| name | tech-dialogue |
| description | Improve dialogue to feel natural and character-specific (subtext, voice, tension, beats). Use when drafting/revising dialogue-heavy scenes or when dialogue sounds generic/AI-like. |
tech-dialogue
Use when (proactive triggers)
Call this skill without being asked when you observe any of the following:
- Dialogue feels generic/AI-like (same cadence for every speaker, over-explains, no subtext).
- The scene is dialogue-heavy but has low tension or no directional pressure.
- A character’s voice is drifting (OOC risk).
Quick start (runbook)
- Diagnose (2 minutes): name the failure mode(s) + cite 1–2 concrete lines as evidence.
- Lock voice (if needed): fill a voice sheet from
assets/templates/voice-sheet.md (or update existing character voice in spec/characters/*).
- Build pressure: write a dialogue intent map using
assets/templates/dialogue-intents.md.
- Apply “surgical” edits: change only the dialogue lines and nearby beats that carry the problem.
- Self-check with
references/CHECKLIST.md, then optionally suggest a gate for verification.
Inputs (what to read)
- The target dialogue passage (chapter/scene).
- Relevant canon constraints when available:
- character boundaries/voice:
spec/characters/*
- style contract:
spec/style/*
- narrative/POV rules (no cheating exposition):
spec/narrative/*
Purpose
This skill helps produce dialogue that is:
- character-specific (voice differences are audible)
- tense (conflict/pressure exists even in “polite” scenes)
- economical (no info-dump “as you know” speech)
Diagnose (common failure modes)
- Same-voice drift: everyone has similar sentence length, vocabulary, politeness level.
- Info dump: characters say things only the reader needs, not what they would say.
- Zero pressure: lines don’t negotiate anything; no leverage, no refusal, no consequence.
- No embodiment: endless back-and-forth with no beats, no space, no physicality.
- Too explicit: motives and feelings are stated instead of implied via subtext.
Surgical edit menu (prefer these over rewriting)
Pick a few actions and apply locally:
- Replace explanation lines with subtext questions or deflections.
- Add one refusal (“I won’t talk about that”) to create a boundary.
- Add one leverage beat (object, timer, third party listening, physical constraint).
- Introduce a micro-turn (new info, misunderstanding, concession, threat, promise).
- Move necessary exposition out of dialogue into action/setting/afterthought.
- Vary turns: interruption, silence, overlap, unfinished sentence, changed subject.
Output expectations
- The revised dialogue passage (minimal, targeted edits).
- Optional artifacts (only if useful):
- A voice sheet:
assets/templates/voice-sheet.md
- A dialogue intent map:
assets/templates/dialogue-intents.md
Output
- Revised dialogue text in the relevant draft/chapter.
- Optional: a short “dialogue intent map” using
assets/templates/dialogue-intents.md.
Applying changes (editing discipline)
- Prefer localized edits (change the dialogue lines and nearby beats) instead of rewriting entire chapters.
- When modifying project files, use
project__apply_edits so the user can review small, controlled diffs.
References
- Revision checklist:
references/CHECKLIST.md
- Common pitfalls + fixes:
references/PITFALLS.md
- Deep techniques (ported reference):
references/TECHNIQUES.md
Self-check
- Can you identify who is speaking without tags?
- Is there subtext (stakes) instead of pure information transfer?
- Do lines create pressure or change something (not just filler)?
Optional: verification gate
If this change is important (voice continuity risk, canon constraints, or large edits), suggest running a verifier gate:
gates/gate-chapter (for prose quality + continuity)
gates/gate-consistency (for cross-file voice/canon consistency)