content-engine

Content Engine

A full content-production pipeline. Given a topic, this skill produces platform-native posts — each post thinks about the topic differently. Same topic, different angle, hook, voice, structure, and format per platform. The reader following all platforms sees distinct pieces, not reformatted copies.

When to Use This Skill

Use when the user asks to:

• Turn a topic into posts for multiple platforms
• Produce publication-ready drafts across LinkedIn / X / Newsletter
• Run the repurposing chain on an idea
• Generate platform-native content (not one post reformatted)
• Get drafts ready for scheduling/publishing

Setup

At the start of every invocation, load the reference files upfront.

Always load:

• references/index.md — identity, niche, node map, execution instructions
• references/voice/brand-voice.md — core voice principles (charisma, respect, trust, never-talk-down) and anti-patterns
• references/voice/platform-tone.md — per-platform voice adaptation
• references/engine/hooks.md — 14 canonical hooks + exclusion list + respect test
• references/engine/repurpose.md — chain order, annoyance test, per-platform change table
• references/engine/scheduling.md — cadence and peak times
• references/engine/content-types.md — 6 format definitions
• references/audience/builders.md — primary audience
• references/audience/casual.md — secondary audience

Load for active platforms (v1 = LinkedIn + X + Newsletter):

• references/platforms/linkedin.md
• references/platforms/x.md
• references/platforms/newsletter.md

Do NOT load (v1 stubs; reference briefly only when outputting stub placeholders in Step 5):

• references/platforms/instagram.md · tiktok.md · youtube.md · threads.md · facebook.md

Step 1 — Topic Intake

The user provides a topic. Before proceeding:

1. Niche alignment check (per index.md Identity section). Does the topic fit "investor-grade judgment applied to AI, trading infrastructure, or market intelligence"?
   • If NO → stop and explain why the topic is off-niche. Do not produce posts.
   • If YES → proceed.
2. Skip clarifying questions for any information already explicit or strongly implied in the user's prompt. Only ask about missing load-bearing inputs (e.g., if the topic is very vague and needs a specific angle first).

Step 2 — Hook Selection

For each of the 3 active platforms, pick 2–3 candidate hooks from the approved taxonomy in references/engine/hooks.md (14 formulas, grouped into three families):

• Evidence-led: Data Lead · Proof · Credibility (Grounded) · Value / Resource Drop
• Insight-led: Framework · Non-Obvious Angle · Collision · Counter-Intuitive Mechanism · Risk-Aware Warning · Eloquence / Aphorism
• Invitation-led: Walk-Through · Honest Correction · Question · Curiosity (Open Loop)

Rules:

• Aim for a mix across platforms AND across families — ideally at least two of {Evidence, Insight, Invitation} are represented across the three drafts.
• Explicitly exclude any hook from the "Hooks Explicitly Excluded" list in hooks.md (smug, clickbait, or talking-down framings).
• Run the respect test on each candidate: "Would Peter be comfortable saying this out loud, verbatim, to a room of peers he respects?" If no, pick a different hook.
• Track which hook fits which platform for Step 3.

Step 3 — Execute Repurposing Chain (v1 = 3 platforms)

Execute in this exact order (per references/engine/repurpose.md):

Step 3a — X (first)

Forces brevity and finds the core. If the idea can't survive compression to 280 chars (or a tight 5–12 tweet thread), it isn't ready.

• Voice: references/voice/platform-tone.md → X section
• Format: per references/platforms/x.md (single tweet OR thread)
• Hook: from Step 2 selection

Step 3b — LinkedIn (second)

Different angle from X — not reformatted. Institutional framing, narrative arc, professional warmth.

• Voice: references/voice/platform-tone.md → LinkedIn section
• Format: per references/platforms/linkedin.md (1,300–2,000 char post)
• Hook: different hook family from X, from Step 2 selection

Step 3c — Newsletter (third)

Deepest version. Full reasoning + counterargument + invalidation condition. Post-mortem element if relevant.

• Voice: references/voice/platform-tone.md → Newsletter section
• Format: per references/platforms/newsletter.md (1,000–2,000 word essay with section headers)
• Hook: often Walk-Through, Honest Correction, or Collision

Step 4 — Annoyance Test (Mandatory Gate)

Reread all three drafts. Ask:

"If a reader followed Peter on all three active platforms, would they feel they're seeing substantially the same content reworded?"

• If YES → at least two of {angle, hook family, structure, depth, voice energy} didn't change enough. Return to the relevant sub-step in Step 3 and rethink.
• If NO → proceed.

Three posts are sufficiently different when at least three of these differ meaningfully:

1. Opening angle (which facet of the topic leads)
2. Hook type (from the approved taxonomy)
3. Structure (thread / long post / essay)
4. Depth (surface / walk-through / full mechanism + invalidation)
5. Voice energy (per platform-tone.md)

Step 5 — Stub Output (Phase 2 platforms)

For Instagram, TikTok, YouTube, Threads, Facebook — output a one-line placeholder per platform:

[Instagram: STUB — not active in v1. Fill when Instagram goes active.]
[TikTok: STUB — not active in v1. Fill when TikTok goes active.]
[YouTube: STUB — not active in v1. Fill when YouTube goes active.]
[Threads: STUB — not active in v1. Fill when Threads goes active.]
[Facebook: STUB — not active in v1. Fill when Facebook goes active.]

Do not attempt full drafts for stub platforms. Doing so violates the design of v1.

Step 6 — Scheduling Recommendation

For each of the 3 active-platform drafts, include:

• Recommended posting day (per references/engine/scheduling.md)
• Recommended posting time window
• If the drafts belong to the same topic run, note a staggered schedule (default: X Tue 8am → LinkedIn Wed 8am → Newsletter Thu 9am)

Step 7 — Iteration Prompt

End the run by inviting Peter to refine the system:

1. Which hooks/angles feel strongest to Peter for this topic?
2. Anything new to add to hooks.md based on what worked?
3. Any voice calibration to refine in platform-tone.md?
4. Any new audience signals to update audience/*.md?

This is the engine's compound-interest effect for content — it gets smarter each week Peter iterates.

Quality Checklist

Before presenting output, verify:

• Topic passes niche alignment check (per index.md)
• Each of the 3 active drafts names the hook type used (from the approved 14)
• Across the 3 drafts, at least 2 of 3 hook families (Evidence / Insight / Invitation) are represented
• No excluded hooks used (check against hooks.md exclusion list)
• All 3 drafts pass the respect test (no talking down, no smug framing, no lecturing)
• All 3 drafts pass the annoyance test (angle/hook/structure/depth/voice all meaningfully differ)
• Peter's clarity gate holds (smart 13-year-old can follow the language; investors still find it sharp)
• Peter's non-negotiables applied: hook first, fact before hype, no fluff, never condescending
• Each draft has at least one counterargument or invalidation condition (non-negotiable for Newsletter; strongly preferred for LinkedIn)
• Specific numbers with units and timeframes where data is cited
• 5 stub placeholders present for Phase 2 platforms
• Scheduling recommendation included per active-platform draft

Output Format

Present output in this order:

1. Niche alignment confirmation (one line: "Topic aligns with [pillar(s)].")
2. Hook candidates per platform (2–3 per platform with hook type labeled)
3. X draft (hook type + full draft + recommended posting day/time)
4. LinkedIn draft (hook type + full draft + recommended posting day/time)
5. Newsletter draft (hook type + full draft with section headers + recommended send day/time)
6. Stub placeholders (5 one-liners)
7. Annoyance + respect test confirmation (one line: "Both tests passed — drafts are platform-native.")
8. Iteration prompt (4 questions for Peter)