reply-draft

Generates full-text reply drafts for inbound inquiries using verified context and personal writing rules.

Updated Jul 7, 2026
One-click install
npx skills add https://github.com/hivgb1-ai/do-better-workspace-v2 --skill reply-draft-hivgb1-ai
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: reply-draft
Source: https://github.com/hivgb1-ai/do-better-workspace-v2/tree/main/.claude/skills/reply-draft
Command: npx skills add https://github.com/hivgb1-ai/do-better-workspace-v2 --skill reply-draft-hivgb1-ai

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Writing replies to inbound inquiries, follow-ups, and confirmations is repetitive and error-prone: it is easy to invent details, echo the sender too much, or make promises you cannot keep. This Skill produces one complete, verified reply draft grounded only in facts confirmed in your workspace, while leaving the actual sending decision to you. ## Core Features & Use Cases - Context-grounded drafting: Searches past replies and progress trackers for similar cases and checks real calendar or schedule data before mentioning dates, so nothing is fabricated. - Rule-based validation: Greps the draft against your personal forbidden-phrases list, verifies any mentioned attachment actually exists on disk, and tags every sentence to a verifiable source. - Full-text output with review trail: Always outputs the complete draft plus a judgment summary, open questions, and a PASS/FAIL verification table; feedback triggers a full rewrite, not word-level patching. - Use Case: A client emails asking about availability and pricing. You paste the original message, and the Skill checks your schedule and past similar replies, then produces a complete draft in your tone with a verification report. You review it and only say "send it" when satisfied. ## Quick Start Paste the sender's original message and say "write a reply to this inquiry" to receive a complete verified draft.

Frequently Asked Questions about reply-draft

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I generate an email reply draft with AI?

Provide the sender's original message and ask for a reply draft. The Skill analyzes the thread stage, searches past replies for similar cases, checks real schedule data, and outputs a complete draft with a verification report before you decide to send.

How does the reply draft avoid making up information?

Every sentence in the draft must trace to one of three sources: something you actually said, a fact in your documents, or a verified fact. Sentences that cannot be tagged to a source are deleted, and schedule mentions require checking real calendar or tracker data.

Can the AI send the reply automatically?

No. The Skill only prepares the full draft; sending always requires your explicit confirmation such as saying "send it". After sending is confirmed, it can propose updating a tracking sheet if one is declared in your workspace.

How do I customize tone and forbidden phrases for replies?

Tone rules, forbidden expressions, signature, and default attachment paths live in a personal dictionary YAML file under the workspace registry. The Skill greps each draft against the forbidden-phrase list and rewrites any matching sentence.

What happens when I give feedback on a draft?

Feedback triggers a full rewrite, not word-level patching. The Skill returns to the context-analysis step and re-runs the entire validation pipeline so the revised draft passes all checks again.