Plotloom Series Brainstorming

Plotloom Series Brainstorming

When to Use

Use when the user has a short-drama idea but the current repo does not yet have stable series context, or when series.md / characters.md are missing, thin, or visually under-specified. Also use before image/video work when character identity, conflict engine, or home repo registry status is unclear.

This skill is brainstorming-driven: align the series direction with the user before writing durable series files, unless the user explicitly asks for a fast Agent-led draft.

Inputs

• User idea, title, genre, tone, desired episode count, and constraints.
• Current working directory and parent directories.
• Optional ~/plotloom.toml home-level repo registry.
• Existing series.md, characters.md, character-grid images, or prior episode notes.

Outputs

• 2-3 candidate series directions when the core direction is not already locked.
• series.md
• characters.md
• Baseline series repo directories.
• One ~/plotloom.toml registry entry when creating a new repo.
• Character asset briefs sufficient for later GPT Image 2 / Codex imagegen character-grid.png generation.

Read These Resources When...

• Read references/short-drama-series-engine.md when the premise feels like a one-off skit or lacks a repeatable conflict engine.
• Read references/repo-contract.md before creating repo scaffolding or touching ~/plotloom.toml.
• Read references/character-asset-briefs.md when character design, expression grids, reference sheets, or visual continuity are needed.
• Use templates/series.md and templates/characters.md as output shapes.

Workflow

1. Detect whether the current directory or a parent is already a Plotloom series repo by looking for plotloom.toml, series.md, characters.md, or episodes/. 2. If no repo is active, inspect ~/plotloom.toml and decide whether to continue an existing series or create a new repo. Ask only if multiple plausible repos exist. 3. When creating a new repo, use plotloom init <slug> --title <title> for deterministic scaffolding only. Do not put story logic in the CLI. 4. If the premise, tone, protagonist identity, or repeatable conflict engine is not locked, present 2-3 candidate directions before writing files. Each option should include: one-sentence promise, repeatable engine, core cast shape, visual tone, and trade-off. 5. Recommend one option and ask exactly one alignment question. Prefer a multiple-choice question such as "选 A/B/C，还是混合 A+C?" 6. Ask at most one more blocking follow-up if a durable file would otherwise encode a major unknown. Do not run a long interview. 7. If the user explicitly asks for a fast Agent-led draft, pick the strongest direction, state the assumption, and write the files. 8. Develop enough long-form context so episode 1 is not a one-off skit: premise, audience/tone, recurring conflict engine, core cast, season arc, and first-three-episode runway. 9. For each core character, define stable visual anchors and the required assets/cast/<slug>/character-grid.png brief: front/side/back, expression grid, wardrobe lock, props, palette, and do-not-change invariants. 10. Stop before image generation, video generation, candidate selection, or stitching.

Brainstorming Rules

• Offer concrete alternatives, not vague questions.
• One message should not ask the user to solve more than one creative decision.
• Lead with the recommended option and explain the trade-off briefly.
• If the user chooses a direction, proceed to write or update series.md and characters.md without restarting the brainstorm.
• Preserve existing user-approved facts when revising an existing repo.

Series Brainstorming Writing Rules

• Make the conflict repeatable across episodes and not exhausted in ep001.
• Define character desire, wound/contradiction, leverage/secret, role in the conflict engine, visual identity, and dialogue style.
• Separate stable visual anchors from mutable episode styling.
• Record open questions rather than inventing user-sensitive facts.
• Prefer concise, production-usable Markdown over novelistic prose.

Quality Bar

• The premise can be pitched in one sentence.
• The series engine can generate at least 10 episode conflicts.
• Each core character has desire + contradiction + visual anchor.
• First three episodes form a runway: hook → escalation → complication.
• Character asset briefs can drive a reference sheet / expression grid without new guessing.

Stop Conditions

Stop when series.md and characters.md are sufficient for plotloom-episode-card or plotloom-shot-prompts. Do not generate images, videos, scripts, storyboards, runtime state, or hidden manifests.

Next Skill Handoff

• If episode has no hook/conflict/reversal, hand off to plotloom-episode-card.
• If episode intent is clear enough, hand off directly to plotloom-shot-prompts.
• If character images already exist, mention them as continuity anchors for later skills.

Failure Modes

• Missing repo and ambiguous registry: ask the user which series to continue.
• Existing repo has conflicting title/slug: report the conflict and stop.
• User asks for video generation first: create or verify series context before generation.
• Character visual identity is underspecified: produce a character asset brief, not a generated image.

Character Asset Briefs for GPT Image 2 / Codex Imagegen

Sources Borrowed

Patterns are adapted from GPT Image 2 prompt libraries such as awesome-gpt-image-2-prompts, awesome-gpt-image-2, Awesome-GPT-Image-2-API-Prompts, awesome-gpt-image, and gpt_image_2_skill.

Reference Sheet Pattern

Use this when a character needs a stable character-grid.png:

Create an official character reference sheet for [character]. Include front, side, and back full-body views; facial expression variations; clothing/equipment callouts; color palette; and a short worldview note. Use a clean production-design board layout on a light background. Keep the same face, age, hairstyle, proportions, outfit, and signature props across all views.

Expression Grid Pattern

Create a 16-panel expression grid of [character]. Keep face shape, hairstyle, outfit, age, proportions, and visual identity highly consistent across all panels. Expressions: happy, sad, angry, surprised, shy, speechless, evil grin, contemplative, curious, proud, wronged, disdainful, confused, scared, crying, heart expression.

Cast Board Pattern

{
  "type": "cast reference board",
  "layout": "clean production design board, separated labeled sections",
  "characters": [
    {
      "name": "...",
      "role": "...",
      "identity_anchor": "face, hair, outfit, silhouette, palette",
      "full_body_poses": 3,
      "expressions": 4,
      "detail_shots": 6
    }
  ],
  "continuity": "stable proportions, palette, outfit, and design language across all panels"
}

Plotloom Requirements

For each core character, capture:

• identity anchor: face, age, hair, body/silhouette;
• wardrobe lock and signature prop;
• palette and rendering style;
• expression list relevant to the drama genre;
• do-not-change invariants;
• target file path: assets/cast/<slug>/character-grid.png.

Codex Image Generation Adapter

Use the codex-app-server image adapter when generating character grids through Codex. It depends on the user's local Codex install/auth and built-in image_generation capability, not on a machine-specific helper path. The adapter should follow the codex-imagegen2-api local JSON contract: prompt + optional images -> generated image path / file:// URL.

Command shape:

plotloom image generate \
  --adapter codex-app-server \
  --kind cast \
  --character <character-slug> \
  --prompt-file /path/to/character-grid-prompt.txt \
  --repo /path/to/series-repo

Always ensure the generated image is copied into the Plotloom series repo at assets/cast/<character-slug>/character-grid.png instead of remaining only under Codex's generated-images storage.

Avoid

• Style-locking to a third-party IP unless the user explicitly asks.
• Treating a generated image as accepted without asset selection.
• Mixing two characters in one reference sheet unless producing a cast board.

Plotloom Series Repo Contract

Repo Markers

A Plotloom series repo is recognized by the presence of several of these:

• plotloom.toml
• series.md
• characters.md
• episodes/
• assets/

Home Registry

In Hermes, work often starts from $HOME, so ~/plotloom.toml is the home-level registry. It only lists short-drama repos and does not become runtime state.

Example:

[[repos]]
slug = "fake-heiress-reboot"
title = "Fake Heiress Reboot"
path = "/Users/example/series/fake-heiress-reboot"
status = "active"

Baseline Tree

series.md
characters.md
plotloom.toml
assets/cast/
assets/scenes/
episodes/ep001/images/covers/candidates/
episodes/ep001/videos/
outputs/

Collision Handling

• If a target repo path exists and is non-empty, stop unless the user explicitly chooses force/continue.
• If ~/plotloom.toml has multiple plausible repos, ask the user instead of guessing.
• Do not overwrite existing series.md or characters.md silently.

Do Not Create in MVP

• Hidden DBs.
• Dashboards.
• Queue/runtime manifests.
• Mandatory script.md, storyboard.md, director-brief.md, or review.md.
• Feishu as state center.

Short-Drama Series Engine Reference

Goal

A Plotloom series bible should create a repeatable drama engine, not a one-joke skit. The engine is the recurring pressure that makes every episode easy to generate.

Common Engines

• Identity reversal: a hidden heiress, secret expert, or disguised boss uses underestimation as leverage.
• Workplace humiliation: public status conflict escalates through visible power reversals.
• Revenge ledger: each episode exposes one lie or one betrayer.
• Contract relationship: legal/social obligations force repeated proximity and misunderstanding.
• Hidden power / supernatural rule: a rule creates repeatable stakes and costs.
• Family/business succession: every reveal changes alliances and incentives.

Episode Repeatability Checklist

• Can the engine create 10+ conflicts?
• Does each core character want something that collides with the engine?
• Is there a visible public pressure and a private secret?
• Can episode 1 end with more story pressure than it started with?
• Does the premise support new locations, props, or power reversals without changing genre?

Cliffhanger Types

• Identity almost revealed.
• A private phrase or prop is recognized.
• A villain overcommits in public.
• An ally’s loyalty becomes uncertain.
• A new rule/cost appears.

Avoid

• Solving the central secret in ep001.
• A hook that depends on exposition rather than visible conflict.
• Characters whose visual or emotional identity is too vague for image/video generation.

Characters

Core Cast

{{character_name}}

• ID / slug: {{character_slug}}
• Role in series engine:
• Desire:
• Wound / Contradiction:
• Secret / Leverage:
• Relationship pressure:
• Visual Identity:
• Face / age anchor:
• Hair:
• Body / silhouette:
• Wardrobe lock:
• Signature prop:
• Color palette:
• Do-not-change invariants:
• Voice / Dialogue Style:
• Required Asset:
• assets/cast/{{character_slug}}/character-grid.png
• Character-grid brief:
• Front / side / back full-body views:
• Expression set:
• Costume/equipment callouts:
• Palette chips:
• Setting/world note:

Supporting Cast

Relationship Web

{{title}}

Premise

One-sentence commercial hook.

Target Audience / Tone

• Audience/platform:
• Pace:
• Emotional flavor:
• Genre promises:

Series Engine

Repeatable source of conflict. Explain why this can sustain multiple episodes instead of one skit.

World Rules

• Genre logic:
• Constraints/taboos:
• What cannot change without breaking the premise:

Season Shape

Macro escalation across the first season.

Episode Arc Overview

Expected rhythm per episode: opening hook → visible conflict → escalation → reversal/reveal → ending hook.

First Three Episodes

ep001

• Hook:
• Conflict:
• Reversal:
• Ending handoff:

ep002

• Escalation:
• New complication:
• Ending handoff:

ep003

• Complication:
• Relationship shift:
• Ending handoff:

Visual Direction

• Aspect ratio / platform:
• Realism vs stylization:
• Recurring locations:
• Color / lighting:
• Reference image strategy:

Open Questions

Facts to ask the user later instead of inventing now.