# Cw Dialogue

> Write natural, subtext-rich dialogue that sounds distinct per character.

- Skill: `matrixx0070/cw-dialogue` (Agent Skill)
- Install (CLI): `npx skillmds add matrixx0070/cw-dialogue`
- Raw SKILL.md: https://api.skillmd.com/api/skills/matrixx0070/cw-dialogue/raw
- Safety review: pending (external: skill-scanner PASS, skillspector PASS)
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Coding & Dev Tools
- Author: Matrixx0070 (https://skillmd.com/u/matrixx0070)
- Updated: 2026-08-19
- Page: https://skillmd.com/skills/matrixx0070/cw-dialogue

---


## When to use

Use this when an exchange reads flat, on-the-nose, or interchangeable — when characters are saying exactly what they mean or all sounding alike. Applies to any conversation carrying tension or revealing character.

**Not for:** deep character construction (use cw-character), the surrounding scene's action spine (use cw-scene), story structure (use cw-story-outline), or copyediting (use cw-edit-pass).

## Method

1. Give each speaker a conflicting agenda so the line-level want opposes the other's. Decision point: if both want the same thing, the scene is an interview, not a conversation — split their goals.
2. Write the subtext first: what each is really after. Then have them speak around it. People rarely say the true thing directly.
3. Differentiate voices on four axes: diction (word choice), rhythm (short vs. winding), directness (blunt vs. evasive), and what each avoids. Decision point: cover the names and reattribute lines — if you can't tell who's who, sharpen one axis.
4. Deploy the tools of indirection: deflection, changing the subject, answering a question with a question, silence, and the interrupted line.
5. Trim to the turn. Cut greetings, filler, and any line that only conveys information the reader has — dialogue is combat, not admin.
6. Attribute lightly: "said" is invisible; skip adverb tags and let the words carry the emotion. Use action beats to break rhythm and ground the body.
7. Read it aloud. Decision point: if you stumble or it sounds written, contract it, fragment it, or cut it.

## Example

Agendas: Ana wants an apology; Theo wants to avoid admitting fault. Subtext (unspoken): Ana — "you humiliated me"; Theo — "if I say sorry, I lose." On page: "You had a good time, then." / "It was fine." / "Fine." She restacked the plates that were already stacked. / "What do you want me to say, Ana." / "Nothing. You've said it." Voices: Ana clipped and cornering; Theo hedging, deflecting into logistics. No line states the real grievance.

## Pitfalls

- On-the-nose lines where characters announce their feelings and the plot.
- Every character sharing the author's cadence — no voice separation.
- "As you know" dialogue that exists only to inform the reader (maid-and-butler talk).
- Cluttered tags — "he exclaimed angrily" — doing work the words should do.

## Output format

```
SCENE STAKE (why this talk matters now): <...>
SPEAKER A: agenda <...> | subtext <...> | voice axes <diction/rhythm/directness/avoids>
SPEAKER B: agenda <...> | subtext <...> | voice axes <...>
EXCHANGE (trimmed to the turn):
  A: "<line>"
  <action beat>
  B: "<line, indirect>"
  ...
TURN LANDS ON: <the line/beat that shifts the power>
READ-ALOUD CHECK: <passed / lines to contract>
```