Write Email

Produce send-ready cold email copy: specific trigger opener, your frame, one CTA, under the word cap. Input: prospect name + company (ideally with an account brief already run; if not, do a 2-minute mini-research pass first).

Before you write

1. Read ${CLAUDE_PLUGIN_ROOT}/references/messaging.md and references/guardrails.md if not already in context.
2. Look for voice-profile.md in the working folder. If found, match its voice traits and phrase bank. If not found: draft in the house style anyway, but tell the rep — "Run /voice-interview once (15 min) and every draft after that will sound like you."
3. Look for learnings.md in the working folder (maintained by the what-works skill). Default to its "Proven" patterns, never reuse anything under "Killed."

The locked structure (from the approved play)

Every cold email follows this shape — it replaced a version that read like a mass sequence, so don't drift back:

1. Trigger line — one sentence, a specific observed fact: "Saw [the specific trigger you found] at [Company] — [one line on why it means the pain is live now]."
2. Frame — the problem, framed our way. For Hook #2: "[your locked outbound frame from messaging.md]."
3. Trap question — a question whose honest answer is the pain: "[a trap question whose honest answer is the pain — see messaging.md] For most teams the honest answer is 'they don't.'"
4. One proof sentence — compressed, no bullets: "[approved proof points from messaging.md]." [⚠️ confirm clearance] until the rep confirms names are cleared.
5. One CTA — "Worth 15 minutes?" or the free-trial angle for Champion types ("Here's the free tier — no sales call required").
6. Optional P.S. — your current campaign (from messaging.md, if one is live), one link, one reason to care.

Hard rules

• ≤80 words for a cold first touch (P.S. excluded). Count them.
• One CTA. One link (two only if one is the campaign P.S.).
• Subject: lowercase, specific, internal-looking — "[specific trigger] at [company]", never "Quick question" or anything with a colon-and-title format.
• Persona check: Champion gets hands-on/technical language, Practitioner gets program-owner language, Economic Buyer gets margin/board language. Never mix (see messaging reference).
• Banned words and openers per guardrails — run the final draft against the banned list before output.
• Follow-up/bump emails: reply in the SAME thread, 2 sentences max, structured like the approved Email 2 ("Wanted to make sure you saw this…" + one link).

AI-writing check (mandatory, before output)

Reread the draft and fix: em-dash overuse, rule-of-three lists, "It's not X, it's Y" constructions, hollow adjectives, any sentence the rep wouldn't say out loud on a call. The test: would a busy buyer believe one specific human wrote this to them specifically?

Output format

For each prospect:

**To:** [Name, Title, Company] · persona: [Champion/Practitioner/Economic Buyer] · hook: #[N]

Subject: [subject]

[body]

P.S. [if used]

---
Personalization source: [the fact + link the trigger line is built on]
Word count: [X]/80 · Checks: voice [✓/no profile] · banned-words ✓ · clearance [✓/⚠️]

If drafting 3+ emails, vary the trigger lines and trap questions so they don't read as one template when forwarded around a team (they get forwarded).