fixed after live testing

• PubMed identifier handling and PMC OAI-PMH full-text retrieval.
• Zotero attachment, note, annotation, and indexed-text handling.
• Windows newline recovery and render-change detection in sync.
• Materials Project CIF export.
• Ai4Scholar figure timeout and artifact saving.
• fal queue/data-URI results and DashScope workspace endpoint checks.