AgentSkillsCN

create-voice

通过写作样本创建语音档案——让 mark-my-words 了解您真实的写作方式。

SKILL.md
--- frontmatter
name: create-voice
description: "Create a voice profile from writing samples — teach mark-my-words how you actually write"
disable-model-invocation: true
allowed-tools: Read, Write, Bash, Glob, Grep, AskUserQuestion
argument-hint: "[voice name]"

Create Voice Profile

You are creating a voice profile from the user's writing samples. The profile will be a compact style guide that captures how they write — not generic AI voice, their actual voice.

Steps

1. Load Configuration

Read .claude/mark-my-words.local.md to get the user's settings. If the file doesn't exist, tell the user:

No configuration found. Please run /mark-my-words:setup first to configure your blog settings.

Then stop.

2. Ensure Voices Directory

Check if .claude/voices/ exists. If not, create it.

3. Get Voice Name

If $ARGUMENTS was provided, use it as the voice name (convert to kebab-case: lowercase, hyphens for spaces, no special characters).

If no arguments, use AskUserQuestion to ask:

What should this voice profile be called?

Suggest names based on common patterns: casual-tech, professional, personal-blog, tutorial-voice. The name becomes the filename and identifier.

4. Check for Existing Profile

Check if .claude/voices/<name>.md already exists. If it does, tell the user:

A voice profile named "<name>" already exists. Use /mark-my-words:update-voice <name> to refine it, or choose a different name.

Then use AskUserQuestion to offer:

  • •Choose a different name
  • •Overwrite the existing profile

If they want a different name, go back to Step 3.

5. Get Description

Use AskUserQuestion:

Describe this voice in one sentence. What does it sound like? (e.g., "Conversational technical writing with dry humor" or "Direct and opinionated, like explaining something to a friend")

6. Gather Writing Samples

Use AskUserQuestion:

How would you like to provide writing samples?

  • •Point to files — I'll give you file paths or a glob pattern
  • •Paste text — I'll paste samples directly into the chat
  • •Both — files and pasted text

You need enough samples to identify patterns. Aim for at least 2-3 samples, ideally 1000+ words total.

If "Point to files":

Use AskUserQuestion to get file paths or a glob pattern (e.g., content/blog/*.md, or specific paths).

Use Glob to find matching files. Read each file. Track the source info:

yaml
- type: file
  path: <absolute path>

For glob patterns, track as:

yaml
- type: glob
  pattern: "<pattern>"
  count: <number of files matched>

If "Paste text":

Use AskUserQuestion to ask for their writing sample. Ask for a label for reference:

What's this sample from? (e.g., "Blog post about React hooks", "README I wrote last week")

Track the source:

yaml
- type: paste
  label: "<user's label>"

Ask if they have more samples. Repeat until they're done.

If "Both":

Do both flows above.

Important: Do not fetch URLs. If the user provides a URL, tell them:

Voice profiles work with local files and pasted text only — no URL fetching. You can paste the content directly or save it to a file first.

7. Analyze and Distill

Read all the gathered samples carefully. Analyze the writing for patterns across the six profile sections defined in references/voice-profile-format.md:

  1. •Tone & Register — formality, emotional range, reader relationship
  2. •Sentence Patterns — rhythm, length, distinctive structures
  3. •Vocabulary & Diction — word choice, technical level, characteristic phrases
  4. •Rhetorical Habits — how they explain, persuade, use humor
  5. •Structural Preferences — intros, sections, conclusions, formatting
  6. •Things to Avoid — anti-patterns that would break the voice

Be specific and concrete. Pull actual examples from the samples. Don't be generic — "uses humor" is useless, "drops self-deprecating asides in parentheses after technical claims" is useful.

8. Present for Review

Show the user the generated profile and ask:

How does this look?

  • •Save it — looks good
  • •Adjust it — I want to tweak some sections
  • •Start over — try again with different samples

If "Save it":

Continue to Step 9.

If "Adjust it":

Use AskUserQuestion to find out what to change. Update the profile and present again.

If "Start over":

Go back to Step 6.

9. Save the Profile

Write the profile to .claude/voices/<name>.md using the format from references/voice-profile-format.md. Include:

  • •Frontmatter with name, description, created date (today), last_updated (today), and sample_sources
  • •All six body sections with the distilled analysis

10. Set as Default?

Use AskUserQuestion:

Should "<name>" be your default voice for new posts?

  • •Yes — use this voice by default
  • •No — I'll pick a voice when I need one

If yes, read .claude/mark-my-words.local.md, update the default_voice field to <name>, and write the file back.

11. Confirm

Tell the user:

Voice profile "<name>" saved to .claude/voices/<name>.md.

If set as default:

It's now your default voice — new posts will use this style automatically.

Otherwise:

You can use it by picking it when creating a post, or set it as default later by updating default_voice in your config.

Run /mark-my-words:update-voice <name> to refine it with more samples.