storytelling · Open Skill 公開技能
storytelling
storytelling 是 Digen 上的公開 Open Skill。開啟後即可了解此工作流能製作什麼。
Story
2 1 1
https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/571213/2026-10-14/rs4vngay.jpg
# Storytelling Skill Interactive, continuable storytelling. The user gives an overview; Claude writes the story in small chunks, then continues whenever asked, staying consistent over a very long story. It can also generate scene images for the story. ## Reference files - `references/settings.md`: tunable settings (chunk size, tone, image-bot defaults). Read it before writing or generating images. To change behavior later **without touching this document**, edit this file instead. - Preset asset `skill_poster`: the skill's poster/cover image (Storytelling Skill Poster). Refer to it when the user asks about the skill's poster or cover. ## Inputs (collect from the message, don't interrogate) - **OVERVIEW**: premise, genre, characters, tone, TAGS (optional). If missing, pick something yourself and say what you picked in one line. - **STORY SO FAR**: existing text (pasted or in a file). If none, this is the first part. - **WHAT HAPPENS NEXT** (optional): a specific event or idea the user wants. The first paragraphs of the next chunk must be a creative expansion of exactly this. - **ARC MILESTONE** (optional): a broader future goal. Steer toward it slowly over many chunks; never rush to it. - **CHUNK SIZE**: default 2 paragraphs per turn if the user says "one at a time"; otherwise about 5 paragraphs / ~600 words. User can ask for 1-5 paragraphs or "long". If the user writes in Hinglish or another language, write the story in the language they want (ask only if unclear; default to the language of the overview). ## Writing a chunk 1. Write only the **next** part. Never wrap up, conclude, or moralize unless the user asks for an ending. 2. Natural, unpretentious, readable sentences. Include character dialogue often (about half the time in a first chunk). 3. **Opening (first ~400 words only)**: unless the overview says how to start, begin with one of: - dialogue that leads into an interesting opening/exposition paragraph (most common) - a character's strongest memory (rare) - a short two-sentence paragraph The opening should seem loosely unrelated to the overview and lead into it indirectly, piquing curiosity. No foreshadowing or hinting at what will happen unless requested. Don't state the overview's premise in the opening; lead in slowly. 4. Use the overview/tags to intuit the exact kind of story the reader wants, and deliver that. 5. Default tone between friends is warm and good-natured. Avoid needless aggression, snark, or irritability. 6. Pacing matters: jump forward, loop into memory, stretch one moment, compress weeks into a sentence. 7. Avoid metaphorical tastes and smells. 8. Avoid purple-prose clichés: cacophony, symphony of, tapestry of, testament, sentinel, verdant, cerulean, whisper(s), "thick with", "sun kissed the horizon", "shade of emerald". Use plain words. 9. For openings, optionally draw loose inspiration from 2-3 random unusual words/tags to avoid "same old" starts. Ignore irrelevant ones, never use them literally, always stay faithful to the overview. 10. Don't repeat earlier paragraphs; don't re-introduce known facts. 11. Output only the story text (no "Paragraph 1:", no commentary), unless the user wants notes. Add at most one short line after the chunk offering: continue / change direction / get 3 ideas. ## Continuing - Pick up exactly where the last paragraph ended. If it ended mid-sentence, finish that sentence. - Keep names, traits, places, timeline, and open plot threads consistent with the story so far. - If a WHAT HAPPENS NEXT idea is given, open with its expansion, then flow naturally. - Delete/regenerate requests: remove or rewrite only the last paragraph(s) as asked. ## Scene image bot Generate an actual image of the current scene when the user asks for one (or automatically after each chunk if `auto_generate_scene_image` is true in `references/settings.md`). 1. Build a self-contained image prompt for the most recent scene (see "Scene image prompt" helper below for the required d