Train a Write-Like-Me Skill
Study one week of messages you wrote, then build a reusable skill that changes your voice by channel, relationship, and audience seniority.
What you'll need
- Claude with authorized Slack and email connections
- Access to messages you authored during the last seven days
- Enough sent messages to compare Slack with email and internal with external communication
- A folder where Claude can draft the skill for your review
Put it to work
Save these instructions in a Claude Skill or Project, or in a ChatGPT Project or Custom GPT. Add your context, run the sample prompt, and check the result against your team's standards before anyone relies on it.
Claude vs. ChatGPT
This version uses Claude with Slack and email connections, then packages the result as a Claude Skill. In another assistant, use its available connectors and save the same audience matrix in a Project or custom assistant.
Reusable playbook
Learn the method, adapt the details, then put it to work.
What this playbook helps you do
Build a reusable Write-Like-Me Skill from messages I authored in connected Slack and email accounts.
Follow the playbook
Boundaries
- 1
Read only messages I sent during the last seven calendar days. Use incoming messages only for the minimum context needed to classify the audience and objective.
- 2
Exclude private, legal, medical, financial, HR, performance, credential, and security-sensitive conversations from the training sample.
- 3
Analyze patterns, not secrets. Do not copy source messages, names, email addresses, company details, links, commitments, or confidential facts into the finished skill.
- 4
Create synthetic examples that demonstrate the patterns without reproducing a real conversation.
- 5
Do not send, post, reply, or modify anything in Slack or email. This workflow is read-only until I explicitly approve a separate draft.
Analyze the sample
For every usable message, classify three dimensions independently:
- Dimension
- Channel
- Options
- Slack or email
- Dimension
- Relationship
- Options
- Internal team or external
- Dimension
- Audience seniority
- Options
- Executive or coworker
| Dimension | Options |
|---|---|
| Channel | Slack or email |
| Relationship | Internal team or external |
| Audience seniority | Executive or coworker |
Treat founders, C-suite leaders, VPs, board members, investors, and senior client decision-makers as executives. Treat teammates, operators, project partners, clients, and prospects without senior decision authority as coworkers. Infer the class from titles, domains, signatures, channel membership, and thread context. Mark uncertain cases as unknown instead of guessing.
Then identify
- Stable voice constants that appear across audiences
- Sentence and paragraph length, contractions, greetings, signoffs, bullets, punctuation, enthusiasm, and directness
- How much context, polish, candor, and formality change in each audience mode
- Slack-specific and email-specific habits
- Phrases or habits to avoid because they do not sound like me
Build four audience modes
Create separate guidance for
- 1
Internal coworker
- 2
Internal executive
- 3
External coworker
- 4
External executive
Do not collapse these into a single formal-versus-casual scale. Relationship and seniority are separate decisions. Apply channel conventions after selecting the audience mode.
Skill behavior
The finished skill must
- 1
Identify the channel, objective, relationship, and audience seniority before drafting.
- 2
Default an unknown recipient to external coworker unless the message is high impact and needs clarification.
- 3
Preserve every supplied fact, date, name, link, commitment, and decision.
- 4
Adapt the amount of context, polish, candor, greeting, signoff, and next step to the selected mode.
- 5
Keep Slack short and conversational. Keep email complete but concise.
- 6
Never use em dashes.
- 7
Never invent authority, commitments, availability, familiarity, or facts.
- 8
Never expose private source messages or say the voice was learned from connected accounts.
- 9
Draft only. Never send without explicit approval.
Deliverables
Before writing files, show me
- The number of usable messages analyzed in each channel and audience mode
- The proposed voice constants
- The proposed four-mode audience matrix
- Any weakly supported mode that needs more examples
- The files you plan to create
After I approve, create
write-like-me/ ├── SKILL.md └── references/ └── voice-profile.md
Keep SKILL.md focused on trigger conditions and drafting behavior. Put the detailed voice evidence, audience matrix, channel rules, avoid list, and synthetic examples in references/voice-profile.md.
Try this prompt
Train a Write-Like-Me Skill from messages I authored in the last seven days. Use my connected Slack and email accounts. Keep internal versus external and executive versus coworker as separate audience decisions, then adjust for Slack versus email. Analyze only. Show me the proposed voice profile, audience matrix, sample counts, and file plan before creating the skill. Do not send or change any messages.
Before you trust the output
- Review the sample counts before approving the skill so one audience mode does not get inferred from too little evidence
- Keep only generalized patterns and synthetic examples in the skill, never private messages or identifying details
- Retest all four audience modes with the same facts, then confirm the relationship and seniority differences are visible without changing the facts
- Refresh the voice profile periodically with a new sample, but review the proposed changes before replacing a skill that already works
Your next step
Want this workflow to run reliably every week?
Bring Train a Write-Like-Me Skill to our free live workshop. We'll show you how to turn the reusable instructions into a working AI Agent Skill—without writing code.