Skip to content

CHANGELOG.md

Latest commit

 

History

History
86 lines (61 loc) · 9.52 KB

CHANGELOG.md

File metadata and controls

86 lines (61 loc) · 9.52 KB
Raw

Changelog

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

The format is based on Keep a Changelog.

[Unreleased]

[1.4.0] - 2026-05-18

Added

  • /poster skill (/poster [paper-dir]): generate a self-contained 1400×900 HTML academic poster + 2× PNG render from a drafted-and-compiled paper. Pipeline adapted from PaperX (arXiv:2602.03866) — DAG intermediate (tools/wiki2dag.py), per-section distillation, interactive figure picker, critique-revise convergence pass, event-driven Playwright render with programmatic DOM overflow check. Reads paper/main.tex + paper/sections/*.tex + paper/figures/ and grounds distillation in wiki/outputs/paper-plan-*.md when available. Authors are cached in paper/.author_display.txt for double-blind workflows. Supports affiliation/conference logos (PNG/JPG/PDF; PDF auto-converted at 200 DPI) and two header layouts (corners / stacked). Export to PDF from the browser's print dialog.
  • TikZ figure rasterization (tools/rasterize_latex.py): figures defined as inline TikZ (without \includegraphics) are auto-compiled via pdflatex + pdftoppm into PNGs under paper/figures/ and flow through the existing poster figure-injection pipeline. TEXINPUTS is augmented with the source paper_dir so \input{figures/data/foo.tex} and pgfplots .dat references resolve correctly. Optional dependency: TeX Live + poppler-utils (graceful failure with _failed/<slug>/ artifacts saved for debugging).
  • Live HTML <table class="poster-table"> rendering: booktabs tabular environments inside \begin{table}…\end{table} become real HTML tables with cell-level scrubbing (\textbf/\emph/\texttt/\textsc, \multicolumn → colspan, citation key → bibliography ordinal, \ensuremath{X} → $X$ for KaTeX). Tables auto-detect booktabs \midrule to split thead/tbody; tables without \midrule render entirely as <tbody>. \input{tables/foo} is expanded inline (comment-aware, so % \input{...} is ignored). Numeric cells wrapped in $..$ for math/text rendering parity. Table font unified to KaTeX_Main.
  • Multi-figure layouts (Step 2.5): per-section figure picker supports selecting ≥ 2 figures with explicit layout choice — side-by-side (inline-multi-side), vertical stack (inline-multi-stack), or after-table.
  • Optional playwright>=1.40 dependency for event-driven rendering with deterministic font/asset wait semantics. Without it, /poster falls back to a system-browser subprocess render path (Chrome/Edge preferred, Firefox at 1× scale).
  • templates/poster/poster_template.html with 3-column auto-fit layout, KaTeX math support, .poster-table CSS (KaTeX_Main font, alternating striping), and an iterative fit() algorithm that shrinks --base-font until content settles inside the poster bounds.
  • wiki/log.md session entries for poster runs (paper title, section/figure/table counts, header config, refine iterations, check-overflow result).
  • README "What's New" entry with anonymized demo image (assets/poster_demo_tikz_tables.png) — placeholder title/authors/prose, real tables/figures preserved to demonstrate the pipeline.

Changed

  • setup.sh adds import smoke-test for tools/wiki2dag.py and tools/poster.py (tools/rasterize_latex.py transitively validated via wiki2dag module-level import).
  • requirements.txt notes playwright>=1.40 as optional with one-shot python -m playwright install chromium setup hint.

No breaking changes

Existing skills (/init, /ingest, /discover, /paper-draft, /paper-compile, /visualize, etc.) are unchanged. The new /poster skill is purely additive — no existing entity schemas, edge types, runtime contracts, or skill outputs are modified.

[1.3.0] - 2026-05-12

Added

  • /discover venue mode (--venue <slug> --year <int>): rank a conference's full paper list by relevance to the existing wiki. Data source is papercopilot/paperlists public JSON — no Semantic Scholar / DeepXiv calls, no extra API keys. Supports NeurIPS (slug neurips aliases to nips), ICLR, ICML, and any other venue covered by Paper Copilot.
  • 20 unit tests in .sleepcode/tests/test_discover_from_venue.py covering URL aliasing, JSON normalization, transient-failure retry, title-based dedup, sparse-wiki guard, and the no-write contract.

Changed

  • /discover skill argument-hint expanded from 3 to 4 seed modes (anchor / topic / wiki / venue) across .claude/, i18n/en/, and i18n/zh/. References (seed-modes.md, ranking-signals.md, wiki-dedup.md) updated for venue-mode picking, scoring, and title-based dedup against candidates that lack arXiv IDs.
  • Venue mode is stricter than other /discover modes: it does not write to wiki/ at all, including wiki/log.md. Sparse wikis (< 20 unique relevance terms) fail with an actionable error rather than returning an unpersonalized ranking.

[1.1.0] - 2026-05-06

Added

  • Knowledge graph visualization (/visualize): two output modes — Web SPA (Cytoscape, served by tools/serve.py) with BFS-highlight, entity/edge filtering, and in-app Reader; Obsidian mode (--obsidian / --canvas) generating color-coded graph config and force-layout Canvas files with labeled semantic edges.
  • Local research dashboard (#/dashboard in web UI): paper ingestion timeline, method-type breakdown, novelty histogram, and experiment status charts.

Changed (BREAKING)

  • Replace claims entity with methods entity. The wiki no longer carries a separate testable-claim entity; verifiable hypotheses now live on ideas pages, and reusable techniques are first-class via wiki/methods/*.md (with name, slug, type enum, source_papers, parent_methods, child_methods, code_repo).
  • Drop experiments.target_claim; require experiments.linked_idea. Every experiment must reference its source idea; /exp-eval now updates the linked idea's status / pilot_result / failure_reason instead of a separate claim entity.
  • ideas.origin_gaps retargets [concepts, topics] (was [claims]). Reverse links land in concepts.linked_ideas / topics.linked_ideas (new frontmatter fields).
  • ideas schema: gains target_venue, novelty_score (1-5 int). Drops domain. Body template adds ## Novelty argument and ## Target venue; drops ## Expected outcome.
  • papers schema: gains tldr, contribution_type (multi-select from a fixed enum), datasets. Drops keywords, domain. Body renames ## Problem → ## Problem & Context, ## Results → ## Experiment & Results.
  • concepts schema: drops part_of. Gains definition (1-sentence machine-readable) and linked_ideas. Body drops ## Formal notation, ## When to use, ## Key papers; gains ## Relationship to foundations.
  • topics schema: drops my_involvement, sota_updated. Gains linked_ideas. Body drops ## Research gaps, ## Key people, ## My position; gains ## Key benchmarks and splits ## Open problems into ### Known gaps / ### Methodological gaps subsections.
  • people schema: drops tags. Gains research_areas and type: object {kind: enum, subtype: str}. Body drops ## Key papers and ## Collaborators.
  • /novelty: adds --write flag — when set with an idea slug as target, persists novelty_score to the idea page (skill is otherwise still read-only).
  • /paper-plan: argument-hint changes from <claim-slugs> to <idea-slugs>. The plan is built from validated ideas + their linked experiments and methods, not from a separate claim graph.
  • CLI: tools/research_wiki.py removes find-similar-claim, query weak-claims, query evidence-for. The query subcommand now exposes only ready-to-test and orphans.
  • Edge workflow rename: claim_evidence → evidence on supports / contradicts (endpoints unchanged at *->*).

Migration notes

  • Pre-existing wiki/claims/*.md content is not migrated; in this repo the directory was empty so it has been removed. Any external repo with content there should run /edit to fold relevant assertions into the appropriate idea / method pages before pulling this change.
  • The experiments.target_claim field is dropped, not migrated; pre-existing experiment pages must be hand-edited to set linked_idea instead.

[0.1.0] - 2026-04-09

Added

  • 20 Claude Code skills for full research lifecycle: /init, /ingest, /ask, /edit, /check, /daily-arxiv, /ideate, /novelty, /review, /exp-design, /exp-run, /exp-status, /exp-eval, /refine, /survey, /paper-plan, /paper-draft, /paper-compile, /research, /rebuttal
  • Wiki knowledge engine (tools/research_wiki.py) with 20 CLI commands
  • 8 entity types: papers, concepts, topics, people, ideas, experiments, claims, summaries (superseded by 0.2.0 — see Unreleased; claims is replaced by methods)
  • Typed semantic relationship graph (graph/edges.jsonl) plus bibliographic citation layer (graph/citations.jsonl)
  • Daily arXiv automation via GitHub Actions
  • Cross-model review via any OpenAI-compatible API (DeepSeek, OpenAI, Qwen, OpenRouter, SiliconFlow, etc.)
  • Multi-source data integration: arXiv RSS, Semantic Scholar, DeepXiv
  • Remote GPU experiment support (tools/remote.py)
  • Structural wiki linter with auto-fix (tools/lint.py)
  • Bilingual support (English + Chinese) with setup.sh --lang switching
  • One-click setup (setup.sh)
  • Obsidian-compatible [[wikilink]] format throughout
  • 2125 tests