# Voice Discovery

> Discover the voice for a specific novel during the foundation phase. Writes 3-5 trial passages in different registers, evaluates which best serves the world and tone, then fills in voice.md Part 2 with the chosen register, exemplars, and anti-exemplars.

- **Type:** Skill
- **Install:** `agentstack add skill-tchr-dev-autonovel-voice-discovery`
- **Verified:** Yes — security-reviewed for prompt injection and unsafe behavior
- **Seller:** [tchr-dev](https://agentstack.voostack.com/s/tchr-dev)
- **Installs:** 0
- **Category:** [AI & ML](https://agentstack.voostack.com/c/ai-and-ml)
- **Latest version:** 0.1.0
- **License:** MIT
- **Upstream author:** [tchr-dev](https://github.com/tchr-dev)
- **Source:** https://github.com/tchr-dev/autonovel/tree/main/.claude/skills/foundation/voice-discovery

## Install

```sh
agentstack add skill-tchr-dev-autonovel-voice-discovery
```

Requires the [AgentStack CLI](https://agentstack.voostack.com/docs/cli). Works with Claude Code, Cursor, and any MCP-compatible agent.

## About

# Voice discovery

The voice should feel like it BELONGS in the world. Le Guin's insight: in fantasy, the language creates the world, not just describes it. A novel about salt-flat priests does not sound like a novel about a tide-bound port city.

## Inputs (from `/`)

- `seed.txt`, `world.md`, `characters.md`
- `voice.md` (template — Part 1 guardrails are already filled; you fill Part 2)

Reference: `framework/voice-guardrails.md` (Tier 1/2/3 banned words and structural slop patterns — these apply to ALL voices, regardless of which register you choose).

## Procedure

### Step 1 — propose 3-5 candidate registers

Based on the world and characters, list candidate voices the novel could have. Examples (do not just pick from these — invent for the specific novel):

- **Mythic and weighty** — like stone tablets being read aloud. Long sentences. Latinate vocabulary. Clauses that echo psalms.
- **Spare and cold** — sentences like knife cuts. Anglo-Saxon vocabulary. Short paragraphs. Heavy white space.
- **Warm and breathless** — like a traveller telling stories by firelight. Run-on sentences. Colloquial register. Sensory rushes.
- **Dry and ironic** — flat affect. Dark humour. Subtext does the heavy lifting.
- **Childlike-strange** — present tense, unreliable register, fairy-tale syntax.
- **Reportorial / forensic** — emotionally restrained, observational, precise to the level of measurement.

Pick three to five candidates that fit THIS world.

### Step 2 — write a trial passage in each register

For each candidate, write a ~150-word scene from the novel. Same scene every time so registers are directly comparable. Pick a scene that touches the world's specifics — magic in use, a key location, a character speaking. Save the trial passages to `/voice_trials.md`.

### Step 3 — evaluate

For each trial passage, list:
- What it achieves
- What it costs
- What kind of novel it implies
- Which audience it fits
- Whether it inherits any AI-tells from `framework/voice-guardrails.md`

Pick the register that best serves the story. Justify in 2-3 sentences.

### Step 4 — fill `voice.md` Part 2

Edit `/voice.md`. Below the `## Part 2: Voice Identity` heading, fill in:

```
### Tone

### Sentence Rhythm

### Vocabulary Register

### POV and Tense

### Dialogue Conventions

### Exemplar Passages

### Anti-Exemplars

```

## Hard rules

- The exemplars must be SPECIFIC to the novel — they reference characters, locations, magic that exist in `world.md` / `characters.md`.
- The anti-exemplars must be SPECIFIC too. "Too flowery" alone is not enough — show a flowery passage that's wrong for THIS novel.
- The voice must avoid the universal Tier 1/2/3 slop in `framework/voice-guardrails.md`. Even the most baroque voice should not say "delve" or "tapestry."

## After writing

Tell the user which register won and why. Suggest running `evaluate-foundation` to score the planning docs as a whole.

## Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

- **Author:** [tchr-dev](https://github.com/tchr-dev)
- **Source:** [tchr-dev/autonovel](https://github.com/tchr-dev/autonovel)
- **License:** MIT

Install and usage instructions live in the source repository linked above.

## Pricing

- **Free** — Free

## Security capabilities

Automated source analysis of v0.1.0 — what this tool can access:

- **Network access:** no
- **Filesystem access:** no
- **Shell / process execution:** no
- **Environment & secrets:** no
- **Dynamic code execution:** no

*"Yes" means the capability is present in the source — more access means more to trust, not that it is unsafe.*


## Versions

- **0.1.0** — security scan: passed — Imported from the upstream source.

## Links

- Listing page: https://agentstack.voostack.com/l/skill-tchr-dev-autonovel-voice-discovery
- Seller: https://agentstack.voostack.com/s/tchr-dev
- Browse the marketplace: https://agentstack.voostack.com/browse

---
Listed on AgentStack — the marketplace for AI agent skills and MCP servers. Every listing is security-reviewed. Creators keep 70%.