x-copywriter

Draft tweets and threads that read like an operator typed them on their phone, not a comms team approved them. You may be new to X · this skill teaches the platform's culture as it drafts (rationale next to output) and intentionally does NOT ground voice in your LinkedIn posts. The voice anchors are three reference accounts: Romàn (@romanbuildsaas, metrics-first), Michel Lieben (@MichLieben, story

• self-roast), Alex Vacca (@itsalexvacca, operator-with-receipts). Their actual top tweets live in reference-corpus.md. Platform culture rules live in culture-brief.md.

When to invoke

Triggers: "write a tweet", "draft an x post", "tweet about Y", "x thread on X", "/x-copywriter".

Do not use for:

• LinkedIn posts. The voices are different and you'll drift if you mix.
• Long-form (newsletter, blog).
• Replies to other people's tweets. Replies are a different game (specificity + speed) and shouldn't go through this skill.

Workflow

1. Brief

One batched prompt:

Quick brief.

1. What's this tweet about? (milestone, lesson, hot take, 
   build-in-public, IP-giveaway, self-roast, etc.)
2. The single specific fact, number, or scar that anchors it. 
   Be concrete. ("crossed $50k MRR yesterday" beats "growing 
   nicely"). If we don't have a number, can we add one?
3. Solo tweet or thread? If unsure, I'll pick.
4. Anything off-limits? (NDA pre-launch features, exact ARR if 
   we're being cagey, etc.)

If the user already gave enough, skip the prompt and parse.

2. Pick the format

Default decisions:

• Solo tweet by default. Most ideas should ship solo. Threads are reserved for ideas that genuinely need 4-7 beats.
• Thread if: tactical breakdown with steps; story arc with a before / middle / after; list with examples per item.
• Two solo tweets beats a 3-tweet thread. If a thread is < 4 tweets, split it into two solos that can each stand on their own.

3. Pick the hook

Match the brief to one of these 8 patterns. All come from the reference corpus. Read reference-corpus.md if you haven't recently.

Hooks to never use on X (LinkedIn imports):

• "Excited to announce..."
• "Thrilled to share..."
• "Big news!"
• "Here are 7 lessons from..." (as solo tweet, not thread hook)
• "Most people think X. They're wrong." (LinkedIn cliché now even on X)
• "Day 1 at..." career update
• Any hook that needs a "1/" because the actual idea hasn't started

4. Pick the body shape

For solo tweets, pick one of:

S1. Bare metric + 1-line context. 60-100 chars total.

"We just hit $1.5M ARR with GojiberryAI" "227,744 views. 3,316 comments. 477 leads. In 7 days. One Claude system."

S2. Number stack with line breaks. 3-5 numbers, one per line, no bullets. Closes with one short opinion or implication line.

"We turned 24 employees into LinkedIn influencers in 90 days.

581 posts. 43,473 reactions. 28,130 comments. 27 new clients worth $153K in MRR."

S3. Quoted strawman + counter. Open with the cliché in quotes. Follow with > or hyphen-prefixed lines that demolish it. Lowercase fine here.

"'services don't scale'

chases the $10k software budget inside companies watches openai kill the feature in 3 months cpo title before a paying user, 18 months burning runway"

S4. Single-line shitpost. 1 line, lowercase, no follow-up. No CTA. The whole point is the line.

"cool so we're just giving away the entire operating system now" "absolute insanity if you ask me."

S5. Numbered listicle. 3-5 short clauses, one per line. Sentence case, period or no period at end (consistent within tweet). Title line first.

"3 things i wish i knew at $0 ARR:

1. distribution > product
2. write the offer before the code
3. one channel, all in"

For threads, pick one of:

T1. Tactical breakdown. Hook tweet (works alone). 4-6 step tweets, each one numbered (2/, 3/, etc.) or unnumbered. Last tweet = CTA: "follow <YOUR_X_HANDLE> for more on building <YOUR_BRAND>" or "comment 'X' and i'll send the doc".

T2. Story thread. Hook = scar / failure stack. Beat tweets walk the arc: bottom → turning point → current state. End with the lesson + soft follow CTA.

T3. List thread. Hook = "X mistakes I see at $Y ARR companies" or "Y plays we ran for 100+ clients". Each subsequent tweet = one item with a one-line example. End with a "want the full doc?" giveaway CTA or a follow CTA.

5. Voice register

Use lowercase when:

• Writing a self-roast, shitpost, or one-line dunk (S4, H4)
• The tweet is a quick reaction to something happening
• You're trying to sound like you typed it on your phone

Use sentence case when:

• Posting a milestone or hard data (S1, S2, H1, H3)
• Authority / hot-take / observation (H8 even though that example uses lowercase, the gravitas register works either way)
• Threads usually start in sentence case in the hook tweet

Mix is normal. A milestone tweet can have a sentence-case hook and a lowercase shitpost line at the end. Don't over-engineer it.

Punctuation:

• Periods at the end of sentences are fine. Many high-performers drop the final period in casual mode.
• Line breaks: 1-2 max in a solo tweet, more freely in threads.
• Em dashes (—): never. Hard ban, project-wide rule.
• Smart quotes: never. Straight quotes only.
• Ellipsis: use ... if needed at all.
• > for nested bullet / quoted-cliché lists (Mich/Alex use this).
• → for sub-arrow bullets in threads (sparingly).

Vocabulary to use:

• Plain English. "use" not "leverage". "help" not "empower".
• Numerals always. "3x", "$2M", "23,000,000 sends".
• Tool names with their canonical casing (Claude Code, Cursor, Instantly.ai, HeyReach, Clay, Apollo, n8n, Stripe).
• "Build", "ship", "drop", "spin up". Active verbs.
• Self-deprecating shorthand fine: "lol", "ngl", "tbh" (use sparingly, not in every tweet).

6. CTA patterns

Most tweets need no CTA. The post is the asset. Forced CTAs are the strongest LinkedIn-import tell.

When you do add one, pick one of:

CTA-A. Comment-and-send (giveaway only).

Comment '<KEYWORD>' and I'll send the [thing].

Variant: "Repost ♻️ + comment 'X' and I'll send it." The repost-gate is X-native and works.

CTA-B. Follow for more (threads). Last tweet of a thread.

follow <YOUR_X_HANDLE> for more on building <YOUR_BRAND> in public.

CTA-C. RT the first tweet (threads). Equivalent of the LinkedIn "share if useful" but X-native.

RT the first tweet if this was useful.

CTA-D. None. Default for solo tweets. Authority posts and shitposts work better with no ask.

Banned closers (LinkedIn import):

• "Agree?"
• "Thoughts?"
• "What do you think? Drop a 👇 in comments."
• "Tag someone who needs this."
• "Save this for later."
• "Like and follow for more value."
• "Hope this helps."

7. Slop scrub (mandatory pre-output)

Before showing the draft, scan it. Any hit → rewrite.

Punctuation:

• ❌ Em dash —. Never. Replace with period, comma, colon, parens, or > arrow.
• ❌ Smart/curly quotes. Straight only.
• ❌ Ellipsis character …. Use ....
• ❌ Multiple hashtags. Or one hashtag. Zero hashtags on X.

Banned phrases: "excited to announce", "thrilled to share", "big news", "in today's fast-paced world", "dive into / deep dive", "delve", "unlock the power", "navigate the landscape", "at its core", "it's important to note", "a testament to", "game changer", "move the needle", "stay ahead of the curve", "cutting-edge", "state-of-the-art", "robust", "scalable", "seamless", "world-class", "leverage" (verb), "synergy", "embark on", "foster", "drive impact", "buckle up", "let that sink in", "read that again", "spoiler:", "plot twist:", "i'm humbled to announce", "couldn't be more proud", "trust the process", "your network is your net worth", "hope this helps".

Banned structural slop:

• ❌ "Here's the thing:" / "Here's the kicker:" / "Here's the catch:"
• ❌ Negation pivot 3+ times in one tweet ("It's not X. It's Y. It's not about A. It's about B.")
• ❌ Tricolons of parallel verbs in every line ("listen, adapt, deliver. faster, cheaper, better.")
• ❌ Throat-clearing transitions: "Furthermore," "Moreover," "Additionally," "That said," "Importantly," "Ultimately"
• ❌ Bold/italic Unicode tricks (𝗯𝗼𝗹𝗱).
• ❌ Keycap emoji as numbered headers (1️⃣ 2️⃣). Plain 1. or 1).
• ❌ Emoji as bullet markers (🚀 ✨ 💡). Inline emoji only, max 1 per tweet.
• ❌ Tagging 5+ people in one tweet.

Length scrub:

• Solo tweet must be ≤ 280 chars. Count it. Show the count in the output metadata.
• If a solo tweet hits 240-280, try cutting 1-2 words. Tight beats full.

8. Output

Show the draft in fenced code blocks. For each piece, include below the block:

• Format: solo tweet · thread · list thread, etc.
• Char count (if solo): 247 / 280.
• Hook pattern used: e.g. "H4 · self-roast comparison"
• Body shape used: e.g. "S2 · number stack with line breaks"
• Voice register: lowercase / sentence case / mixed
• Slop scrub: "0 hits" or list any flags caught + how rewritten
• Why it works (X-specific): 1-2 sentences explaining what makes this read as native X, not LinkedIn (this is the teaching layer for a newbie).
• One alt hook the user can swap in.

For threads, output every tweet in its own code block, char-counted individually. Note the hook tweet must work standalone before moving on.

Default output: 1 solo tweet + 1 alt solo tweet from a different hook pattern. If the brief asks for a thread, deliver the thread plus 1 solo tweet variant (in case the thread idea works better as a single tweet).

Do not add commentary like "Hope this works!" or "Let me know if you want changes." Just the drafts + metadata.

Content graph (close the loop)

X posts written here land in content/posts/ and feed the content knowledge graph (content/CONTENT-WIKI.md). Many X posts repurpose a longer original that is already a node.

Before drafting: open the relevant note in content/concepts/ and build on its canonical lines instead of starting blank.

After the post is saved: add it to the concept's ## Appears in (mark (X) in the angle), 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. Pure repurposes of a post already in the graph can skip this.

Editing mode

If the user pastes a draft and asks to "make it sound less LinkedIn" or "fix the slop":

1. Read the draft.
2. Run the slop scrub (§7). Flag every hit inline.
3. Rewrite to a hook + body shape from §3-§4.
4. Output the rewrite + a diff-style list of what changed and why.

Repurpose context (when called from a repurpose orchestrator)

If invoked with a source reference (master content row, transcript excerpt, or external URL):

1. Treat the source as the insight bank. Pick one hook / data point / hot take · don't rephrase the whole thing.
2. Open with the punch (a number, a contrarian line, a one-word setup), not with "I just published a webinar." X has zero patience for self-promo openers.
3. If the source is rich (multi-step framework, breakdown), prefer a thread. If it's a single insight, prefer a solo tweet.
4. Soft-reference the master only if it earns the slot (e.g., a thread tail with "full breakdown: <link>") · don't bolt it on otherwise.
5. Output the draft as usual. The orchestrator handles creating the tracking row and pasting Source: <URL> at the top of the page body · don't include that line yourself.

Cadence guidance (when user asks)

If the user asks "what should I post today?" or "how often should I tweet?", point them at culture-brief.md §6 (3-5 posts/day under 1k followers, 70/30 reply-to-original ratio, Tue-Thu mornings for B2B threads, pin a thread, don't quit at month 2).

Do not write a daily content calendar inside this skill. Drafting is per-tweet. Strategy is in culture-brief.md.

Reference files

• reference-corpus.md. Voice anchor. Top 15 tweets per reference creator (Romàn, Mich, Alex), pulled from the X API and sorted by weighted engagement. Re-read before drafting if you've been off-task. Refresh quarterly with the curl command at the bottom of this file.
• culture-brief.md. Platform rulebook. What X rewards / kills, voice register, cadence, reference-account stylistic reads. Pointer here when the user asks platform questions.
• This SKILL.md. Drafting workflow + hook/body/CTA patterns + slop blacklist.

Important: Your LinkedIn corpus (e.g. <YOUR_LINKEDIN_CORPUS_PATH>) is not a voice source for X. The two platforms are different games. Don't import LinkedIn voice fingerprints (the :) smiley, [1] [2] walking-CTAs, "Lead Gen Bros", → heavy bullets) into X drafts. Your default lane on X starts closest to @itsalexvacca's operator-with-receipts register, with seasoning from @MichLieben's self-roast and @romanbuildsaas's bare-metric format. Default voice anchors. You can swap or weight differently as you find your own taste over time.

Refresh the reference corpus

Run quarterly. Updates reference-corpus.md.

# Requires X API v2 bearer token in env as <YOUR_X_BEARER_TOKEN>
# e.g. export X_BEARER_TOKEN=AAAA...

# user IDs (cached from initial setup):
# romanbuildsaas: 1893653481858433024
# MichLieben:     512156315
# itsalexvacca:   859850213015597056

for uid in 1893653481858433024 512156315 859850213015597056; do
  curl -s "https://api.twitter.com/2/users/${uid}/tweets?max_results=100&tweet.fields=public_metrics,created_at&exclude=retweets,replies" \
    -H "Authorization: Bearer ${X_BEARER_TOKEN}" \
    -o /tmp/x_${uid}.json
done

Then sort by like_count + 3*retweet_count + 2*reply_count + 3*quote_count, pick top 15 per creator (filter URL-only tweets), and overwrite reference-corpus.md keeping the same shape.