hank-x-drafter - SKILL.md Agent Skill

name: hank-x-drafter description: Draft tight X (Twitter) posts for Hank the Pro. Enforce 220-char hard limit, lowercase fragments, stripped narration. Use structural templates, verify char counts, batch-update to Airtable Drafts table. version: 1.1.0 author: Ace (CMO, Hank the Pro) trigger: | When asked to draft X posts (especially for Hank marketing, founder voice, or industry observations), or when Jonathan asks to "write for X" / "tweet this" / "draft 3 X variations."

CRITICAL: Length and structure rules (overrides earlier sections if conflict)

X posts are NOT short LinkedIn posts. They are different format entirely.

Hard length limits (must enforce)

Single post target: 100 to 220 characters. Hard ceiling: 270 characters. If you hit 270, you've failed the format. Cut.

Thread tweets: each piece 100 to 240 characters. Don't pack one piece. Spread the idea.

Character math (do this before writing)

Before drafting a single post, decide: this is one post (under 220 char) or a thread (multiple under-240 pieces). Don't write a single post that should have been a thread.

Test: read your draft. Can you cut a sentence and lose nothing? You're over.

Structural templates (use one, don't invent)

Template 1: Observation + Twist "watched [scene]. [implication]." Example: "watched a CSR field 47 calls in 4 hours yesterday. she's not a bottleneck, she's performing miracles." Char count: 117. Posts clean.

Template 2: Stat + Reframe "[number]. [reframe of what that means]." Example: "78% of homeowners hire whoever calls back first. trades software still treats phone calls as a backwater feature." Char count: 138.

Template 3: Setup + Punchline "[short setup]. [twist that lands]." Example: "6-truck plumbing shop just told 3 customers 'maybe wednesday.' that's $30k of follow-up work guessed away in 90 seconds." Char count: 152.

Template 4: Hot Take "[contrarian opinion stated flat]." Example: "ServiceTitan is right that the trades need software. wrong that it has to cost $50k and 6 months." Char count: 119.

Template 5: Question / Frame "[question that flips assumption]." Example: "why does dispatch software treat the phone call as the backwater? it's the entire game." Char count: 105.

Forbidden patterns (you keep doing these, stop)

DON'T explain the implication of your numbers. "47 calls in 4 hours yesterday. that's one call every 5 minutes. no break. no system." The reader did the math. Cut the math.

DON'T stack 4 short sentences when 2 would work. Pick the strongest 2.

DON'T put the punchline on the last line of a long setup. Lead with it or split into a thread.

DON'T use lowercase headings like "DRAFT 1 (Volume shock)" — reader doesn't see your scaffolding. Just write the post.

DON'T over-narrate. "heres what actually happens:" — show, don't say "here's what happens."

DON'T use ≠, ≈, or other symbols. Use plain words.

When to thread instead of single

If you have:

More than one beat (setup + escalation + punchline)
A list of 3+ items
A story with arc
A take that needs evidence

Use a thread. Each beat = its own tweet. 3 to 5 tweets ideal. 7+ is showing off.

Forced cut step (do this every time)

After drafting, cut 30 to 50 percent of words. Specifically:

Cut every "actually," "really," "just," "kind of"
Cut explanatory clauses ("which means...", "in other words...")
Cut ratio comparisons unless mathematically necessary
Cut the SECOND example when one example does the job

If your draft was 280 chars, target 180 after cuts. If 480 chars, you needed a thread.

Character verification and Airtable workflow (before updating records)

Always follow this checklist when saving X drafts:

Count each variation programmatically. Use Python: len("your text here") — paste each post and verify the count displays under 220 chars. If any exceed, flag it and don't save.
When updating Airtable Drafts table: Use MCP mcp_airtable_update_records or terminal curl to batch-update existing record IDs. Do not create duplicate records. Populate: Body (multilineText), Variant (single select: V1/V2/V3), Platform (single select), and Title (tracing label).
Never publish or post live. All drafts stay in Airtable with Status: "Todo" until Jonathan approves via Telegram. Post URL and Posted At fields remain empty until he gives the go-ahead.
Batch updates matter. When you have 3 record IDs ready, update all three in a single API call. It saves quota and prevents partial saves.

Example Airtable update payload:

{
  "records": [
    {"id": "recXXX", "fields": {"Body": "your post text here (168 chars verified)", "Variant": "V1", "Platform": "LinkedIn", "Title": "CSR 47 calls - V1"}},
    {"id": "recYYY", "fields": {"Body": "second post (174 chars verified)", "Variant": "V2", "Platform": "LinkedIn", "Title": "CSR 47 calls - V2"}}
  ]
}

Voice fingerprint check (per draft, before saving)

Read your final draft. Ask: does this sound like:

Alex Finn (fast, present-tense, lowercase fragments) ✓
Tommy Mello (operator-grounded, dollar specific) ✓
ChatGPT trying to sound human ✗
LinkedIn post copied to X ✗

If it reads like 3 or 4, redraft.

When variations should differ

When generating 3 variations, vary by:

LENGTH (one short ~120 char, one mid ~180, one approaching limit ~240)
STRUCTURE (one observation+twist, one stat+reframe, one hot take)
ANGLE (each surfaces a different beat from the same input)

NOT all 3 variations of the same beat in different lengths. That's lazy.

Humor calibration (add wry observations, sparingly)

Sterile drafts get ignored. But humor every post reads as trying too hard. Aim for: roughly 1 in 5 drafts has a humor beat. Not 1 in 5 are jokes. Just a slight wink.

What works in this lane

Dry observations: "watched a CSR put a customer on hold to take a call from her own husband asking what's for dinner. small business is mostly dinner logistics."

Self-deprecation that reveals something real: "shipped a feature today that broke 30% of our integrations. the support tickets are educational."

Industry inside jokes (only if you'd know it from being on the floor): "every plumber knows what 'i'm 5 minutes out' actually means and it's not 5 minutes."

Absurd specificity: "a $4M roofer just told me his best lead source last quarter was the homeowners next door to a job that went well. that's it. that's the system."

The pattern: humor comes from a TRUE observation that's slightly absurd, not from punchlines you wrote.

What doesn't work (avoid)

Don't write any of these flavors:

Pun jokes or dad jokes ("AI: artificially incompetent")
Twitter-bro snark ("ratio'd," "skill issue," "based," "anon")
Self-aggrandizing humor ("just shipped at 3am, who else gets it?")
Generic AI-takes-your-job jokes
"If you know, you know" / "iykyk"
Forced wordplay
Comparing trades to silicon valley as a lazy contrast joke

Humor placement rules

Never put humor in the hook (first 100 chars). Hook earns the read. Humor is the close or middle beat.
Never explain the humor. If you wrote "i'm 5 minutes out," don't add "you know what i mean."
Humor lives in ONE post per batch of 3, not all 3. If you write 3 humor drafts, you've drifted. Pick 1.

When to add humor

Skip humor entirely on:

Hank product reality posts (these need to feel honest, not winking)
Hot takes on competitors (humor reads as catty, not confident)
Stat-led posts where the number is the punch
Anything about customer money problems (their losses aren't funny)

Use humor on:

Founder lessons (self-deprecation lands)
Industry observations (dry noting works)
Behind-the-scenes building moments (everyone screws up, humanize)

Variation rule update

When generating 3 variations on a topic:

Variant A: straight observation, no humor
Variant B: tighter / different angle, no humor
Variant C: optional wry tone (pick this one only if topic supports humor per the rules above)

If the topic shouldn't have humor at all, all 3 variants stay straight. Don't force a wry version onto a serious topic.