writewell-onboard - SKILL.md Agent Skill

name: writewell:onboard version: 1.0.0 description: | Create a personal voice profile for /writewell by analyzing writing samples. Provide individual post URLs or a blog homepage URL with a post count. Fetches the content, distills the writing style, and generates voice/SKILL.md. allowed-tools: - Read - Write - Edit - WebFetch - WebSearch - AskUserQuestion - Glob

Write Well: Onboard

Create a personal voice profile by analyzing writing samples from the web.

Input Modes

Ask the user which mode they want to use:

Mode A: Individual post URLs

The user provides a list of URLs pointing to individual blog posts or articles. Fetch each one.

Mode B: Blog homepage + count

The user provides:

The URL of a blog's homepage or archive page
How many posts to analyze (default: 5)

Fetch the homepage, find links to individual posts (in reverse chronological order), then fetch that many posts. Use these heuristics to identify post links:

Look for <article> elements, .post / .entry classes, or similar blog patterns
Prefer links inside the main content area, not sidebars or footers
If the page is an archive/index, look for date-ordered links
If the homepage doesn't list posts directly, look for an /archive, /blog, /posts, or /writing page and try that instead
If you can't find individual post links, tell the user and ask them to provide direct URLs (fall back to Mode A)

Analysis

For each fetched post, extract the main prose content (skip navigation, ads, footers). Then analyze across all posts for these dimensions:

Sentence structure

Average sentence length (short/medium/long)
Sentence variety (uniform vs. mixed lengths)
Use of fragments vs. complete sentences
Comma style (sparse vs. liberal)
Semicolons, colons, em dashes -- how often and for what purpose
Exclamation marks -- frequency and context

Paragraph structure

Typical paragraph length (sentences per paragraph)
Single-sentence paragraphs (common or rare)
How paragraphs open (observation, example, question, assertion)
Persuasive structure patterns

Tone

Formality level (casual / professional-warm / formal / academic)
Humor style (none / dry / playful / self-deprecating)
Directness (blunt / direct / hedged / indirect)
How the writer handles disagreement or criticism
Self-awareness and vulnerability

Vocabulary

Signature words and phrases that recur across posts
Jargon comfort level (avoids it / uses it naturally / leans on it)
Words and patterns the writer clearly avoids
Contractions (frequent / occasional / never)

Openings

How posts typically begin (thesis / context / anecdote / question / greeting)
First-sentence patterns

Closings

How posts typically end (CTA / question / summary / sign-off / just stops)
Last-sentence patterns

Formatting

Header usage (none / sparse / extensive)
List style (bullets / numbers / inline)
Bold/italic usage patterns
Links -- how they're woven in
Use of images, tables, code blocks

Rhetorical devices

Signature moves (rhetorical questions, analogies, data-heavy arguments, stories)
Parenthetical asides (frequent / occasional / rare)
How opinions are signaled vs. facts

Point of view

Default pronoun (I / we / you / one)
When the writer switches POV and why

Emotional register

How enthusiasm is expressed
How criticism is delivered
How the writer handles admitting mistakes

Distinctive touches

Anything unique that doesn't fit the categories above
Recurring structural patterns
Cultural or domain-specific references

Output

Generate voice/SKILL.md in the writewell skill directory using this template:

---
name: writewell:voice
version: 1.0.0
description: |
  Personal writing voice for [name or description]. Defines sentence structure,
  tone, vocabulary, openings, closings, formatting, rhetorical devices, and
  distinctive touches. Referenced by the main /writewell skill at Part 2.
---

# Personal Voice: [Name]

This file defines the writer's personal style. It is read and applied by the main
`/writewell` skill during Part 2 (Apply personal voice).

Source material: [list the URLs analyzed]

---

## Sentence Structure
[findings]

## Paragraph Structure
[findings]

## Tone
[findings]

## Vocabulary
[preferred words/phrases and words/patterns to avoid]

## Openings
[patterns found, with examples from the source material]

## Closings
[patterns found, with examples from the source material]

## Formatting
[findings]

## Rhetorical Devices
[findings]

## Point of View
[findings]

## Emotional Register
[findings]

## Distinctive Touches
[findings]

Process

Use AskUserQuestion to present this choice: Question: "How would you like to set up your writing voice?" Options:
- "Post URLs — I'll paste links to individual posts"
- "Blog homepage — analyze my N most recent posts"
- "Skip — use writewell without a personal voice" 1a. If the user picks "Skip", write voice/SKILL.md with the content below, then tell the user: "Got it — writewell will use general writing principles only. You can run /writewell:onboard any time to create a personal voice profile." Then stop.
```
---
name: writewell:voice
version: 1.0.0
description: No personal voice — use general writing principles only.
---
# No Personal Voice
No personal voice profile has been configured. Proceed without applying
any personal style rules. To create one later, run `/writewell:onboard`.
```
Fetch the content using the appropriate mode.
If any URL fails to fetch, report it and continue with the ones that worked. Need at least 2 posts to produce a useful profile.
Analyze all posts across the dimensions above.
Write the voice profile to voice/SKILL.md in the writewell skill directory.
Show the user a summary of the voice profile -- the key traits that define this writing style, in 5-8 bullet points.
Ask if they want to adjust anything before finalizing.

Important

The voice file should contain rules and patterns, not example text. Write it as instructions that an LLM can follow, not as a literary analysis essay.
Use concrete examples from the source material to illustrate rules, but frame them as "write like this" not "the author wrote this."
If the analyzed posts have inconsistent styles (e.g., some formal, some casual), note the contexts where each style appears rather than averaging them out.
If a voice file already exists, warn the user before overwriting it.