word-jumble（单词混乱）

Word Jumble

Generates a complete Word Jumble puzzle: 4 scrambled words → circled letters → final idiom answer, with a cartoon hint image and a printable rendered screenshot.

This skill does not post anywhere. Posting is the caller's responsibility.

Output

All output goes into an output/ directory (create if needed, default: ~/.openclaw/workspace/word-jumbles/):

• output/YYYY-MM-DD-puzzle.json — raw puzzle data
• output/images/YYYY-MM-DD-cartoon.jpg — tall cartoon hint image (9:16)
• output/images/YYYY-MM-DD-printable.png — rendered 1200×900 printable screenshot

Puzzle Format

type Puzzle = {
  scrambles: {
    scrambled: string[];      // shuffled letters
    unscrambled: string[];    // correct word, letter by letter
    circled: number[];        // 1-indexed positions in unscrambled
    clue_letters: string[];   // letters at those positions (must match)
    clue: string;             // wordplay clue, no answer giveaway
  }[];
  final_puzzle: {
    scrambled: string[];      // circled letters shuffled (spaces for word breaks)
    clue: string;             // clue for the idiom, no answer giveaway
    solution: string[];       // answer letters + spaces
    imageUrl: string;         // relative path to cartoon image
  };
}

Validation Invariants (must all pass)

1. sorted(scrambled) === sorted(unscrambled) for every word
2. clue_letters[i] === unscrambled[circled[i] - 1] for all i (1-indexed)
3. sorted(all clue_letters) === sorted(solution letters excluding spaces)

Always validate in code before proceeding. Use scripts/validate_puzzle.py:

python3 scripts/validate_puzzle.py output/YYYY-MM-DD-puzzle.json

Step-by-Step Process

1. Generate the puzzle JSON

Prompt the LLM (yourself) with this structure:

Make a Word Jumble puzzle whose final answer is a common idiom. Use 4 words (8–10 letters each). Each word has 2–3 circled letters. Verify every circled position extracts the right letter. Verify circled letters anagram to the final answer. Format as JSON matching the Puzzle type.

Then validate the output in code using validate_puzzle.py. Regenerate if invalid — do not proceed with a broken puzzle.

2. Generate the cartoon image

Use image_generate with a 9:16 aspect ratio. The image must:

• Illustrate the idiom literally and visually (e.g. "bite the dust" → cowboy falling face-first)
• Be black and white, hand-drawn newspaper comic style
• Contain no text, no labels, no words that could give away the answer
• Be a strong visual hint — funny and evocative

Save to output/images/YYYY-MM-DD-cartoon.jpg.

3. Render the printable

Bake the puzzle JSON and cartoon path into assets/puzzle-template.html:

• Replace __PUZZLE_JSON__ with the JSON string
• Replace __CARTOON_IMAGE__ with the cartoon filename (relative, same directory when served)

Serve the HTML locally (localhost only):

cd <tmpdir containing html + cartoon> && python3 -m http.server 7891 --bind 127.0.0.1

Open http://localhost:7891/puzzle.html in the browser tool, resize viewport to 900×900, screenshot (fullPage: false), save to output/images/YYYY-MM-DD-printable.png. (Layout is fluid — the body fills 100% width, so the viewport width controls the final puzzle width.)

Kill the server after screenshotting.

4. Return results

Hand back:

• Path to the printable PNG
• Path to the cartoon image
• The puzzle JSON (or path to it)

Do not post anywhere — the caller decides what to do with the output.

Notes

• Pick fresh idioms each run — avoid repeating recent answers
• Clues should be clever wordplay, never a synonym of the answer
• The cartoon image should be funny and slightly absurdist — it's the soul of the puzzle
• If image_generate is unavailable, skip the cartoon and note it in output

Security

• The local HTTP server binds to 127.0.0.1 only — not accessible from the network
• The answer key is printed in plain text, upside-down, in small print at the bottom of the page. This is intentional — it's a puzzle, not a secret. The answer being technically present in the DOM is fine: it's visually hidden by rotation, and the four scrambled clue words are not shown anywhere, so there's still plenty of solving to do.
• Image generation sends prompts to an external API — callers should be aware of this