reading-cloze-builder

SKILL.md

Reading Cloze Builder

Create a completed Day_test-style worksheet from an article and keep the five tasks internally consistent.

Workflow

  1. Read the source markdown and identify the article body plus the five task slots.
  2. Read assets/group_format_template.md before generating Task 1.
  3. Extract 20 core words that most affect comprehension.
  4. Mark all 20 words in the full article with ==word==.
  5. Select 10 words from the 20-word group for testing.
  6. Replace those 10 words once each with numbered blanks in reading order.
  7. Produce Task 5 by marking those same 10 words with ==word== in the original article.
  8. Save the completed markdown as a new file instead of overwriting the original unless the user explicitly asks.

Selection Rules

Task 1: 20 core words

Follow these rules:

  • Exclude proper nouns such as people, countries, dates, and publication names.
  • Keep difficulty at B1 or below when possible.
  • Prefer words that affect comprehension of policy, action, cause/effect, attitude, and evidence.
  • Keep inflected forms if the article uses them and that form matters for downstream masking.
  • Strictly follow assets/group_format_template.md for the Group block layout.
  • Keep the same visual style as the template, including the heading pattern, frequency tag, Chinese gloss line, example sentence line, optional collocation/translation/derived-word lines, and separator line.
  • If some fields are unavailable, preserve the template structure as closely as possible and omit only the missing content itself.

Task 2: full-group masking

  • Mask every Task 1 word in the article with ==word==.
  • Preserve the original paragraphing and punctuation.
  • Keep the article readable; do not restructure sentences.

Task 3: 10 assessment words

Select 10 words from Task 1 using these constraints:

  • Do not select more than one word from the same sentence.
  • Spread selections across the article as evenly as possible.
  • Mix parts of speech when possible.
  • Favor words that test comprehension, spelling, grammar form, or context.
  • Keep the exact surface form that will later be blanked or highlighted.

Task 4: numbered cloze blanks

  • Replace each of the 10 Task 3 words exactly once.
  • Number blanks in reading order as `1________`, `2________` and so on.
  • Use one blank per chosen word and do not blank the same word twice.
  • Preserve all other wording, punctuation, and paragraph breaks.

Task 5: final 10-word highlighting

  • Return to the original article.
  • Mark only the 10 Task 3 words with ==word==.
  • Do not carry over Task 1-only masking.

Consistency Checks

Before delivering:

  • Verify Task 3 words are a subset of Task 1.
  • Verify Task 4 contains exactly 10 numbered blanks.
  • Verify each Task 4 blank corresponds to one Task 3 word in order.
  • Verify Task 5 highlights the same 10 words selected in Task 3.
  • Verify no sentence contributes more than one Task 3 word.
  • Verify Task 1 visually matches the Group template style.

Output Pattern

Unless the user gives a stricter format, use this structure:

  • A Group block that matches assets/group_format_template.md for Task 1.
  • A fully rendered article block for Tasks 2, 4, and 5.
  • A numbered list for Task 3.

Resources

assets/

  • assets/day_test_template_v2.md: blank worksheet template mirroring the provided Day_test layout.
  • assets/group_format_template.md: required formatting reference for the Task 1 Group block.

scripts/

  • scripts/build_output.py: assemble a final markdown file from prepared Task 1-5 outputs. Use this when you already completed the reasoning steps and want a clean deterministic export.

Run example:

python3 scripts/build_output.py \
  --template assets/day_test_template_v2.md \
  --article article.md \
  --group task1.txt \
  --task2 task2.txt \
  --task3 task3.txt \
  --task4 task4.txt \
  --task5 task5.txt \
  --output completed.md
Weekly Installs
4
First Seen
12 days ago
Installed on
opencode4
gemini-cli4
amp4
cline4
github-copilot4
openclaw4