style-examples

name: style-examples description: Create style example files for session_doc.py narration from the user's campaign summaries. Use when the user asks to create examples or style references for a campaign. tools: Read, Glob, Grep, Bash, Agent, Write

Style Examples Generator

Create examples/*.md files for session_doc.py narration by extracting representative passages from the user's own writing.

What These Files Are For

session_doc.py Pass 5 (narration) loads all *.md files from the --examples directory as style references. The model reads them before writing narration to match the user's tone, sentence structure, POV handling, humor, and pacing. These must be verbatim excerpts from the user's writing — not rewritten, not summarized, not improved.

Required Information

Before starting, you need:

1. Campaign directory (detect from CWD or ui_config.yaml, or ask)
2. Session summaries file — the source of the user's writing
3. How many chapters/sessions to read (default: first 10)

Ask the user for anything you can't detect automatically.

Workflow

Phase 1: Read the source material

Read the summaries file thoroughly (at least the first 10 chapters/sessions). Use Explore agents in parallel for large files. As you read, identify passages that showcase distinct writing qualities:

Look for these style dimensions:

• Character introspection: Internal monologue, italicized thoughts, how characters observe and assess
• Dialogue and political maneuvering: Conversations where characters lie, negotiate, manipulate — with internal commentary running underneath
• Combat and consequences: How violence is described (terse? elaborate?), aftermath and emotional cost
• Ensemble comedy: Multi-character scenes with humor, rapid NPC introductions, escalating absurdity
• Suspense and action: Chases, stakeouts, infiltration — pacing and tension techniques
• POV switching: How the writer moves between character perspectives within a chapter
• NPC voice: How the writer gives distinct personalities to NPCs through dialogue and description

Also identify the writer's signature patterns:

• Sentence rhythm (staccato fragments? flowing prose? both depending on mood?)
• How they handle the gap between internal monologue and spoken dialogue
• Humor style (dry? dark? absurdist? observational?)
• Use of sound effects, italics, formatting
• First person vs. close third person and when they switch
• How they describe physical spaces and people

Phase 2: Select and extract passages

Choose 4-6 passages that together cover the full range of the writer's style. Each passage should be:

• Self-contained: Makes sense without context (or with minimal context)
• Verbatim: Copied exactly from the source, not rewritten or edited
• Substantial: Long enough to demonstrate a pattern (typically 500-2000 words each)
• Distinct: Each example showcases a different facet of the writing

Good example categories (pick 4-6 that fit this writer's strengths):

1. Introspection and observation — a character alone with their thoughts
2. Political maneuvering — dialogue with subtext and internal commentary
3. Ensemble comedy — multiple characters bouncing off each other
4. Combat and consequences — violence and its emotional aftermath
5. Suspense/action — a chase, infiltration, or stakeout
6. World-building — how the writer describes places and introduces new settings
7. Emotional weight — a death, a betrayal, a moment of genuine feeling

Phase 3: Write the example files

Write each example to <campaign>/examples/<descriptive_name>.md.

Each file should:

• Start with # Style Example: <category> as a title
• Use --- separators between non-contiguous passages
• Preserve the original character headings (### Zephyr, #### Sequoia, etc.)
• Preserve all formatting: italics, bold, blockquotes, line breaks
• Include NO commentary, NO annotations, NO "this shows the writer's tendency to..."
• Be pure source material — the model reads these as writing to emulate, not writing to analyze

Key Principles

• Never rewrite: These are the user's words. Copy them exactly. Typos, unconventional formatting, and all. The model needs to learn the ACTUAL voice, not a cleaned-up version.
• Variety over quality: A single brilliant passage teaches one thing. Five good passages covering different modes teach the model to match the writer across all situations.
• Show the range: If the writer does staccato combat AND flowing political dialogue AND dark humor, the examples need all three. Don't just pick the "best" writing — pick the most representative.
• Include the weird stuff: If the writer uses sound effects (Crack, Twang), animated vegetables, six repetitions of "Shit" — include those. The unusual choices ARE the voice.
• Multi-POV is a feature: If the writer switches between character perspectives within scenes, the examples should show this. It's a deliberate technique the model needs to learn.

Output

List the files created with a brief note on what each one teaches the model. Remind the user they can add more examples over time — the model reads all *.md files in the directory.