zms618/paper-reading

v0.7.0

A figure-first paper-reading mentor that helps readers understand research and practice scientific thinking.

Release notes

Add paper-grounded research discovery while reading, with four evidence-based entry points: real failures, reverse-engineering successful mechanisms, apparent contradictions, and deployment needs.
Add an optional paper-anchored workflow that connects located experiments to falsifiable problems, null explanations, and low-cost probes while preserving open-problem and single-question routes.
Add separate scientific-depth and research-velocity assessments, along with asset/reproduction checks and conditional GO/STOP planning.
Expand the bilingual project README with the new research workflow, keeping the installation tutorials and acknowledgements.

Advance both Research Idea Discovery plugin manifests and the project guide to v3.5.0 while preserving the stable plugin ID and repository paths.
Add new scout, portfolio, execution-planning, explanation, and method-design Skills, references, examples, checks, and regression tests.

Extend the first overview stage to connect the research problem, paper story, main conclusion, and a verified result snapshot with its limits.
Add an optional research-thinking choice after the third overview stage; readers can discuss problem necessity, key assumptions, and simple alternatives, or proceed directly to deep reading.
Add a post-reading evidence review using Claim → Evidence → Alternative → Falsifier → Next Experiment, with optional research-question cards and presentation practice.
Keep research thinking evidence-led and user-controlled; do not force answers or present an unverified idea as established novelty.

Refresh the bilingual project guides and plugin metadata to describe paper reading as both paper understanding and research-thinking practice.
Add references/research-thinking.md with discussion flows, candidate-question criteria, and optional reporting formats.
Preserve the stable plugin identifier paper-reading, ChatGPT installation guides, screenshots, and acknowledgements.

Build a semantic argument map for all paper figures, tables, algorithms, and teasers, including their roles, first-use stages, and later reuse.
Track all essential visuals for the current stage before declaring it complete; explain cross-figure argument links and allow multiple necessary figures.
Distinguish a true architecture diagram from a pipeline or algorithm listing, and state clearly when no architecture diagram is present.
Add check_figure_coverage.py to audit a human-reviewed figure plan and stage records. It checks supplied records; it does not parse PDFs or verify that a client displayed an image.

Keep the stable plugin ID paper-reading while synchronizing the v0.6.0 reading workflow and helper scripts.
Update project guides and installation links for the renamed no-heartburn-academic-skills repository.

Add conditional one-click next-stage navigation while preserving explicit reader consent and the seven-stage reading flow.
Prefer verified, inline paper figures; distinguish an actual image display from sandbox download links and provide download links only as a fallback.
Add the verified sandbox download-link helper to the PDF crop script.
Add an in-repository screenshot example of a Transformer Self-Attention explanation. Its diagram is an explanatory schematic, not an original paper figure.

Make ChatGPT the primary use case while retaining Codex compatibility; clarify that account/model and client limits still apply.
Keep the package identifier paper-reading and the academic-skills collection structure while updating to the supplied v0.5.3 workflow.
Add tests for stage navigation, preview/download wording, and verified fallback links.

A seven-stage paper-reading guide with a clear boundary between the first three overview stages and the subsequent deep-reading stages.
An optional seventh stage for evidence review, reviewer-style analysis, and research transfer.
crop_pdf_asset.py for rendering a full PDF page or a user-selected crop.
make_figure_card.py for adding a short reading guide around an unchanged, verified figure crop, with a pixel-preservation check.
Plugin manifests, dependency declarations, automated checks, and usage documentation.

The plugin is named paper-reading; its client-facing display name is 论文带读.
The reading framework is preserved from the supplied v0.5.0 source. This release organizes the plugin under the academic-skills repository and aligns its package name, skill path, metadata, and documentation.
The plugin provides instructions and two image utilities. PDF interpretation, web verification, image rendering in chat, and conversation continuity depend on the host AI client and its tools.
No user papers, datasets, chat logs, credentials, or model weights are included.