Skip to main content

Agent pack

Support reply drafter

Support replies drafted from your real answers, sent only by your people.

By
Caleb and Ryan, the founders
Updated
September 2, 2026
Version
0.1, in the library
  • Claude Code
  • Anthropic API

Reads each incoming support message, finds how your team has answered the same situation before, and drafts the reply from those answers in your support voice. New situations get flagged as new instead of improvised. Your team reviews and sends; the drafter never talks to a customer alone.

Built, documented, running.

Part of the working library we deploy from on every engagement.

Bring this into a project

Runs on

  • Claude Code
  • Anthropic API

2 requirements, below.

  • Complete and documented: the same files we run ourselves.
  • Deployed and tuned to your stack as part of an engagement.

The file

Draft excerpt
# Always straight to a person anger: any message reading as upset, no draft, human firstmoney: anything touching billing or commitmentslegal: anything mentioning terms, data, or lawyersnew: no precedent found -> flagged as new, never improvised

What is inside

3 items. Includes deployment guide.

  • AGENT.mdThe drafter: find precedent, draft, flag the new
  • escalation-rules.mdWhat must always reach a person untouched
  • guide.mdSeeding from your ticket history

Who it is for

  • The same questions get re-answered from scratch daily.
  • Answer quality depends on who picks up the ticket.
  • You want faster support without a bot going rogue on customers.

Requirements

  • A history of answered tickets or emails
  • A review step your team actually keeps

Where this sits

  1. This system

    Support reply drafter

    In the library, running.

  2. In your stack

    Deployed on an engagement

    The audit finds which system earns its place first; we install and tune it around how you work.

Changelog

  1. September 2, 2026In the library. Version 0.1 runs in our own operation.

Related

  1. Agent pack

    Research analyst agent

    A research question becomes a sourced brief, with every claim carrying its source

    Works with Claude Code, Anthropic API.

    Built and running

  2. Agent pack

    Inbox drafting agent

    Every email that needs a reply gets a draft in your voice, waiting in your drafts folder

    Works with Claude Code, Anthropic API.

    Built and running

  3. Agent pack

    Document review agent

    Every outbound document gets the careful second read, against your checklist

    Works with Claude Code, Cursor.

    Built and running