newsletter-writer

Draft emails (newsletters, onboarding sequences, broadcasts) that read like you, not like ChatGPT. Renders into your brand email template and (optionally) pushes to Resend.

This skill fails its job if the output reads like generic SaaS copy. That is the test. Smoothness is the failure mode.

SETUP (run this once before drafting)

The voice comes from the same corpus the writer skills share:

1. Populate skills/linkedin-copywriter/corpus.md (copy from corpus.md.example and fill with your top posts). The newsletter voice is anchored to those posts.
2. Fill the placeholder tokens this skill uses verbatim in CTAs and sender lines: <YOUR_NAME>, <YOUR_BRAND>, <YOUR_DOMAIN>, <YOUR_EMAIL>, <YOUR_CAL_LINK>.
3. If you ship through Resend, set the env + IDs in step 8: RESEND_API_KEY, <YOUR_RESEND_AUDIENCE_ID>, <YOUR_RESEND_FROM>. Skip this whole block if you only want the drafted copy.
4. (Optional) Point <YOUR_EMAIL_TEMPLATE> at your master HTML email chrome. No template? The skill drafts the copy and you paste it wherever you send from.

When to invoke

Triggers: "write a newsletter", "draft the welcome email", "day-X check-in email", "broadcast about X", "/newsletter-writer".

In scope:

• Lifecycle / onboarding sequence emails (welcome, activation nudges, founder check-ins, win-back)
• One-off newsletters / broadcasts
• Transactional emails with marketing copy (receipts, account events that double as touchpoints)

Out of scope (use the other skill):

• LinkedIn posts → /linkedin-copywriter
• Long-form blog / docs → /long-form
• YouTube scripts → /youtube-script

Workflow

1. Brief

Quick brief.

1. Email type: welcome, activation nudge, founder check-in, broadcast/newsletter, win-back.
2. The single specific anchor: a number, a tool name, a person, a moment, or a real observation. Concrete beats generic. ("5 days in" / "you ran the install yesterday" beats "thanks for joining").
3. Primary CTA: install, book a call, browse resources, reply, share, upgrade.
4. Audience: all subscribers, free trial, paying, segment.
5. Send timing: trigger event + delay, or scheduled date.

Parse what the user gave; only ask for missing fields.

2. Voice anchors

Read ../linkedin-copywriter/corpus.md if you haven't this session. The top 5 posts there are your viral anchors. Internalise:

• Punchy opens. Often one short sentence or fragment.
• Numerals everywhere ("$480/year", "80k emails", "5 days in").
• Specific tool names (the real ones in your stack).
• Specific people by name + context.
• Opinionated takes that signal stake ("They are wrong.").
• Self-aware moments ("still a bit mindblown", "at least that's the plan").
• Uneven pacing. Fragments mixed with longer thoughts. One-word sentences.
• → arrows for choice-lists.
• Direct asks, not generic CTAs.

3. Adapt for email

• Subject: lowercase, < 50 chars, concrete. "5 days in." / "anything stuck?" / "welcome to <YOUR_BRAND>" beats "Welcome to the <YOUR_BRAND> Family!".
• Preheader: < 90 chars, specific hook. The inbox preview line.
• Body: 3-6 short blocks. One idea per paragraph. Three sentences is a long paragraph.
• CTA: one primary, button-styled. Optional inline secondary link. The button label is a specific verb, not "Get started" or "Learn more".

4. Slop blacklist — hard rejects

The model has trained on a million SaaS emails. These patterns are its defaults and they all signal AI. Reject every one. Re-read the draft hunting for them before approving.

Structural patterns to reject:

Voice patterns to reject:

5. Texture requirements — every draft must have

These are what make copy feel hand-typed, not generated:

• At least one specific number or proper noun in the first 30 words.
• At least one piece of texture: a specific tool name, a real observation, a self-aware aside, a time anchor, or a named person.
• At least one direct opinion or stake, especially in newsletters. ("They are wrong." / "Don't bother with X." / "This was a mistake.")
• Sentence-length variation. Drop in at least one fragment or one-word sentence somewhere.

If a draft has zero of these, the draft is slop. Rewrite.

6. Self-audit before approving

After drafting, do a re-read with this checklist:

[ ] Opens with a number, fragment, name, or claim (not setup phrasing)
[ ] No "The [adj] way to..." anywhere
[ ] No "X, Y, and Z" parallels
[ ] No "No X, no Y, no Z" enumerated negatives
[ ] No banned words
[ ] No em dashes
[ ] At least 1 specific number / proper noun in first 30 words
[ ] At least 1 texture moment (tool name / observation / aside)
[ ] CTA label is a specific verb, not generic
[ ] Sentence lengths vary
[ ] If sender name isn't dynamic, no "Hey [name]" greeting
[ ] If From carries identity, signature block is optional, often skipped

Any unchecked item = rewrite that section before showing the user.

7. Before / after examples

SLOP (what the LLM defaults to):

Hey,

The fastest way to feel what <YOUR_BRAND> does is to clone someone else's setup.

Templates are pre-built workflows · tables, integrations, and
dashboards that drop into your workspace in 30 seconds. No yaml, no
glue code, no chasing the right combination of providers.

The full library lives on the resources page · featured by us, plus
what the community is sharing.

[Browse the resources →]

Build something worth sharing? Hit reply with the slug and
I'll feature it.

What's wrong: "Hey," empty greeting · "The fastest way to feel" stem cell opener · "tables, integrations, and dashboards" three-parallel · "No yaml, no glue code, no chasing" enumerated negatives · "lives on the resources page · featured by us, plus what the community" smooth corporate transition · no specific numbers · no texture.

TIGHTER (your voice):

You can read about <YOUR_BRAND> all day. Easier to just clone someone's
setup and run it.

That's what the templates are. Pre-built workflows you drop into your
workspace. 30 seconds, no yaml.

The library is on the resources page · what we've shipped, plus what
people are sharing.

[Browse the resources →]

Built something useful? Reply with the slug. If it's good I'll feature
it on the page.

What changed: opens with a specific stake ("You can read about <YOUR_BRAND> all day"). One negative not three. Specific number ("30 seconds"). "If it's good" is an opinion. Same length but more weight.

8. Render + push to Resend (optional)

After the user approves the draft. Skip this block entirely if you just want the copy.

# Compose the HTML using your master template chrome (<YOUR_EMAIL_TEMPLATE>).

# Render preview
"/Applications/Google Chrome.app/Contents/MacOS/Google Chrome" \
  --headless --disable-gpu --no-sandbox \
  --window-size=700,1200 --hide-scrollbars --virtual-time-budget=2000 \
  --screenshot=lifecycle/<flow>/<NN-slug>-preview.png \
  "file://$(pwd)/lifecycle/<flow>/<NN-slug>.html"

# Push to Resend (newsletters = broadcasts, lifecycle emails = templates)
set -a; source .env; set +a

# For broadcasts:
python3 -c "
import pathlib, json
html = pathlib.Path('lifecycle/newsletters/YYYY-MM-DD-<slug>.html').read_text()
pathlib.Path('/tmp/bc.json').write_text(json.dumps({
    'audience_id': '<YOUR_RESEND_AUDIENCE_ID>',
    'from': '<YOUR_RESEND_FROM>',     # e.g. '<YOUR_NAME> <hello@updates.<YOUR_DOMAIN>>'
    'reply_to': '<YOUR_EMAIL>',
    'subject': '<SUBJECT_LINE>',
    'preview_text': '<PREHEADER>',
    'html': html,
}))
"
curl -s -X POST -H "Authorization: Bearer $RESEND_API_KEY" \
  -H "Content-Type: application/json" \
  --data-binary @/tmp/bc.json https://api.resend.com/broadcasts

# For lifecycle templates (then attach to an automation step):
# POST /templates with {name, subject, html, from, reply_to}
# POST /templates/{id}/publish

Persist resource IDs to a local state file (e.g. lifecycle/resend.json).

Quality bar

Before declaring done:

• Self-audit checklist all passed.
• Reads aloud like you on a call. If you imagine saying it to a customer over Zoom, ship it.
• One specific number or name in first 30 words.
• Single primary CTA, specific verb label.
• Renders correctly in the local Chrome screenshot (if you templated it).

Content graph (close the loop)

When an issue is saved as a content/posts/ file, it feeds the content knowledge graph (content/CONTENT-WIKI.md). Before drafting, pull the relevant note from content/concepts/ and build on its canonical lines instead of starting blank. After the issue ships, deposit its 1-3 core ideas back: add the post to each concept's ## Appears in (create a note if the idea is genuinely new), update any content/entities/ or content/mocs/ it touches, and append to content/log.md. Edges live in the concept notes (Obsidian backlinks are automatic), so don't edit the post body to add them.

Related

• Voice corpus (shared with /linkedin-copywriter): ../linkedin-copywriter/corpus.md
• Reference creators: ../linkedin-copywriter/reference-creators.md
• Brand voice rules: <YOUR_BRAND_DOC>
• Booking link for CTAs: <YOUR_CAL_LINK>