---

name: linkedin-write description: Write a LinkedIn post in Anuj's voice — sharp, contrarian-but-grounded, first-person, no engagement-bait. Asks length intent before drafting and returns post body plus 2 alternate hooks.

linkedin-write

Drafting a LinkedIn post for Anuj Sadani. LinkedIn is one of Anuj's two sharp/contrarian-edge formats (the other is Substack). Lead with a punchy hook, deliver one idea, close clean.

Step 1 — Load context (mandatory)

Read these files:

1. shared/voice-rules.md
2. shared/voice-samples.md
3. shared/pet-peeves.md
4. shared/topic-modes.md
5. shared/platform-styles.md
6. references/hook-formulas.md

Step 2 — Ask length intent

Ask the user ONCE:

"Length? short (<500 chars, single observation+implication), medium (~1300 chars, hook+2-3 paras+payoff), or long (~2000+ chars, mini-essay with sections)?"

Wait for the answer. Default to medium if no response.

Step 3 — Classify topic mode + apply platform style profile

Apply topic-modes.md. If security → forensic voice. If architecture/agent → sharp technical opinion. If leadership/governance → measured but still owned.

Then apply the linkedin row from shared/platform-styles.md: technical depth 2 (concept over code, at most one config line), headline aggressiveness 4 (scroll-stopping first line, never engagement-bait), density 2 (white space, short paragraphs). Honor the precedence rules in that file — voice-rules.md/pet-peeves.md are absolute, topic mode can raise (never lower) substance, and the style profile sets the per-platform baseline within those.

Step 4 — Pick a hook pattern

Choose ONE pattern from references/hook-formulas.md that fits the topic. Do not name the pattern in the output; just use its shape.

Step 5 — Draft the post

Constraints:

• First person. "I" — owned opinions.
• One idea per post. Resist listing 5 takeaways.
• Scannable spacing. Short paragraphs (1-3 lines), single blank lines between. No paragraph longer than 4 lines on mobile.
• No more than 3 hashtags, and only if they are topically meaningful (#AIAgents is fine; #TechLife is not).
• Emoji: follow the Emoji Control section in shared/voice-rules.md. Default level is low — most LinkedIn posts should have zero emoji. If the user says "use medium emoji" / "high emoji" / "no emoji" override accordingly. Per the per-format ceiling: even at medium, no emoji as bullet leads.
• Closer: crisp takeaway OR one genuine open question. Never engagement-bait ("What do you think?", "Drop a comment", "Hit follow").

Length targets:

• short: ≤500 characters total. Hook + one observation + one implication.
• medium: 1100-1400 characters. Hook + 2-3 short paragraphs + payoff line.
• long: 1800-2400 characters. Hook + 4-6 short paragraphs, optionally with > quote lines or 2-3 short numbered bullets, + closer.

Step 6 — Pre-delivery checks

1. Run the regex sweep from shared/pet-peeves.md. Any hit → rewrite.
2. Confirm character count matches the requested length tier.
3. Confirm: no em-dashes joining clauses, no rule-of-three, no engagement-bait CTA, first-person.
4. Confirm: style profile applied — depth, headline, and density match the linkedin row in shared/platform-styles.md (topic-mode overrides may raise depth).

Step 7 — Save and deliver

Save to file. Derive a kebab-case topic slug from the user's topic (e.g., "Why agentic systems fail in production" → agentic-systems-prod-failures). Get today's date in YYYY-MM-DD form. Save the post body to:

./drafts/<YYYY-MM-DD>-<topic-slug>/linkedin.md

If the file already exists (re-running on the same topic same day), suffix with -v2, -v3, etc. Create directories as needed.

Also print in chat in this shape:

─── POST ───
<post body>

─── ALTERNATE HOOK 1 ───
<single hook line that could replace the opener>

─── ALTERNATE HOOK 2 ───
<a different angle hook line>

─── META ───
characters: <N>
hashtags: <list or "none">
hook pattern used: <e.g., dichotomy / demoting / pragmatic-flex / observation-flip>
emoji level: <none | low | medium | high>
saved to: <path>

Do not post anywhere. The file is for review and manual copy into LinkedIn's compose window.