Skip to content
T1KTheOneKit

human-writing-before-send

Activate Human-Writing Before Any Email / Message / Post / Pitch Deck / Proposal / Presentation

Section titled “Activate Human-Writing Before Any Email / Message / Post / Pitch Deck / Proposal / Presentation”

Always-loaded companion to the human-writing-reminder.cjs hook. The hook fires only on UserPromptSubmit; mid-turn injected user messages (the “The user sent a new message while you were working” path) are NOT UserPromptSubmit events, so the hook reminder never fires for them. This rule closes that gap — it is in context every session regardless of the hook.

Rule

Section titled “Rule”

Before you draft, finalize, or send ANY email, chat message, LinkedIn/X post, pitch deck, proposal, slide deck, presentation, one-pager, cover letter, or other human-facing prose, activate the t1k-human-writing skill (guide mode for a fresh draft, lint mode for an existing one). This applies even when:

  • The request arrived as a mid-turn injected message (no hook reminder appeared), OR
  • You are composing the draft yourself rather than editing a pasted one, OR
  • The message “looks routine” (invoice cover notes, quick replies, acknowledgements).

Why

Section titled “Why”

AI-authored business prose reliably accumulates AI tells — the most common being the em-dash (—), which alone makes an email read as machine-written. A single lint pass before finalizing removes them. In a 2026-07-06 session, two invoice cover emails were drafted directly (skill skipped, because the request came as a mid-turn injection) and both filled with em-dashes; the miss was only caught on explicit user prompt.

2026-07-13 — the pitch-deck gap. A user asked to “create a pitchdesk” (a client pitch deck) and the hook stayed silent, so the skill never activated — a whole client deck was authored without a human-writing lint pass. Root cause: the hook’s ACTION regex had no create/build/make, and its ARTIFACT regex had no deck/pitch deck/slides/ presentation/proposal. Neither matched. The fix widened both regexes and added a deliverable-only matcher (fires on the noun alone, catching the “pitchdesk” typo too), and this rule’s scope now covers decks/proposals/presentations explicitly. Human-facing prose is not just email — a pitch deck’s headline, body copy, and card endings carry the same AI tells.

How to apply

Section titled “How to apply”
  1. Recognize the trigger: any email/message/post/pitch deck/proposal/presentation is about to be drafted or sent.
  2. Activate t1k-human-writing FIRST — do not draft prose before it.
  3. Hard check before finalize: zero em-dashes in outbound prose bodies; no slop vocabulary, no rule-of-three, no throat-clearing opener, no summary-conclusion closer.
  4. Slide-deck nuance: on a deck, an em-dash in a badge/title/stat separator (“Growth — 15%”) is fine — it is a layout device, not a prose tell. The real deck tell is a repetitive “statement — punchline” ending on card after card, plus rule-of-three bullets and slop headline vocab. Kill the repetition and vary the card endings; don’t blanket-ban the dash in separators.
  5. Match the author’s real voice (voice-inject from their past sends when available).
Section titled “Related”
  • ~/.claude/skills/t1k-human-writing/SKILL.md — the skill itself (guide + lint modes)
  • hooks/human-writing-reminder.cjs — the UserPromptSubmit hook this rule backstops
  • rules/image-analysis-routing.md — the parallel always-loaded-rule + hook pattern