agentleFS
bybren-llcbybren-llc/story-systems-template/.claude/skills/writers-room/SKILL.md

writers-room

Skill66 starsChanged 4 months agoMITView on GitHub

This skill convenes a 6-seat collaborative pre-production session — Story Architect, Story Analyst, Dialogue Writer, Scene Writer, Standards Reviewer, and Research Specialist pitch their creative visions before writing begins. Optionally convenes the seats across MULTIPLE models/providers (Claude + self-hosted vLLM + cloud) for genuine diversity.

  • Reads credentials

1 other project uses a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

8 rules. About 0 are common, 6 are distinctive, and 2 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈6

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 2

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Before starting a new screenplay or major rewrite

    line 22
  2. Distinctive:

    When reimagining existing material

    line 23
  3. Distinctive:

    When the creative direction is uncertain

    line 24
  4. Distinctive:

    When you want multiple expert perspectives before committing

    line 25
  5. Very specific:

    solo (default): all six seats = their routed Claude models.

    line 113
  6. Very specific:

    Guest (non-Claude) seats: call consult_model(model_id, seat, system, prompt) (MCP), or node scripts/model-consult.js consult <model_id> …

    line 129
  7. Distinctive:

    Degrade gracefully — never fake: If a guest seat's provider is unreachable or not configured, the connector returns a structured error (not …

    line 135
  8. Distinctive:

    Diverse Perspectives: No single creative blind spot

    line 214

≈ “Common” is an estimate: at least 1 in 5 of the 372,429 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Writer's Room Skill

Purpose

A collaborative pre-production workflow where multiple specialized agents pitch their creative visions before any writing begins. This creates diverse perspectives, healthy creative tension, and a synthesized approach that draws from the best ideas.

When to Use

  • Before starting a new screenplay or major rewrite
  • When reimagining existing material
  • When the creative direction is uncertain
  • When you want multiple expert perspectives before committing

Participants

The Writer's Room convenes 6 specialized agents, each with a distinct role:

Story Architect

Focus: Structure and form

  • Alternative three-act structures
  • Non-linear possibilities
  • Scene sequence options
  • Climax variations
  • Pacing strategies

Story Analyst

Focus: Character and meaning

  • Character arc potential
  • Thematic depth opportunities
  • Aristotle's Six Components evaluation
  • What the story is really about
  • Theophrastus archetype analysis

Dialogue Writer

Focus: Voice and tone

  • Character voice distinctions
  • Comedic/dramatic rhythm
  • Language approach (profanity, dialect, period)
  • Signature lines and moments
  • Subtext opportunities

Scene Writer

Focus: Visual storytelling

  • Key visual setpieces
  • Physical comedy/action beats
  • Power dynamics per scene
  • Compression opportunities
  • "Show don't tell" moments

Standards Reviewer

Focus: Quality and originality

  • What makes each proposed element A-grade?
  • Weak points to avoid
  • Cliche identification
  • Fresh vs. derivative analysis
  • Genre boundary awareness

Research Specialist

Focus: Authenticity and grounding

  • Setting accuracy requirements
  • Technical/professional accuracy
  • Historical or cultural considerations
  • Real-world logic checks
  • Reference material needs

Process

Step 1: Brief the Room

Provide all participants with:

  • The core premise/logline
  • Any existing material (V1 script, source material, notes)
  • Target tone and audience
  • Constraints (runtime, budget considerations, rating)
  • What's "sacred" vs. open for reimagining

Step 2: Individual Pitches

Each agent prepares their perspective:

  • What excites them about the material
  • Their proposed approach
  • Specific recommendations
  • Potential concerns or warnings

Step 3: Synthesis

Combine the best elements from all pitches into:

  • A unified creative direction
  • Key scenes that must exist
  • Tone and voice guidelines
  • What's explicitly off-limits
  • The "North Star" for the project

Multi-Model Council (opt-in)

By default every seat is played by its routed Claude model (see .claude/model-routing.md) — "diverse perspectives" from diverse prompts. The council upgrade lets the room convene diverse minds: some seats on Claude, some on a self-hosted vLLM model, some on another cloud. This is additive — the default behavior is unchanged.

Modes

  • solo (default): all six seats = their routed Claude models. No registry or endpoints needed. This is exactly the classic Writer's Room.
  • diverse: seats spread across providers, plus optional guest seats on rostered non-Claude models (e.g. a local Qwen/Llama wildcard, a Gemini seat).
  • custom: an explicit seat→model roster provided at invocation.

How the council runs

The Session Manager conducts:

  1. Read the roster. node scripts/model-consult.js list (or the list_models MCP tool if the model-consult server is enabled) returns the available models, redacted, with a ready flag per model. Assign each seat a model_id — default is the seat's affinity → Claude model.
  2. Gather pitches in parallel — one mind per seat:
    • Claude seats: dispatch the seat's native subagent ("Acting as the Story Architect…") on its routed model, as today.
    • Guest (non-Claude) seats: call consult_model(model_id, seat, system, prompt) (MCP), or node scripts/model-consult.js consult <model_id> --seat <seat> --prompt "…". The system prompt is the seat's role framing; the prompt is the brief from Step 1.
  3. Synthesize with provenance. A single Claude Opus pass consolidates all pitches into one direction and credits which seat/model each surviving idea came from — so you can see whether the local model or the cloud guest actually moved the needle.
  4. Degrade gracefully — never fake. If a guest seat's provider is unreachable or not configured, the connector returns a structured error (not a completion). Skip that seat and record it under "Seats skipped" with the reason. The room continues with the seats it has; the roster header states who spoke and who was skipped (no silent drops).

Configuring guests

Guest endpoints and keys live in .env (see .env.example); the registry (.wtfb/ai-harness/model-registry.json) references only the env-var names. To enable the MCP path, install the SDK and register the server per scripts/model-consult-mcp.mjs. With nothing configured, the room simply runs in solo mode.

Output Format

# Writer's Room Creative Direction

## Project: [Title]
## Date: [Session Date]
## Mode: [solo | diverse | custom]

### Council Roster
| Seat | Model | Provider | Spoke? |
|------|-------|----------|--------|
| Story Architect | claude-opus-4-8 | anthropic | ✅ |
| Story Analyst | claude-opus-4-8 | anthropic | ✅ |
| Dialogue Writer | claude-fable-5 | anthropic | ✅ |
| Scene Writer | claude-fable-5 | anthropic | ✅ |
| Standards Reviewer | claude-opus-4-8 | anthropic | ✅ |
| Research Specialist | claude-opus-4-8 | anthropic | ✅ |
| *(guest)* Wildcard | vllm-local | openai-compatible | ✅ / ⏭️ skipped |

*Seats skipped:* [none] — or "Wildcard (vllm-local): NETWORK_ERROR — endpoint unreachable".

---

## The Pitch (From Story Architect · claude-opus-4-8)
[Structure and form recommendations]

## The Heart (From Story Analyst · claude-opus-4-8)
[Character and thematic core]

## The Voice (From Dialogue Writer · claude-fable-5)
[Tone and language approach]

## The Spectacle (From Scene Writer · claude-fable-5)
[Visual setpieces and key moments]

## The Standard (From Standards Reviewer · claude-opus-4-8)
[Quality benchmarks and originality notes]

## The Foundation (From Research Specialist · claude-opus-4-8)
[Authenticity requirements]

---

## Synthesized Creative Direction

### Core Approach
[Unified vision statement]

### Must-Have Scenes
1. [Scene description]
2. [Scene description]
3. [Scene description]

### Tone Guidelines
- [Guideline]
- [Guideline]

### Off-Limits
- [What to avoid]

### North Star
[Single guiding principle for all creative decisions]

Benefits

  1. Diverse Perspectives: No single creative blind spot
  2. Quality Bar: Standards Reviewer prevents lazy choices
  3. Grounded Creativity: Research Specialist prevents plausibility errors
  4. Unified Vision: Synthesis creates alignment before writing begins
  5. Reusable Asset: Creative Direction document guides entire production

Integration

The Writer's Room output feeds directly into:

  • /theme-discovery - Validates thematic direction
  • /character-interview - Guides character development priorities
  • /story-check - Establishes success criteria
  • /scene-writer agent - Provides visual direction
  • /dialogue-writer agent - Sets voice parameters

Example Invocation

/writers-room

Project: Racoon Rescue V2
Premise: A drunk Navy petty officer uses an unconscious raccoon to bypass his breathalyzer interlock.
Existing: V1 screenplay (15 min short, dark comedy)
Target: Smart R (Fargo-level), complete reimagining
Sacred: Core premise only - everything else open
Constraints: Keep at ~15 min, sophisticated adult humor

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.