Skip to content

/reply-from-superhuman

Draft a reply, a follow-up, or a new email in Superhuman in the user's own voice — read the whole thread first, reuse an existing draft instead of adding a second one, keep the right people on the thread, and leave sending to the user unless they asked for it. Use when the user

BOOST
From plugin
superset
15k34 skills1 agent1 command
Install
$ npx -y skills add superset-sh/superset --skill reply-from-superhuman --agent claude-code

How it fires

How this skill gets triggered: by you, by Claude, or both.

  • Fires itselfAuto-invocation. Claude auto-loads it when your prompt matches the work.Auto-invocation is when the right skill fires by itself at the right moment, driven by a FLOW.md router and a hook, instead of you invoking it by name. It is the difference between a skill being installed and a skill actually getting used.Read the full definition →
  • You can call itInvoke it directly when you want it.
  • Slash command/reply-from-superhuman

Context preview

The summary Claude sees to decide when to auto-load this skill.

Draft a reply, a follow-up, or a new email in Superhuman in the user's own voice — read the whole thread first, reuse an existing draft instead of adding a second one, keep the right people on the thread, and leave sending to the user unless they asked for it. Use when the user

SKILL.md

reply-from-superhuman.SKILL.md
name: reply-from-superhuman
description: Draft a reply, a follow-up, or a new email in Superhuman in the user's own voice — read the whole thread first, reuse an existing draft instead of adding a second one, keep the right people on the thread, and leave sending to the user unless they asked for it. Use when the user says reply to, answer, follow up with, draft an email to, or send, or when a triage turned up threads that need a response.
argument-hint: the thread or person to write to, and what the email should say
allowed-tools: mcp__superhuman__list_accounts, mcp__superhuman__list_threads, mcp__superhuman__get_thread, mcp__superhuman__get_message, mcp__superhuman__get_attachment, mcp__superhuman__list_drafts, mcp__superhuman__get_draft, mcp__superhuman__get_draft_send_status, mcp__superhuman__create_or_update_draft, mcp__superhuman__discard_draft, mcp__superhuman__list_snippets, mcp__superhuman__get_snippet, mcp__superhuman__undo_send

Write it as the user would, and let them send

A draft in Superhuman is the user's to send; the job is to make it the one they would have written. Read the whole thread, draft once, and stop at the draft.

1. Read the thread end to end

`get_thread` with drafts and comments included. The last message is not the whole ask: an earlier message often holds the question, a team comment may say how to answer it, and the question itself may sit in an attachment. `get_attachment` returns images and audio directly; for anything else it returns a download link, so a PDF or a spreadsheet is unread until the user opens it. When the thread is trimmed, raise `message_limit`. Note who is on the thread and who spoke last; the reply goes to the person waiting, not to whoever wrote first.

2. Look for a draft before writing one

`list_drafts` filtered by `thread_id`. If a draft exists, `get_draft` and read it: the user may have started a reply in Superhuman, and the right move is to finish theirs, not to add a second one. Check `get_draft_send_status` before touching it. `pending` means it is already on its way, and a draft the user sent from Superhuman minutes ago looks exactly like one that still needs work.

3. Draft with instructions, not prose

`create_or_update_draft` with `type` `reply` or `reply_all` and the thread id for a thread, or `type` `new` with `to` for a fresh email, where `subject` can be left for the writer and `from` picks one of the user's aliases. Give it `instructions` (what to say, what to ask, what tone the thread calls for) and let it write in the user's style and signature. Pass `body` only when the user dictated exact wording, or when forwarding, where `body` holds the intro and the forwarded message is appended for you.

  • `reply_all` when others on the thread need the answer; `reply` when the user

is answering one person. Keep the CCs the thread already had.

  • Omit `subject` on a reply; it inherits the thread's.
  • One question per email. Say what was done and what is asked, and nothing

about what you plan to do later.

  • Team snippets (`list_snippets`, `get_snippet`) hold wording the team already

agreed on for common replies; start from one when it fits.

To revise, pass `draft_id` and `thread_id` back so the draft updates in place. Recipient fields replace rather than merge, so pass the full list each time. An update can rewrite the subject, so when it carries a prefix or a ticket tag that must survive, discard the draft and recreate it instead.

4. Stop at the draft

Report the draft: to, cc, subject, the body, and that it is in the user's Drafts in Superhuman. Sending is the user's action, so this skill does not pre-approve `send_draft` and a send always asks first. When the user does ask you to send, set an `undo_timeout` and report the undo token, and call `undo_send` the moment they say stop. `send_at` schedules a send; `smart_send` lets Superhuman pick the time.

Anti-patterns

  • **Drafting from the snippet.** The ask is in the thread, not the preview.
  • **A second draft on a thread that has one.** Finish the existing one.
  • **Replying to the wrong person.** The reply goes to whoever is waiting; the

thread's starter may have left it long ago.

  • **Sending because the user said "reply".** Reply means draft. Send means

send.

  • **Writing in your voice.** Instructions produce the user's style; a pasted

body produces yours.

Read more
Ships withsuperset

Superset is an agentic IDE to orchestrate 100+ coding agents in parallel. Run any agent with your own subscription.

Get the whole plugin, auto-invoked

Other skills on superset.