Run2 pdf calendar parsing
[COLM'26] SkillLearnBench is the first benchmark for evaluating continual learning methods that automatically generate agent skills.
npx -y skills add cxcscmu/SkillLearnBench --skill run2_pdf-calendar-parsingAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
What its author says it does
Copied from the file, not written here
Extract calendar events from PDF by mapping colored rectangles to a 15-min grid using PyMuPDF.
SKILL.md
1.6 KB, as published. Nobody here has run it
PDF Calendar Parsing with PyMuPDF
Installation
pip install pymupdf
Approach
- Use
page.get_drawings()to extract all drawn shapes. - Identify the 15-min grid: horizontal lines with
width > 100andheight < 3. Sort by y-position. The first line is time 0 (12:00 AM), each subsequent line is +15 min. - Identify event blocks: filled rectangles with
height > 5andwidth > 100. - Map each block's top/bottom y to the nearest grid line:
slot = round((y - grid[0]) / spacing), time =slot * 15minutes from midnight. - Blue blocks: check if
fill[2] > 0.6andfill[0] < 0.55(blue-dominant). These are low-priority and overwritable.
Grid line to time mapping
lines = sorted([d["rect"].y0 for d in drawings if d["rect"].width > 100 and d["rect"].height < 3])
spacing = lines[1] - lines[0] # ~7.5 px
def y_to_slot(y):
return round((y - lines[0]) / spacing)
def slot_to_time(slot):
total_min = slot * 15
h, m = divmod(total_min, 60)
return h, m # 24-hour format
Block padding
Blocks have ~1px padding on each side, so round() to nearest slot gives correct results. Always verify by checking that durations are multiples of 15 minutes.
Identifying block colors
- Gray (out of office):
fill ≈ (0.38, 0.38, 0.38) - Red/Pink (meetings):
fill ≈ (0.90, 0.49, 0.45) - Blue (low-priority):
fill ≈ (0.47, 0.53, 0.80)— OVERWRITABLE - Yellow/Green (labels):
fill ≈ (0.75, 0.79, 0.20)— these are small label squares inside blocks, ignore