SmoothTalker

Helps your AI agent write smoother, kinder, more effective messages. Declining, apologising, negotiating, following up, delivering bad news, setting boundaries, answering customers.

What it does

SmoothTalker is a library of 89 conversation playbooks. When you ask your agent to write, reply to, soften or improve any message, it consults the matching playbook and gives you two ready-to-send versions, one warmer and one more direct. Every playbook has the same shape: goal, structure, principles, examples at several tones, and what to avoid.

Smooth means clear, warm and low-friction. It never means slippery or manipulative. When you are being wronged, SmoothTalker makes you clearer, not softer.

Things you can ask

"My landlord wants to raise rent by $300. Help me push back."
"I forgot my best friend's birthday. Text her."
"Reply to this customer DM, she's asking for a discount."
"Draft a follow-up to the client who hasn't paid the invoice."
"My roommate sent a passive-aggressive text about the dishes. Reply."
"Counter this job offer, the base is below market."
"Text Maya after our first date, I want a second one."
"Make this email less cold."

How it works

Read-only. No account, no login, no data collected. The agent fetches a static JSON library once a day and applies it locally. Nothing you write is sent to SmoothTalker.

Categories	Core (8): how to use, principles, tone calibration, anti-patterns, phrase bank, reply to a pasted message, cross-cultural notes, situation router Personal (44), Work (26), Commerce (11)
OpenAPI	openapi.json
Library	data/all.json (whole library) · data/index.json
Auth	None
Source	github.com/fengyiqicoder/smoothTalker, content CC BY 4.0

Use it with Muse today

Until it appears in the Muse connector directory, you can add it as a custom connector: copy the prompt from MUSE_PROMPT.md and send it to Muse.

Use it with other agents

Claude Code, or any host that reads Agent Skills. Install the skill once. It loads the index and fetches only the playbook each request needs.

mkdir -p ~/.claude/skills/smoothtalker/reference
cd ~/.claude/skills/smoothtalker
curl -fsSLO https://raw.githubusercontent.com/fengyiqicoder/smoothTalker/main/skill/SKILL.md
curl -fsSL -o reference/principles.md https://raw.githubusercontent.com/fengyiqicoder/smoothTalker/main/skill/reference/principles.md

ChatGPT, Claude or Gemini projects. Download all.json, add it as a file to a project (or a custom GPT's knowledge, or a Gem), and paste this into its instructions:

When I ask you to write, reply to, soften or improve a message, use the SmoothTalker playbooks in all.json. Always follow the entries 00-how-to-use and 01-principles. Pick the one or two playbooks whose triggers and summary fit my situation (06-situation-router if unsure), build the message from the playbook's structure with my real details, and give me two ready-to-send versions, one warmer and one more direct, unless I ask for one. If I say what I want, do that. Never make me softer when I'm being wronged, and never write anything manipulative. End with the id of the playbook you used.

Download a fresh copy now and then; the library keeps growing.