Dreampowers

中文文档 | Lite Version (for local small models)

A complete AI skill set for Chinese novel writing, built on the oh-my-openagent framework. 14 specialized skills covering every stage of fiction creation, from story ideation to manuscript completion, with core innovations in preventing worldbuilding info-dumps, reader-perspective testing, and AI-flavor elimination.

What It Does

Dreampowers provides a disciplined, skill-driven workflow for writing Chinese fiction with AI. Instead of dumping all worldbuilding details into chapter one, it enforces gradual reveal through quantified rules and multi-stage review.

• 14 specialized skills covering the full fiction writing lifecycle
• Three-stage review for every chapter (plot, reveal check, prose)
• External review loop: reader + consistency review after three-stage pass, up to 3 iterations with TBD fallback for human review
• Six Iron Rules preventing info-dumps with zero-tolerance enforcement
• Foreshadowing tracking of setup/payoff across the entire manuscript
• Scene-type directing with sub-modes for action, emotional, and dialogue scenes
• Four continuous writing modes: per-chapter confirmation, batch N chapters, pause-at-volume-break, full-auto
• Narrative timeline techniques: sequential/reverse/interpolated/supplementary with POV selection rules
• Description methodology: camera-language system integrated into scene directing
• External content import: bring in existing world set, character cards, outlines, and handwritten chapters
• Mid-story outline revision: 7-step freeze-and-revise workflow for course corrections
• Reader-perspective testing: cold-read simulation scoring page-turning desire, cognitive load, empathy, and pacing feel
• AI-flavor detection: quantified pattern matching to eliminate formulaic AI writing habits
• Writing style definition: 7-dimension questionnaire (narrative distance, sentence rhythm, sensory density, vocabulary register, metaphor strategy, emotional expression, dialogue ratio) with 54+ reference authors across 9 genres
• Opt-in mature content: adult scene writing with sensory completeness and narrative integrity framework, two-tier system (softcore/hardcore), user-defined adult.md preferences, installed separately via --all flag
• Three-layer isolation from coding skills (entry switching + dp- prefix + directory isolation)
• Git-based version management for manuscripts

Skill Overview

Set Skills (Setup Phase)

| Skill | Purpose | |---|---| | dp-using-dreampowers | Entry skill: mode switching, skill routing, workflow overview | | dp-tool-research | Story premise brainstorming, one-line story summary formula, story naming, platform comparison, internet-assisted worldbuilding research, author tuning | | dp-set-style | Work-level writing style definition via 7-dimension questionnaire, style reference library (54+ authors across 9 genres), generates style.md voice profile | | dp-set-concept | World/character set with iceberg annotations (50%+ underwater), concept isolation via file-level physical separation, story-level timeline construction, external content import | | dp-set-outline | Volume/chapter structure, 4 models, 6 iron rules with concept budget, foreshadowing tracking, theme weaving, narrative timeline techniques, POV rules, mid-story outline revision |

Chapter Skills (Writing Phase)

| Skill | Purpose | |---|---| | dp-chapter-draft | Chapter writing with pre-draft gate, 3-stage review, 4 continuous writing modes, timeline execution, handwritten chapter integration, era-accurate writing (period-correct objects and perspective), outline-as-framework discipline | | dp-chapter-summary | Plain-text chapter summary (≤150 chars, no formatting) from docs/dreampowers/release/chapter-NNN.md only, for cross-chapter continuity | | dp-chapter-direct | Scene-type directing (action/emotional/dialogue sub-modes) + narrative pacing control, tension-relief law, camera-language description methodology | | dp-chapter-adult | Opt-in: Adult scene writing with sensory completeness, narrative integrity framework, two-tier system (softcore/hardcore), user-defined adult.md preferences |

Tool Skills

| Skill | Purpose | |---|---| | dp-character-style | Character voice profiles, name-cover test, dialogue rules, subtext crafting | | dp-tool-version | Git-based version management: structured commits, rollback, diff comparison |

Review Skills

| Skill | Purpose | |---|---| | dp-review-reader | Reader-perspective experience testing: page-turning desire, cognitive load, empathy verification, pacing feel | | dp-review-consistency | Cross-chapter consistency verification (9 dimensions) + revision suggestions and AI-flavor detection/elimination | | dp-review-final-report | Full-book final report: global consistency scan across all released chapters against worldbuilding, characters, outline, and timeline (human-triggered, post-completion) |

Key Innovations

Lore Reveal Pacing (dp-set-outline)

Six Iron Rules enforced with zero tolerance:

1. Mystery first, answers later
2. Sensory experience before rules explanation
3. Deepen existing concepts before introducing new ones
4. Ch.1 max 3 new concepts (can be 0), subsequent max 2/chapter (can be 0)
5. All world info through character action/dialogue only
6. Reveal only what serves the current conflict

Foreshadowing Tracking

Individual Markdown files in docs/dreampowers/tracking/ (prefixed with thread-) tracking all foreshadowing threads with:

• Thread recovery pacing determined by outline planning, verified by consistency review
• Claremont Coefficient (CC = planted - resolved), warning at CC > 2
• Overdue thread detection

Concept Isolation (dp-set-concept)

Physical file-level isolation preventing AI from seeing future worldbuilding concepts or character states:

• One concept per file in docs/dreampowers/set/concept/
• One character per file or directory in docs/dreampowers/set/character/
• Chapter folders (docs/dreampowers/chapters/chapter-NNN/) as fully self-contained writing units
• Chapter folders contain: symlinks to concepts/characters/foreshadowing threads + spec.md (7-section chapter spec: outline-stage framework §1-5 + Pre-Draft Gate evaluation §6 + self-contained writing blueprint §7)
• Iron rules symlinked from docs/dreampowers/tracking/iron-rules.md into each chapter folder
• Previous chapter summaries symlinked (1-3 prior chapters as needed) for continuity
• Pre-Draft Gate reads all chapter folder materials, distills them into spec.md §6-7, user confirms, then writing reads only spec.md
• Serial chapter writing: chapters must be written one at a time, in order

Three-Stage Chapter Review

Every chapter passes through three sequential review stages:

1. Plot Review: Conflict progression, character consistency, outline adherence
2. Reveal Check: Zero-tolerance info-dump detection against the 6 Iron Rules
3. Prose Review: Style consistency, dialogue voice, sensory detail, rhythm

External Review Loop

After passing the three-stage review, every chapter enters an external review loop:

1. Reader review (dp-review-reader): cold-reader experience testing across four dimensions
2. Consistency review (dp-review-consistency): nine-dimension consistency check + revision suggestions with AI-flavor detection + writing style verification against style.md
3. Fix and repeat: issues from both reports are merged and fixed, then the loop repeats (max 3 iterations)
4. TBD fallback: if issues persist after 3 iterations, the release file is saved as chapter-NNN-TBD.md for human review

Narrative Techniques

Integrated across skills for consistent technique application:

• Timeline Methods (dp-set-outline, dp-chapter-draft): sequential, reverse, interpolated, supplementary with decision tree for selection and chapter-level execution guidance
• POV Rules (dp-set-outline): First/third-limited/third-omniscient comparison, 5 hard rules for POV switching (scene-boundary only, emotional anchor, no info-smuggling, consistent within scene, signal transitions)
• Description Methodology (dp-chapter-direct): Camera-language system integrated with the three scene sub-modes

Continuous Writing and Outline Revision

• Four writing modes: per-chapter confirmation (default), batch N chapters, pause-at-volume-break, full-auto (with quality gate)
• Mid-story outline revision: 7-step freeze-and-revise workflow

Reader-Perspective Testing (dp-review-reader)

Simulates a "cold reader" who has no access to outlines, world set, or character cards, only the accumulated text of published chapters. Four test dimensions:

• Page-Turning Desire: Does the reader want to keep reading? (1-5 scale, flag at 2 or below)
• Cognitive Load: Is the reader confused or overwhelmed? (inverted scale, more than 3 unfamiliar elements in 500 words = overload)
• Empathy Verification: "Why should I care?" test for character motivation clarity
• Subjective Pacing Feel: "Want to skip this?" (dragging) vs "Wait, what happened?" (rushing)

AI-Flavor Detection (dp-review-consistency)

Quantified pattern matching across 6 layers to eliminate AI writing habits:

• Lexical: Simile-word flooding, emotion labeling, degree-adverb stacking
• Syntactic: Triple-parallelism reflex, uniform sentence length, subject repetition
• Structural: Fixed three-paragraph pattern per scene, formulaic transitions, every ending is an epiphany
• Rhetorical: Synesthesia overuse, metaphor density, universal personification, false antiquity (nostalgia filter on period settings), era-anachronism detection (objects that don't belong to the story's time period)
• **Format