1. Memory Check: Silently read the local files vault/brand/brand-voice.md and vault/brand/owner-profile.md from the local file system to recall your consolidated core memory, brand positioning, and absorbed feedback. (Do NOT use mcp.lark.drive or mcp.lark.docs to read these; read the local copies directly. Do NOT read the raw ownerreview.md log — rely on what has been distilled into brand-voice.md). 2. Asset Check: Silently check vault/media/media-index.md for any existing approved assets that match the current content topic. Note which assets are available before drafting. 3. Audience Check: Read vault/brand/audience-profile.md. Note the current audience layer distribution, new-vs-returning ratio, and platform-specific audience defaults. Decide which audience layer this content should target. 4. Winner Library Check: Read vault/analytics/winner-library.md. Find any activated patterns matching the current platform(s) and product category. Note the recommended hook_type(s). 5. Cross-Skill Check: Scan your active OpenClaw tools (e.g., mcp.bash for AMC Kanban publish calls) and any installed external Skill Packs (e.g., marketing, video-creation, copywriting). 6. Decision Output: At the very top of your response, BEFORE any platform drafts, output your memory and skill routing:

[Memory Loaded] I noted recent feedback: [Brief Summary]. I will apply it here. [Assets Available] Matching assets: [filenames or 'none found']. [Audience Target] Layer: [Discoverer|Considerer|Regular|Advocate] | New/Return ratio this week: [X%/X%] | Recent signal: [saves|shares|comments]. [Winner Pattern] Recommended hook for this topic: [hook_type] — based on [N] winning posts. [Pillar] This content belongs to: [Pillar Name] | Week ratio: [actual%] → target [target%] | Gap pillars this week: [pillar names or 'none']. [Skills Activated] For this task, I will additionally invoke: [Skill Name 1], [Skill Name 2].

Step 1 · Instagram (write FIRST — visual hook)

Write Instagram first. This forces you to find the visual story before anything else.

Platform DNA: Aspirational, polished, aesthetic-driven. Audience scrolls fast — you have 0.5 seconds to stop them.

DO NOT:

Start with "Introducing..." or "We are proud to present..."
Use generic filler like "Delicious food awaits!" or "Come and try!"
Write more than 125 characters before the first line break
Use more than 8 hashtags

Output format:

[HOOK — ≤125 chars, must be scroll-stopping. Options: bold claim, surprising fact, sensory description, or question]

[1-2 lines of blank space]

[Body — 2-4 short paragraphs. Sensory details: colour, texture, smell, temperature. Tell a micro-story.]

[Soft CTA — one line, conversational. "Link in bio to order." / "DM us to book your table."]

[Hashtags — 5-8 only. Mix: 2 broad (#chinesefood), 2 niche (#[city]eats), 1-2 brand-specific]

Content format: 7-10 slide carousel (preferred) OR single hero image + caption

📷 After drafting this platform's content, immediately output a Visual Brief following [[core/visual-brief]] — Instagram format (1:1 or 4:5, carousel specs).

Step 2 · TikTok (raw, native, movement-first)

Do NOT adapt the Instagram caption. Start from scratch with a video-native brain.

Platform DNA: Raw > polished. Authentic > branded. Fast-paced, sound-on, native editing style. If it looks like an ad, it loses.

DO NOT:

Write a script that starts with the brand name
Use formal or marketing language ("premium quality", "authentic experience")
Plan a shot longer than 3 seconds without a new angle
Include text overlays that cover the food

Output format:

HOOK (0:00–0:02): [One-word or one-action hook — sizzle sound, sauce pour, steam rising, crunch reveal]

BUILD (0:02–0:15): [The "how" or "wow" — process, transformation, or unexpected detail]

PEAK (0:15–0:22): [The money shot — final dish, portion reveal, or reaction]

CTA (0:22–0:30): [Price on screen + one-line text: "Order at [link]" or "DM us 🔥"]

Text overlays: [List any captions to overlay — keep to ≤5 words each]
Sound direction: [Trending audio / natural kitchen sounds / voiceover — specify]

📷 After drafting this platform's content, immediately output a Visual Brief following [[core/visual-brief]] — TikTok format (9:16 video, include Shot List).

Step 3 · 小红书 RedNote (personal diary, not marketing copy)

Platform DNA: 小红书 users trust peer reviews, not brand accounts. Write as if a real person is sharing their genuine experience — not as a brand promoting itself.

DO NOT:

Write in third person ("The restaurant offers...")
Use promotional superlatives ("best", "most authentic", "must-try")
Ignore emoji — they are native to the platform
Forget to include prices (users expect them)

Output format:

标题 Title (≤20 chars, emoji optional): [Hook — question, bold claim, or personal confession]

正文 Body:
第一段 Para 1: [Personal hook — "上周去吃了这家..."] 
第二段 Para 2: [Sensory experience — taste, texture, portion size, value]
第三段 Para 3: [Honest opinion — what's great, what could be better]
第四段 Para 4: [Practical info — price per dish, address, hours]

话题标签 Tags (3-5): [Mix of broad 美食 and local 城市美食]

Length: 300–500 characters (not words). Tight and scannable.

📷 After drafting this platform's content, immediately output a Visual Brief following [[core/visual-brief]] — 小红书 format (3:4 carousel, cover must have large readable title).

Step 4 · Facebook (community conversation, not broadcast)

Platform DNA: Facebook feeds community, not trend. Users are older, more conversational, and respond to nostalgia, belonging, and discussion questions.

DO NOT:

Copy-paste the Instagram caption
Use hashtags heavily (2 max on Facebook)
Post without a conversation hook at the end

Output format:

[Opening — 1 sentence that feels like you're talking to a friend, not an audience]

[Body — 3-5 sentences. Context, story, or memory trigger. More text is OK here.]

[Question CTA — must invite a real response. "What's your go-to order?" / "Tag the person you'd bring here 👇"]

[Optional: Tag a local community group or page if relevant]

Hashtags: 1–2 max

📷 Visual: Reuse the Instagram carousel asset (note which files). Only output a new Visual Brief if Facebook needs a different crop or format.

Step 5 · YouTube (deep storytelling — IF APPLICABLE)

Only create YouTube content if the idea genuinely benefits from longer treatment. Skip if it doesn't.

Qualifying topics: recipes, kitchen tours, chef stories, dish origin stories, supplier visits, "how it's made" Skip for: deal alerts, lunch specials, single-dish promos

Platform DNA: YouTube is search-driven. Title and thumbnail are 80% of the battle. Write for someone searching "best [dish] in [city]" — not for existing fans.

Output format:

Title (SEO): [Dish name + location keyword + hook — e.g., "How We Make Authentic Dan Dan Noodles | [City] Chinese Food"]
Thumbnail text: [3–5 bold words max]

Outline:
[0:00–0:30] Hook — what's the payoff? Show the finished dish first.
[0:30–2:00] Context — brand story or dish origin in ≤90 seconds
[2:00–6:00] Core content — process, interview, or deep-dive
[6:00–7:30] Taste/reaction moment
[7:30–8:00] CTA — subscribe, order link, next video

Description (first 150 chars are visible): [SEO-rich, keyword-dense first sentence]
Tags: [10–15 specific search terms]

📷 After drafting this platform's content, immediately output a Visual Brief following [[core/visual-brief]] — YouTube format (16:9 video + thumbnail brief with text ≤5 words).

Step 6 · X / Twitter (sharp take, not promotion)

Only post to X if there's a contrarian, witty, or industry angle worth a sharp line. Skip purely promotional content entirely.

Platform DNA: X rewards opinions, not announcements. If it sounds like a press release, delete it.

DO NOT:

Post "We're excited to announce..."
Restate what's already on Instagram
Use more than 2 hashtags

Qualifying angles: Industry observation / hot take / behind-the-scenes fact / price transparency / cultural commentary

Output format:

Option A — Single tweet (≤280 chars):
[Sharp observation or hot take — state it plain, no fluff]

Option B — Thread (if the angle needs depth):
1/ [Hook tweet — the thesis, ≤280 chars]
2/ [Evidence or story]
3/ [Counter-argument or nuance]
4/ [Conclusion + CTA]

📷 Visual: Reuse existing assets. If none available, a branded text-on-image card is sufficient — no new shoot needed.

Step 7 · Google Maps (reputation layer — implicit)

Not part of the content publish chain. After every piece of content, ask:

"Does this content highlight something worth a Google review?"

Surface those moments naturally. Never explicitly ask for reviews in exchange for anything (FTC + Google Terms violation).

Gates (run on EVERY platform draft, in this order)

1. Compliance Gate → [[compliance/fda-ftc-rules]] + [[compliance/allergen-gate]]
   RED hit  → block, create owner-approval entry, do NOT proceed
   YELLOW   → flag, hold for owner sign-off

2. Bilingual Gate → [[localization/bilingual-gate]]
   Confirm language per platform decision tree
   Ensure canonical dish names on first mention

3. Owner Gate → [[operations/owner-approval]]
   Only triggers on flagged drafts (Tier 1 or 2 conditions)

4. De-AI Gate → [[compliance/de-ai-gate]]
   Scrub out "AI-isms", robotic phrasing, and symmetrical templates.
   Silently rewrite any offending sentences to be colloquial and human-like before final output.

The Litmus Test

Before finalizing any set of platform versions, ask:

"If someone followed us on ALL platforms, would they be annoyed seeing the same thing?"

Yes → you're reformatting, not rethinking. Go back to Step 1.

No → you've made unique pieces from one idea. Ship.

Batch Workflow (recommended)

1. Pick 2-3 menu items or themes for the week
2. For each idea, run Steps 1-6 in ONE session (not spread across days)
3. Pass each draft through both Gates BEFORE moving to the next step
4. Schedule via [[core/scheduling]] to stagger across the week
5. Track performance per platform; adjust the chain weekly

Weekly batch target: 15-25 posts covering all active platforms for the week.