Turn the creator's intent into something that can continuously generatechoices, costs, and state changesstory system. Preserve the creator's genre, ending, and scale choices; do not treat popular tropes as the answer to the plot.
Quick Start
Offline verification of precise indexing, slicing, and source file change detection for a complete multi-episode draft:
python3 {技能目录}/scripts/selftest.py
python3 {技能目录}/scripts/episode_intake.py index <多集整稿> --out <episode-intake-index.json>
Before starting
This skill can be installed and executed independently. First read the materials explicitly provided by the user and the direct inputs for this task; if the current directory is
short-drama project and the project tools are available, you may read status and use its release lifecycle, but the absence of core
or any other skill is not a blocking condition for development work.Stage Contract provides
the boundaries of this stage, production form inputs, and the rules table, with no need to read files from other skills.
First determine the entry point
- Only an idea, genre, or emotional goal: complete direction exploration, then establish the creative brief and story engine.
- Existing synopsis, novel, adaptation material, or series notes: preserve the source and first list the committed facts, modifiable scope, source fragments, and items to be confirmed. For long materials, first do semantic segmentation, a function ledger, and candidate episodes; do not use regex name/noun hits to directly publish episodes or assets; do not substitute the original meaning.
- Direction is set, only episode planning is needed: read the existing brief and engine, and directly produce or revise the episode map.
- Existing single-episode script: do not invent development materials just for process completeness. Writing/revising the script is handed to
$short-drama-write; splitting assets or subsequent production can enter directly from the corresponding skills. - There is already a complete multi-episode draft, and an episode map needs to be generated or supplemented: keep the original file, only load Multi-episode full-draft intake and resumable continuation. Let the Agent decide the boundaries and the current batch according to the file's actual structure; the tools only perform precise indexing, single-episode slicing, validation, and continuation, and do not write creative judgments on behalf of the user.
Work directly when the information is sufficient. Only gaps that would change the protagonist, conflict engine, ending promise, or adaptation boundaries need to be asked about.
Each execution
1. Lock the creator contract
Read the creator input and accepted files, and distinguish:
- Facts, themes, characters, and ending promises that cannot be changed;
- Blank spaces that can be explored;
- Form constraints and production boundaries;
- The emotional, informational, or power payoff the target audience can repeatedly receive.
For a new project, copy creative-brief.md. If changing accepted facts, first show the semantic differences and downstream impact; do not directly overwrite.
2. Explore truly different directions
When the direction is undetermined, propose a small number ofmechanically differentcandidates: change the protagonist's strategy, the source of opposition, the cost, or the mode of ongoing conflict, rather than merely swapping the title, profession, and wording. When the direction is clear, deepen it directly; do not manufacture false differences just to fill out options.
For each candidate, state: the dramatic promise, what the protagonist pursues, the opposition mechanism, the recurring payoff, the state that can escalate, the long-term termination condition, and the main production burden. Let the creator choose or combine them, and record the rationale for the trade-offs; options not chosen do not enter established facts.
When the full method is needed, read story-craft.md. When the genre is already determined, you may take from genre-cards.md by indexonegenre card, to calibrate the pressure sources, scene granularity, and production burden more commonly used in that genre; it is a reference that the creator can override with a single sentence, does not replace the direction choice, and does not enter the deliverables.
If the setup has the protagonist carrying a past experience that has already happened, or having acquired a set of externally authorized abilities, that is a premise devicelayered on top of the genre; also read premise-devices.md—a device removes resistance, so it must first be given boundaries and costs. A device does not replace the genre card, nor does it warrant building a separate genre for it.
3. Build the story engine
Copy story-engine.md, and write clearly:
- how the pressure appears, and how the protagonist usually responds;
- how the opponent or system counters, and why the protagonist cannot easily exit;
- what each round of conflict changes, rather than how it repeats more loudly;
- each character's goals, leverage, bottom lines, and relationship pressures;
- which character transformation or revelation of fact would end this engine;
- visuals, sound, scene scale, and content boundaries.
For serials, long-form adaptations, or when vertical character change is heavier, additionally read serial-character-and-memory.md: first build a backstory reserve for creator decisions only, choose the narrative entry window where the old strategy begins to exact a cost, then record the beliefs the characters defend, pressure tests, information permissions, and cross-episode memory. Do not pour the backstory reserve directly into the first episode's dialogue.
4. Ground the series movement into episodes
Read episode-design.md, then copy episode-map.jsonl. For each episode, record the entering state, opening, episode pursuit, resistance, directional turn, episode outcome, information release, outgoing pressure, and the facts that the next episode must inherit.
First ensure each episode produces a local dramatic outcome and that adjacent episodes can hand off precisely, then discuss the dual-track rhythm of external pressure/emotional load. For the Episode 1 entry point, first compare a payoff entry with a cold open; for adaptation projects, first read the original work analysis's highlight table, and place the highlights at the earliest position the audience can get them. The episode count, hook types, whether there are reversals, and climax placement are decided by the creator and the project.
Once the production form is settled, propose a project rhythm profile based on the defaults in the episode design, write it into the creative brief for the creator to accept;
after acceptance, have $short-drama write it into creator_authority.rhythm_profile。
5. Do an owner check and hand off
Before submitting the preview, check:
- Referenced episodes, setups, and payoff records can be found, and open items are clearly marked;
- The candidates genuinely change the mechanism, and the chosen direction runs through the brief, engine, and episode map;
- Each escalation can point to a change in power, information, relationships, exposure, resources, time, or cost;
- Each episode first pays off part of that episode's promise, then leaves a specific decision, danger, or question;
- In adaptation projects, the core characters' stances, core relationships, and main storyline have not been rewritten; all additions not present in the original work are registered as
add, listed in the brief; - did not pass off formatting preferences as universal rules.
These are owner self-checks, not final review. Present a creator-readable summary of additions/changes; the creator must accept it before it can be used as a downstream source;
when a quality conclusion is needed, make $short-drama-review a separate follow-up request, preferably performed by a reviewer who was not involved in creating the current version.
If the project needs a director's statement, this skill only produces 项目开发/director-brief.md candidates and marks the visual_direction / production_profile semantic differences they are intended
to be written into. After the creator accepts, the
$short-drama router promotes the selection to creator authority; development skills do not directly rewrite that authority configuration.
Rule tiers
structural_invariant: locally provable references, IDs, explicit state contradictions; can be blocked by validators.reviewed_invariant: semantic obligations such as whether an escalation is real or a promise is fulfilled; judged by a reviewer citing evidence.craft_default: generally helpful practices; can be overridden if the creator states a reason.taste_option: choices such as hooks, arcs, point of view, ending tone, etc.; must not block on their own.
Do not replace causal judgment with fixed plot points, turning moments, length ratios, or quantity recipes; pacing numbers appear only as the project pacing profile accepted by the creator.
Outputs and boundaries
This skill only owns:
项目开发/creative-brief.md项目开发/story-engine.md项目开发/director-brief.md(when the project needs it; only a creator authority candidate)项目开发/adaptation-map.jsonl(when adapting long material; retain only input locator/span, de-referenced functional summary, candidate destinations, and open items; do not copy the original text; see example in adaptation-map.example.jsonl)项目开发/series-arc.json(when the project needs it)项目开发/episode-intake-index.json(when integrating a full multi-episode draft; a rebuildable mechanical index, not a creative fact)项目开发/episode-map.jsonl
It does not write scene action or dialogue, does not break down assets, does not write image/video prompts, does not generate media, and does not issue final review conclusions. Script semantics are taken over by $short-drama-write .
Load on demand
- promises, engines, character pressure, escalation, and setup payoff:story-craft.md
- episode contracts, causal beats, inter-episode handoffs, first-episode entry, threshold episodes, and project pacing profile:episode-design.md
- there are already multiple episodes of complete scripts/scattered drafts, and the full draft must be kept out of context while allowing interruptible resumption: multi-episode-intake.md
- character drive, backstory and entry point, cross-episode change, information permissions, dual-track pacing, and memory recovery: serial-character-and-memory.md
- there is raw material and it needs to be compressed around payoff points, advance or cold-open later text, merge character scenes, and visualize information: adaptation-craft.md
- the creator provides reference works, sample scripts, or prompts and wants to learn the mechanism rather than imitate the expression: creative-reference-intake.md
- within a unit story the same mechanism is run repeatedly, or the middle section starts repeating and gets thinner as it goes: mechanism-loop.md
- need to distinguish reveal, twist, payoff, and hook, or choose pressure mechanisms by genre: reveal-reversal-payoff.md
- qualitative case methods for choosing conflict progression by genre, designing openings, or planning episode-end hooks: genre-and-hook-playbook.md
- the genre is already determined, and the more commonly used pressures, scene granularity, hook orientation, and production difficulties for that genre are needed: genre-cards.md(index and recall conventions; read only one card at a time)
- the protagonist starts with a known outcome, or gains an externally authorized ability: premise-devices.md(device layer, layered on top of genre cards)
- draft project-level visual direction and production rules (director's statement): director-brief-craft.md
- what this stage owns, inherits, and does not overstep:stage-contract.md