# Write and Edit in Your Own Voice

Generic rewrites can lose your writing voice or change important details. Reword uses chosen samples, saved preferences, and local checks to guide personalized drafting and editing.

## Install

```bash
npx skillstore add starrybodies/reword
```

## Metadata

- Status: approved
- Slug: starrybodies-reword
- Skillstore revision: r1
- Version status: missing
- Tree hash: 2b80dfe8cdb618b4d8ff28362fbd505bff8c7252589236bd5b5d8156f6e6b6a7
- Author: starrybodies
- GitHub username: starrybodies
- License: MIT
- Repository: https://github.com/starrybodies/reword/tree/8ffe91a664be6053dcdfbb53e18c330496c058f1/
- Ref: ee739ad38b1e70d7cf051387f911aed2de09ae2b
- Supported tools: Claude, Codex, Claude Code
- Audit status: complete
- Agent install advisory: allowed
- Manual install advisory: allowed
- Artifact signature: available
- Audit attestation: unavailable
- Human verification: not\_verified
- Risk factors: external\_commands, network, env\_access, filesystem, scripts
- Quality score: 88
- Quality tier: featured
- Public page: https://skillstore.pages.dev/skills/starrybodies-reword
- Manifest: https://skillstore.pages.dev/api/skills/starrybodies-reword/manifest

## Capabilities

- Build private voice collections from deliberately supplied samples, with separate authored and reference labels.
- Assemble bounded editing and voice-learning requests for the host model without bundled network calls.
- Apply destination guidance for email, chat, text messages, LinkedIn, cover letters, essays, and web copy.
- Compare drafts for changes to numbers, dates, quotations, code, links, locked phrases, and selected claim relationships.
- Save explicit preferences and author revisions, with consent required for automatic feedback capture.
- Inspect collection coverage, export compact voice cards, pause learning, forget evidence, and restore retained profile versions.

## Use Cases

- Keep Professional Messages Personal: Rewrite emails and team messages using saved preferences while checking dates, requests, and commitments against the original.
- Draft Applications Without Invented Claims: Use a personal voice guide and supplied career facts to draft cover letters with suitable destination conventions.
- Maintain an Editorial Voice: Build a compact guide from selected authored pieces, then refine essay or web-copy guidance using deliberate revisions.

## Prompt Templates

### Rewrite One Email

```
Rewrite this email for clarity while keeping my tone. Preserve all dates, names, and commitments. Return finished copy: [draft].
```

### Build a Private Voice Guide

```
Build my voice from these authored samples: [samples]. Save a named collection in my chosen private location. Ask before enabling ongoing learning.
```

### Draft for a Specific Destination

```
Use my selected voice collection to draft a cover letter from [factual brief] for [reader]. Use only supplied facts. Apply my explicit preferences: [preferences].
```

### Refine and Verify a Voice Guide

```
Save this author revision once: [original and revision]. Keep ongoing learning disabled. Refresh email guidance within the evidence budget. Show pending evidence and export the compact card.
```

## Limitations

- Writing and semantic voice distillation require a host model; Python helpers do not generate prose or train model weights.
- Preservation checks are heuristics, not proof of truth or complete semantic equivalence; many language-sensitive checks target English.
- Prose commands accept plain text and Markdown; rich documents, binary formats, and source code require other editing tools.
- Voice fidelity is not guaranteed, and character budgets do not guarantee exact token counts.

## Best Practices

- Choose relevant authored samples, label reference inspiration separately, and keep collections outside public repositories.
- State the destination, reader, factual brief, and explicit preferences before drafting or editing.
- Preserve originals and review the comparison using the same voice and destination settings before replacement.

## Anti Patterns

- Treating sample experiences or reference writing as factual support for a new draft.
- Enabling ongoing learning without consent or treating accepted model prose as authored evidence.
- Assuming a clear mechanical check proves factual accuracy, complete meaning preservation, or an exact voice match.

## Security Audit

- Audited at: 2026-11-16T20:24:13.577\+00:00
- Summary: All 267 static alerts are false positives in the reviewed source context, including defensive injection fixtures, Markdown examples, and explicit local storage operations. No evidence found of malicious instructions, credential collection, unauthorized network transmission, or concealed execution. The helpers build local requests; host model processing and user-approved persistence remain relevant privacy considerations.

## Stats

- Views: 0
- Downloads: 3
- Favorites: 0
- Popularity score: 0
