Cards with pictures

If the exam shows it as a picture, the card shows a picture. A DFS order asked on an edge list is a bad card; the same question on a drawn graph is a good one.

When a picture beats text

• Algorithms on structures: graph and tree traversals, shortest paths, heaps, AVL rotations, hash tables, sorting states.
• Automata, state machines, recursion trees, circuits.
• Function plots, the geometric meaning of a theorem, vectors and transformations.
• Anatomy, pathways and any labelled diagram: use image occlusion.
• A figure from the lecture or a past exam: cut it out instead of describing it.

Keep text cards for definitions, statements and reasons. A picture has to carry information, not decorate.

How images work

Each image tool saves a PNG under the name you give it into the user's images folder (the result says where), shows you a thumbnail and returns the <img> tag for a card field. Anki gets a copy only when the card is committed. In the chat preview the user sees the file name and can open the file in that folder.

Look at every thumbnail. Redo the image if something is cut off, too small to read on a phone, or gives the answer away.

If overview and the other anki tools are not among your tools, call get_device_info (search for it if it is not loaded) and search for the anki tools again. In the Claude app, a chat reaches the tools on the user's computer reliably only if the user added a folder to it before sending the first message. If the tools are still missing, ask the user to start a new chat, add a folder with the Add folder button (for example the folder with their study material) and send the request again. Only if that does not help either, follow the check-setup skill.

The anki tools run on the user's computer and open files there by absolute path. A file attached to the chat, or a plot you made in a sandbox of your own, is not on that computer: save it there first if you can, otherwise ask the user where the file is.

Getting an image

• Graphs, trees, automata: render_graph with DOT source and a name. For a front and a back version (for example the visit order only on the back), call it twice with the same nodes and edges and labels of the same length, so both drawings line up.
• Figures from a PDF: pdf_figures lists the figures on the given pages with id, region and caption. show_page=true adds a picture of the page with the figures boxed and a 10 % grid. Then call stage_image with the PDF and the figure id, or with page and a region [left, top, right, bottom] read off the grid as fractions.
• Image files and URLs: stage_image with the path or URL. region crops it.
• Plots: draw them with matplotlib in this style, save the PNG on the user's computer and pass its path to stage_image. If you need a Python with matplotlib, uv run --no-project --with matplotlib python plot.py gives you one.

  import matplotlib.pyplot as plt
  
  plt.rcParams.update({
      "font.size": 15, "lines.linewidth": 2.4, "axes.grid": True, "grid.alpha": 0.3,
      "axes.spines.top": False, "axes.spines.right": False,
      "axes.prop_cycle": plt.cycler(color=["#0072B2", "#D55E00", "#009E73", "#CC79A7", "#E69F00"]),
  })
  fig, ax = plt.subplots(figsize=(6, 4))
  ax.plot(...)
  fig.savefig("tangent.png", dpi=200, bbox_inches="tight", facecolor="white")

Image occlusion

draft_add_occlusion turns a saved image into an Image Occlusion note (Anki 23.10 or newer).

• A mask is {"label": "Mitralklappe"} for text on a PDF figure or a graph node, or {"region": [left, top, right, bottom]} in fractions of the image for anything else.
• Every mask becomes its own card. Masks with the same group are asked together on one card.
• hide_all=true (the default) hides all masks and asks for one. Use false when the other labels are needed as hints.
• Check the returned picture: every mask has to cover exactly its label. The user finds the same picture as <name>-masks.png in the images folder.
• If there is no Image Occlusion note type, follow the check-setup skill.

Style

• White background, black lines, at most three accent colours.
• Readable on a phone: labels at least 14 pt at the saved size.
• Nothing on the front may give the answer away: no visit order, no highlighted path, no solution in a caption.
• One idea per picture. Crop away slide titles, page numbers and logos.