Skip to content

Releases

0.10.0

Choose a tag to compare

@github-actions github-actions released this 27 Oct 20:04
c21d2ef

Added

  • A PDF viewer setting chooses where PDFs open: in Obsidian, as before, or in your system's default PDF app. With System default app, clicking a link to a PDF, its canvas card or its name in the file list opens the file in that app instead of an Obsidian tab. PDFs embedded in a note still show inline.

Fixed

  • PDFs: download now saves to the Default download folder you set, even on a canvas that has downloaded before. Until now a canvas kept using the folder of its last download, so changing the setting had no effect on it. The folder a canvas remembers is still used when the setting is empty.

0.9.0

Choose a tag to compare

@github-actions github-actions released this 22 Oct 21:42
df1b766

Added

  • Canvas: update from collection adds the references you have put in a canvas's Zotero collection since the canvas was created, so a canvas you have rearranged or annotated no longer has to be rebuilt to catch up. Papers already on the canvas are skipped, nothing on it moves, and the new cards arrive in a New from Zotero group beside it, with literature notes and citation arrows to the papers already there. Relayout and Group papers by tag remove that group again when they rearrange the papers. The closing notice says how many papers were added and how many could not be found on Semantic Scholar.
  • Canvas: group papers by tag rearranges the open canvas into one labelled group per tag, read from the tags frontmatter of each paper note. Groups overlap, so a paper with up to three tags sits inside the groups of exactly its tags. A paper with more tags, or one whose combination of tags no arrangement of rectangles can give a region of its own, goes in a separate group and is named in a notice. Untagged papers go in an Untagged group, and arrows, reading-status colours and your own nodes are left as they were. New paper notes start with an empty tags list, and Canvas: relayout removes the tag groups when it returns the canvas to the timeline.
  • A paper's PDF stored in Zotero can now be opened from its note. Whenever papers are added to a canvas, each one is looked up in the Zotero desktop app by DOI, and a PDF Zotero holds for it is linked from the note's pdf property (a vault link when the file is inside the vault, a file:// link otherwise). An existing pdf value is never replaced. The new command PDFs: link from Zotero runs the check again for every paper on the canvas, for instance after starting Zotero; the setting Link Zotero PDFs turns the automatic check off.
  • Downloading a PDF now links it from the paper's note, in a pdf property, so the file opens straight from the note. A link already there is kept, so downloading again adds nothing.
  • A linked PDF, downloaded or found in Zotero, also gets an Open PDF link in the note body, below the arXiv line, so it can be opened from the paper's card on the canvas. A note whose pdf property was set earlier gets the line the next time a PDF is linked for it, for instance by running PDFs: link from Zotero.

Changed

  • Downloaded PDFs no longer keep #, ^, [ or ] from the paper's title in their filename, since a link to the file could not contain them. A PDF downloaded earlier under such a name is no longer recognised as downloaded.

0.8.0

Choose a tag to compare

@github-actions github-actions released this 07 Oct 15:35
5efc297

Added

  • The Expand paper window can sort its list by citation count, year, title, or the order the paper cites its references, each ascending or descending. The order in the paper comes from Crossref, so it is available for papers with a DOI whose publisher deposited a reference list; for a paper expanded before this version, run Papers: expand paper (force refresh) once to fetch it.

Fixed

  • Expand paper no longer saves an incomplete reference list when Semantic Scholar cannot be reached, which used to leave the papers only it knows about missing from every later expand of that paper. The window now says the list may be incomplete, and expanding again later fetches the full list. Resolve missing citation edges likewise reports such papers and asks again on its next run.
  • Sorting the Expand paper window by order in the paper puts nearly every reference in its place. References the publisher deposited without a DOI are now placed by title, and so are those whose deposited DOI belongs to a different paper, which used to put the wrong paper in that slot and the right one at the end.
  • A reference that two sources list once with a DOI and once without, such as a book or an arXiv preprint, now appears once in the Expand paper window instead of twice.

0.7.2

Choose a tag to compare

@github-actions github-actions released this 07 Oct 13:12
b3e4add

Changed

  • The README, which Obsidian shows on the plugin's page, is now a short overview: what the plugin does, how to start, what to set up, and what it sends where. Everything it used to cover in detail is in the full guide.

0.7.1

Choose a tag to compare

@github-actions github-actions released this 07 Oct 08:16
b42cd65

Fixed

  • Write summary with the Claude CLI now reads the paper's PDF. The PDF's path was passed in a way the CLI ignores, so the model never saw the file and wrote its summary from the title and metadata alone. The path is now part of the prompt, and the summary is based on the paper's content.

Changed

  • The Claude CLI now runs with only the tools a command needs: reading files, limited to the PDF's own folder, for Write summary, and web search for Recommend papers when Search the web is on. It no longer gets a shell, editing tools or MCP servers, and your own Claude Code settings, permission rules and hooks are not applied to these runs. A PDF carrying hidden instructions for the model can no longer make it read other files on your computer.

Added

  • Release files now carry GitHub build attestations, so you can check that the main.js, manifest.json and styles.css you downloaded were built from this repository by its release workflow (gh attestation verify main.js --repo TheR4iner/obsidian-citation-graph).

0.7.0

Choose a tag to compare

@github-actions github-actions released this 06 Oct 22:27
55ca37f

Changed

  • PDFs are now kept in a folder inside your vault, and the plugin reads and writes only vault files. Download saves to a vault folder (created if it does not exist, with your folders suggested as you type), and Write summary reads PDFs back from it. A download folder you set in an earlier version keeps working if it is inside the vault, even when written as an absolute or ~ path. One outside the vault is refused with a message saying what to do: choose a vault folder in the download dialog or in Default download folder, and move the PDFs you already have into it. If you do not want the PDFs committed to git or synced, add the folder to your .gitignore or to Obsidian Sync's excluded folders; the README shows how.
  • Every Citation Graph setting now appears in Obsidian's settings search.
  • Citation Graph now needs Obsidian 1.13.0 or later. Older versions of Obsidian keep receiving 0.6.2.
  • The Claude CLI is now found by searching ~/.local/bin first and then PATH when Claude CLI path is blank, as before, but a configured path is no longer checked for existence ahead of time. A path to a missing file is reported when Write summary or Recommend papers runs, with the path it tried.

0.6.2

Choose a tag to compare

@github-actions github-actions released this 06 Oct 21:33
cee0e2e

Fixed

  • No change in what the plugin does. Obsidian's community directory showed around three hundred code-quality warnings on this plugin's listing; they came from its automated review not finding the type definitions for Node.js, which the plugin uses to read and write PDFs and to talk to Zotero. The code now points to those definitions itself, so the review checks it as written and the warnings are gone.

0.6.1

Choose a tag to compare

@github-actions github-actions released this 06 Oct 15:31
4f78470

Fixed

  • Add paper by DOI or arXiv now accepts an arXiv ID written with its arXiv: prefix (arXiv:2409.00998), a doi: prefix, an arxiv.org PDF link, and a markdown link copied out of a note. Before, any of these was looked up as if it were a DOI and reported as not found, even for a paper every source knows.
  • Adding a paper by DOI or arXiv ID, expanding a paper, and resolving missing edges no longer look frozen while Semantic Scholar is rate limiting. Semantic Scholar sometimes refuses requests for a while, and the plugin waits and retries for up to about a minute before giving up. During that wait these commands showed nothing, so a lookup that was about to succeed looked like one that had failed silently, and running it again only queued a second lookup behind the first. They now keep a notice up with an elapsed-time clock and say when they are waiting out a rate limit, and every such wait is written to the log file.

0.6.0

Choose a tag to compare

@github-actions github-actions released this 03 Sep 08:27
eac93a7

Changed

  • The Claude CLI is now given only the environment it needs. A program started by another normally inherits every variable in it, which meant handing the CLI every key you had exported: your Zotero key, your OpenAI key, anything else in the shell Obsidian was launched from. It now receives PATH, HOME, the temporary directory and its own ANTHROPIC_* and CLAUDE_* settings, and nothing else. It is still launched without a shell, so nothing in a prompt or a paper title can become a command, and the path you configure now has to be an absolute path to a file that exists, or the bare name claude, rather than being passed through as typed.
  • Every path the plugin touches outside your vault is now built and checked in one place, and has to resolve inside a folder you named. A filename that would escape its folder is refused before anything is opened. Paper titles, author names and arXiv IDs all arrive from remote services and end up in filenames, so this is checked rather than assumed.
  • The README now says what all of the above means for you, alongside the existing note on which services are contacted.

Fixed

  • Expanding a paper no longer trips OpenAlex's, CrossRef's or arXiv's rate limits. Each of those has a minimum gap the plugin is supposed to leave between requests, and the spacing collapsed whenever two requests were started together: both measured the gap from the same moment, waited the same time, and then fired at the same instant. Expanding any paper that has a DOI does exactly that, asking OpenAlex for its references and the works citing it at once, so the gap was never actually left. Requests are now queued one behind the next.
  • A malformed reply from OpenAlex, CrossRef, Zotero or a language model now costs the field it was carrying rather than the whole command. A value of an unexpected type reads as missing, so an expansion or a summary finishes with one field blank instead of stopping partway with an unhelpful error.
  • A paper whose CrossRef record dates it only by its print or online publication date now shows that year, rather than no year at all.
  • The download window's Download to box now fills the width of the dialog, so a long path is readable without scrolling inside a narrow field.

0.5.0

Choose a tag to compare

@github-actions github-actions released this 01 Sep 12:37
fa4c592

Fixed

  • Cycling the reading status of a paper that has a summary or notes in it now works. Such a paper was labelled Read + notes written whatever its actual status, so cycling it appeared to do nothing, or to flip between two labels at random. Two separate causes: Read + notes written was derived from the note's content alone, which meant the plugin's own Write summary silently marked papers as read; and the cycle read the current status from Obsidian's metadata cache, which lags behind writes, so a second press could decide from the status before the first press.
  • Read + notes written now means both halves: a paper reaches it when it is marked Read and its note has content of its own. A paper you have never opened is no longer labelled as read because something was written into its note.

Changed

  • Reading status is now drawn entirely from the plugin's stylesheet, so a theme or a CSS snippet can restyle any status without the plugin's help. Nothing looks different: the same labels, the same thicker frame on a paper you have started, the same dashed and faded treatment for an abandoned one. The status reaches the canvas as a citation-graph-status-* class on the literature note instead of being inferred from the node's colour, which has two visible consequences. Two statuses sharing a colour now read differently, by their labels, where before they were indistinguishable. And a status you left on No colour is now labelled correctly, where before it fell back to "To read".

  • Citation Graph now requires Obsidian 1.7.2 or later. The manifest previously claimed 1.4.0, which was never true: the settings tab has always needed 1.4.10, deleting a paper has needed 1.6.6, and the reference cache and log file need 1.7.2. On an older Obsidian the plugin failed to load rather than saying why. versions.json has been corrected for every past release for the same reason.

  • The README now sets out, in one place, every remote service the plugin contacts and what is sent to it, why PDFs live outside your vault, that the claude binary is the one already on your machine and nothing is ever downloaded or installed for you, and which features bill to your own account. There is no telemetry of any kind.

  • A few buttons and labels now read in sentence case: Download all, Skip these, Proceed with all, and the download picker's Downloaded and No ID yet badges.