diff --git a/content/instagram/2026-07-06-collection-gap/caption.txt b/content/instagram/2026-07-06-collection-gap/caption.txt
new file mode 100644
index 000000000..b26044e5e
--- /dev/null
+++ b/content/instagram/2026-07-06-collection-gap/caption.txt
@@ -0,0 +1,9 @@
+Your business isn't unprofitable. It's unpaid.
+
+59% of small businesses are getting paid late right now. The average one is owed $17,500. Meanwhile the follow-up is a polite email and an awkward call. Whoever shouts softest gets paid last.
+
+We call it the collection gap. The work is done. The cash is missing.
+
+comment the word "cash" and I'll send you the Get Paid Checklist (free PDF): day-0 invoicing, automatic 3-7-14 reminders, one live ledger.
+
+#smallbusiness #cashflow #invoicing #accountsreceivable #operations #automation #businessowner
\ No newline at end of file
diff --git a/content/instagram/2026-07-06-collection-gap/resource_brief.md b/content/instagram/2026-07-06-collection-gap/resource_brief.md
new file mode 100644
index 000000000..38462e8a4
--- /dev/null
+++ b/content/instagram/2026-07-06-collection-gap/resource_brief.md
@@ -0,0 +1,13 @@
+# Lead magnet — "Video 7 - Comment CASH - Get Paid Checklist.pdf"
+
+3 pages, Arqentia Instrument document system (per the V5/V6 resource series).
+
+1. **Cover** — COMMENT CASH pill; title "Get Paid Checklist"; flow INVOICE → REMIND → LEDGER →
+   PAID; core question "How much of your revenue is sitting in unpaid invoices right now?";
+   pass standard "Invoices go out day 0; reminders run themselves; one ledger shows who owes what."
+2. **Step 1 · Find the gap** — checklist: invoices sent days after work closes / follow-up lives in
+   someone's head / no standard reminder cadence / awkward calls instead of process / discounts
+   given just to get paid / nobody knows total outstanding today.
+3. **Step 2 · Close it** — checklist: invoice the day work closes / automatic reminders day 3-7-14 /
+   escalation step at day 21 (call script) / late fee policy stated on every invoice / one live AR
+   ledger reviewed weekly / stop work triggers for chronic late payers.
diff --git a/content/instagram/2026-07-06-collection-gap/script.md b/content/instagram/2026-07-06-collection-gap/script.md
new file mode 100644
index 000000000..bb2eca4de
--- /dev/null
+++ b/content/instagram/2026-07-06-collection-gap/script.md
@@ -0,0 +1,39 @@
+# Arqentia — "The Collection Gap" (Video 7) — keyword: cash
+
+**Format:** 9:16 IG Reel ~50s. Remotion atelier + stock B-roll (per arqentia-video5-video6-build).
+**Grounding:** Bluevine SMB study — 59% of small businesses experience late payments; average SMB
+owed $17,500 (https://www.bluevine.com/blog/small-business-late-payment-gaps); QuickBooks Jan-2026
+Insights: 42% report cash-flow problems.
+
+## Spoken VO (HeyGen — VO only)
+
+Your business isn't unprofitable. It's unpaid.
+
+Look at your receivables. 59 percent of small businesses are getting paid late right now. The average one is owed seventeen thousand five hundred dollars.
+
+Meanwhile you chase invoices by hand. A polite email here. An awkward call there. Whoever shouts softest gets paid last.
+
+We call it the collection gap. The work is done. The cash is missing.
+
+So we don't start with software. We audit the operation first. Invoice the day the work closes. Automatic reminders on day three, seven, and fourteen. One ledger that shows who owes what, at all times.
+
+You did the work. The system should collect it.
+
+comment the word cash, and I'll send you the get paid checklist.
+
+## Beat map (9 beats)
+
+| # | Beat | Line | Visual |
+|---|------|------|--------|
+| 1 | Hook title | "isn't unprofitable. It's unpaid." | title over avatar: "Not unprofitable. / Just unpaid." em unpaid. |
+| 2 | Exhibit 01 | "Look at your receivables" | stock: ledger/bookkeeping office (graded) |
+| 3 | Stats | "59 percent … seventeen thousand five hundred" | MG stat cards 59% → $17,500 |
+| 4 | Stamps | "polite email / awkward call / paid last" | word stamps over avatar |
+| 5 | Name it | "the collection gap" | MG typographic moment + sub |
+| 6 | Method | "audit first… invoice day-0… reminders 3·7·14… one ledger" | MG convergence: INVOICE/REMIND/LEDGER → COLLECTS ITSELF |
+| 7 | Exhibit 02 | "You did the work. The system should collect it." | stock: cash counting / register (graded) + device seal |
+| 8 | CTA | "comment the word cash" | COMMENT [CASH] pill |
+| 9 | End card | — | Arqentia. wordmark |
+
+Signature device: **the aging bar** — receivables aging (days) climbing during problem beats,
+snapping to "day 0 · collected" at resolution. (Meter family; distinct labels/motion from V4-V6.)
diff --git a/content/instagram/2026-07-07-dead-stock/caption.txt b/content/instagram/2026-07-07-dead-stock/caption.txt
new file mode 100644
index 000000000..d45fadfcc
--- /dev/null
+++ b/content/instagram/2026-07-07-dead-stock/caption.txt
@@ -0,0 +1,9 @@
+Your most expensive product is the one that never sells.
+
+Every shelf of slow movers is cash you already spent, frozen. You reordered it because a spreadsheet said you were low. The spreadsheet was wrong. Retailers lose $1.7 trillion a year to stock they bought wrong.
+
+We call it dead capital. Money you can see but cannot spend.
+
+comment the word "stock" and I'll send you the Dead Capital Checklist (free PDF): track what moves, flag slow movers the week they stall, one source of truth for every unit on hand.
+
+#smallbusiness #inventory #cashflow #operations #ecommerce #wholesale #businessowner
\ No newline at end of file
diff --git a/content/instagram/2026-07-07-dead-stock/resource_brief.md b/content/instagram/2026-07-07-dead-stock/resource_brief.md
new file mode 100644
index 000000000..b89c4e6ea
--- /dev/null
+++ b/content/instagram/2026-07-07-dead-stock/resource_brief.md
@@ -0,0 +1,16 @@
+# Lead magnet — "Video 8 - Comment STOCK - Dead Capital Checklist.pdf"
+
+3 pages, Arqentia Instrument document system (V5/V6/V7 resource series).
+
+1. **Cover** — COMMENT STOCK pill; title "Dead Capital Checklist"; flow TRACK → FLAG → LEDGER →
+   MOVES; core question "How much of your cash is frozen on shelves that will never sell?";
+   pass standard "You know what moves, slow movers get flagged the week they stall, one source of
+   truth for every unit on hand."
+2. **Step 1 · Find the frozen cash** — checklist: reorders driven by a spreadsheet, not real
+   sell-through / no flag when an SKU stops moving / stock counts differ from the system / slow
+   movers discovered at year-end, not weekly / cash-value of dead stock unknown today / buying to
+   hit supplier minimums, not demand / no owner for inventory accuracy.
+3. **Step 2 · Thaw it** — checklist: track real sell-through per SKU / auto-flag SKUs with zero
+   movement in 30/60/90 days / one live inventory ledger reconciled weekly / reorder points set
+   from demand, not gut / a clear the-deck rule for aged stock (markdown/bundle/return) / a named
+   owner for inventory data.
diff --git a/content/instagram/2026-07-07-dead-stock/script.md b/content/instagram/2026-07-07-dead-stock/script.md
new file mode 100644
index 000000000..1eb101ab2
--- /dev/null
+++ b/content/instagram/2026-07-07-dead-stock/script.md
@@ -0,0 +1,39 @@
+# Arqentia — "Dead Capital" (Video 8) — keyword: stock
+
+**Format:** 9:16 IG Reel ~50s. Remotion atelier + B-roll (avatar ≤20% per 2026-07-06 rule: hook + CTA
+only; exhibits/MG carry the rest). Grounding: retailers lose $1.73T/yr to overstocks + out-of-stocks
+(Shopify/IHL retail inventory-distortion research); dead stock commonly ties up 20-30% of SMB inventory
+value.
+
+## Spoken VO (HeyGen — VO only)
+
+Your most expensive product never sells.
+
+Walk your warehouse. Every shelf of slow movers is cash you already spent, frozen. You reordered it because a spreadsheet said you were low. The spreadsheet was wrong.
+
+Retailers lose one point seven trillion dollars a year to stock they bought wrong. Too much of the wrong thing. Too little of the right thing.
+
+The overstock is the part you can see. We call it dead capital. Money you can see but cannot spend.
+
+So we don't start with software. We audit the operation first. We track what actually moves. We flag slow movers the week they stall. Then one ledger, every unit on hand.
+
+Your shelves should hold product. Not frozen cash.
+
+comment the word stock, and I'll send you the dead capital checklist.
+
+## Beat map (avatar ≤20%: only hook ~3s + CTA ~3s)
+
+| # | Beat | Line | Visual |
+|---|------|------|--------|
+| 1 | Hook | "most expensive product never sells" | avatar + title "Your most expensive product / isn't selling." |
+| 2 | Exhibit 01 | "walk your warehouse… cash frozen" | B-roll: dim warehouse racks / pallets (graded) |
+| 3 | Exhibit 02 | "reordered it because a spreadsheet said" | B-roll: spreadsheet / reorder screen macro (graded) |
+| 4 | Stats | "$1.7 trillion… too much wrong / too little right" | MG stat card $1.7T + two-line contrast |
+| 5 | Name it | "dead capital" | MG typographic moment + sub |
+| 6 | Method | "audit first… track what moves… flag stalls… one source of truth" | MG convergence TRACK/FLAG/ONE LEDGER → MOVES ITSELF |
+| 7 | Exhibit 03 | "shelves should hold product, not frozen cash" | B-roll: goods moving / forklift flow (graded) + device seal |
+| 8 | CTA | "comment the word stock" | avatar + COMMENT [STOCK] pill |
+| 9 | End card | — | Arqentia. wordmark |
+
+Signature device: **the freeze gauge** — % of inventory frozen climbing during problem beats, snapping
+to "moving" at resolution. Distinct labels/motion from the V4-V7 devices.
diff --git a/content/instagram/2026-07-08-swivel-chair/caption.txt b/content/instagram/2026-07-08-swivel-chair/caption.txt
new file mode 100644
index 000000000..821e9f349
--- /dev/null
+++ b/content/instagram/2026-07-08-swivel-chair/caption.txt
@@ -0,0 +1,11 @@
+Your team isn't slow. Your systems just don't talk to each other.
+
+One order comes in. Someone keys it into accounting. Then inventory. Then the CRM. Same data, three times, by hand.
+
+The average employee burns 9+ hours a week just moving data between systems. That's more than $28,500 per person, every year, on typing a computer should do. And half of it comes back wrong (Parseur, 2025).
+
+We call it swivel-chair work. We don't fix it with more software. We audit the operation first, find every place a number gets entered twice, then connect your systems so it flows once. Entered one time. Everywhere it needs to be.
+
+Comment the word retype and we'll send you the Swivel-Chair Audit: the checklist we use to find every double-entry in your operation.
+
+#smallbusinessoperations #businessautomation #dataentry #systemsintegration #operationsmanagement #smbtech #workflowautomation
diff --git a/content/instagram/2026-07-08-swivel-chair/resource_brief.md b/content/instagram/2026-07-08-swivel-chair/resource_brief.md
new file mode 100644
index 000000000..c8f171398
--- /dev/null
+++ b/content/instagram/2026-07-08-swivel-chair/resource_brief.md
@@ -0,0 +1,34 @@
+# Lead-magnet PDF — "The Swivel-Chair Audit"
+
+**Comment keyword:** RETYPE
+**Promise:** the checklist Arqentia uses to find every place your team re-enters the same data — and what
+it's costing you.
+**Format:** 3 content pages, rs/Arqentia document system (ink cover + paper content + ink back), 1080×1350.
+
+## Page 1 — Cover (ink)
+- Title: **The Swivel-Chair Audit**
+- Subtitle: Find every place your team types the same data twice.
+- `COMMENT "RETYPE"` pill + Arqentia wordmark.
+- Anchor stat strip: "$28,500 / employee / year · 9+ hrs/week moving data · 50% comes back wrong" (Parseur 2025).
+
+## Page 2 — The 12-point double-entry sweep (paper)
+A checklist of the usual re-entry hotspots to walk your operation against. Each = one line + a check box:
+1. Orders retyped from email/PDF into the ERP or accounting system
+2. The same order re-keyed into inventory / stock counts
+3. Customer details re-entered into the CRM
+4. Invoices/receipts typed into the bookkeeping tool by hand
+5. Shipping/tracking numbers copied between carrier portal and orders
+6. Spreadsheet "bridges" someone maintains between two systems
+7. Reporting numbers re-typed into a dashboard or board deck
+8. Employee/payroll data entered into more than one HR tool
+9. Quotes rebuilt from scratch instead of pulled from a record
+10. Supplier data re-keyed per PO
+11. Returns/RMAs logged in a separate sheet from the system of record
+12. Any field a person reads off one screen and types into another
+> Scoring: count your checks. 3+ = a real integration ROI. 6+ = you are paying a full salary in retyping.
+
+## Page 3 — Enter it once: the fix (ink back)
+- The Arqentia method in four steps: **Audit the operation → Map the double-entry → Connect the systems →
+  Enter it once (source of truth + automated flow).**
+- One-line proof: "9 hours a week back per employee is more than a full workday returned."
+- CTA: book an operations audit + Arqentia wordmark / arqentia.com.
diff --git a/content/instagram/2026-07-08-swivel-chair/script.md b/content/instagram/2026-07-08-swivel-chair/script.md
new file mode 100644
index 000000000..2c114fe25
--- /dev/null
+++ b/content/instagram/2026-07-08-swivel-chair/script.md
@@ -0,0 +1,47 @@
+# Arqentia — "Swivel-Chair Work" — keyword: retype
+
+**Format:** 9:16 IG Reel ~50s. Remotion atelier + stock B-roll (avatar ≤20% per 2026-07-06 rule: hook +
+CTA only; exhibits/MG carry the rest). Grounding: U.S. employees spend 9+ hours a week manually moving
+data between systems; manual data entry costs an average of **$28,500 per employee per year**, and 50.4%
+of respondents hit errors/delays from it (Parseur 2025 survey, parseur.com/blog/manual-data-entry-report).
+Distinct from the "handoff tax" (that = cross-team boundaries; this = ONE person re-keying the SAME datum
+into disconnected systems).
+
+## Spoken VO (HeyGen — VO only)
+
+Your team isn't slow. Your systems don't talk to each other.
+
+Watch what happens when one order comes in. Someone types it into accounting. Then the same numbers into inventory. Then again into the CRM. Same data. Three times. By hand.
+
+The average employee spends over nine hours a week moving data between systems. That's more than twenty eight thousand dollars a year, per person, spent retyping what a computer already knows. And half of it comes back wrong.
+
+We call it swivel chair work. You spin from one screen to the next, retyping what you already know.
+
+So we don't start with software. We audit the operation first. We find every place a number gets entered twice. Then we connect your systems so it flows once.
+
+Comment the word retype. I'll send you the swivel chair audit.
+
+## Beat map (avatar ≤20%: only hook ~3s + CTA ~3s)
+
+| # | Beat | Line | Visual |
+|---|------|------|--------|
+| 1 | Hook | "team isn't slow… systems don't talk" | avatar + title "It's not your team. / It's your systems." |
+| 2 | Exhibit 01 | "one order comes in… types it into accounting" | B-roll: office data entry, person at keyboard (graded) |
+| 3 | Exhibit 02 | "same numbers into inventory… again into CRM… three times" | B-roll: hands typing / multiple monitors / form fields (graded) |
+| 4 | Stats | "nine hours a week… $28,500 a year… half comes back wrong" | MG stat card $28,500 + "9 hrs/week" + "50% error" support |
+| 5 | Name it | "swivel chair work" | MG typographic moment + sub "you spin screen to screen, retyping what you already know" |
+| 6 | Method | "audit first… find every place entered twice… connect so it flows once" | MG method: AUDIT → MAP DOUBLE-ENTRY → CONNECT → FLOWS ONCE |
+| 7 | Exhibit 03 | "so it flows once" (resolution) | B-roll: clean automated data flow / single dashboard (graded) + device snap |
+| 8 | CTA | "comment the word retype" | avatar + COMMENT [RETYPE] pill |
+| 9 | End card | — | Arqentia. wordmark |
+
+Signature device: **the re-entry counter** — a small HUD stamping the SAME record ×1 → ×2 → ×3 as it is
+keyed into each system during the problem beats, then snapping to "entered once · ×1" at the fix. Distinct
+labels/motion from the V4–V8 devices (freeze gauge, showWindows, throughput line, etc.); vary the MG scene
+grammar per the anti-template rule (do not clone the dead-stock ledger).
+
+## TTS-safety notes
+- No em-dashes anywhere (periods only). Numbers spelled where spoken ("nine hours", "twenty eight thousand
+  dollars"); stat CARD shows exact "$28,500".
+- Brand name never spoken in VO. CTA is one short sentence, lowercase "comment the word retype".
+- Keyword RETYPE reads cleanly as "ree-type" (no homophone risk). Concept "swivel chair" ≠ keyword, by design.
diff --git a/content/instagram/topics_log.md b/content/instagram/topics_log.md
new file mode 100644
index 000000000..fc4458c8d
--- /dev/null
+++ b/content/instagram/topics_log.md
@@ -0,0 +1,16 @@
+# Arqentia Instagram topics log
+
+Source of truth for what's been scripted/produced/posted. The live IG grid outranks Zernio history
+(Rafael posts natively too — never draft content already on the grid).
+
+| date | topic | keyword | status |
+|------|-------|---------|--------|
+| 2026-06-16 | AI needs a workflow | WORKFLOW | posted |
+| 2026-06-16 | Mapping workflow leaks | LEAK | posted |
+| 2026-06-17 | AI traceability / black box risk | TRACE | posted |
+| 2026-07-01 | The silent leak (shrinkage/downtime found late) | LEAK | posted |
+| 2026-07-02 | The one-person spreadsheet / single point of failure | EXCEL | posted |
+| 2026-07-02 | The handoff tax | HANDOFF | posted |
+| 2026-07-06 | The collection gap (late invoices / AR) | CASH | produced · awaiting review (Drive 7/6/26; zernio draft deleted per review-first directive) |
+| 2026-07-07 | Dead stock / dead capital (frozen inventory) | STOCK | produced · awaiting review (Drive 7/7/26) |
+| 2026-07-08 | Manual data re-entry / swivel-chair work | RETYPE | produced · awaiting review (Drive 7/8/26) |
diff --git a/content/tiktok/2026-07-06-forward-deploy/caption.txt b/content/tiktok/2026-07-06-forward-deploy/caption.txt
new file mode 100644
index 000000000..063630212
--- /dev/null
+++ b/content/tiktok/2026-07-06-forward-deploy/caption.txt
@@ -0,0 +1,9 @@
+Amazon just spent $1,000,000,000 admitting something: AI does not deploy itself.
+
+AWS is embedding thousands of engineers inside customer companies to build agents with them. Not selling tools. Shipping outcomes. The model was never the bottleneck. Deployment is.
+
+Steal the play: one workflow, one builder inside it for two weeks, shipped against a real metric.
+
+comment "deploy" and I'll send you The Deployment Brief (free PDF).
+
+#ai #aws #aiagents #automation #buildinpublic #operator #deployment
diff --git a/content/tiktok/2026-07-06-forward-deploy/resource_brief.md b/content/tiktok/2026-07-06-forward-deploy/resource_brief.md
new file mode 100644
index 000000000..70c444302
--- /dev/null
+++ b/content/tiktok/2026-07-06-forward-deploy/resource_brief.md
@@ -0,0 +1,7 @@
+# rs - Forward Deploy - Comment DEPLOY - The Deployment Brief.pdf
+4 pages, rs_ document system. 1) ink cover: "The Deployment Brief" + COMMENT "DEPLOY" pill +
+"run an AWS-style forward deployment in your own business, in two weeks". 2) paper §01 "pick the
+beachhead": one workflow, expensive + repetitive + measurable; write its current cost; name one
+owner; define the single metric. 3) paper §02 "the two-week embed": week 1 shadow + map + build
+smallest agent; week 2 run supervised, fix daily, measure; ship or kill on day 14; document what
+it touches. 4) ink back: "tools don't transform companies. deployment does." + @rafa_flows.
diff --git a/content/tiktok/2026-07-06-forward-deploy/script.md b/content/tiktok/2026-07-06-forward-deploy/script.md
new file mode 100644
index 000000000..fefecae5b
--- /dev/null
+++ b/content/tiktok/2026-07-06-forward-deploy/script.md
@@ -0,0 +1,9 @@
+# rs_ — "Forward Deploy" (TikTok) — keyword: deploy
+
+~33s full-frame rs_ cut. Trend: AWS $1B Forward Deployed Engineering org (aboutamazon.com,
+2026-06-30) — thousands of engineers embedded with customers to co-build agentic AI.
+Hook "$1B. ONE ADMISSION." → cutaway (engineers at work) on "embedding thousands of engineers" →
+chips NOT SELLING TOOLS / SHIPPING OUTCOMES → stamp "// the bottleneck is deployment" →
+numbered chips ONE WORKFLOW / ONE BUILDER · TWO WEEKS / A REAL METRIC →
+thesis "tools don't transform companies. / deployment does."
+Resource: The Deployment Brief (4pp). CTA: comment "deploy".
diff --git a/content/tiktok/2026-07-07-context-engineering/caption.txt b/content/tiktok/2026-07-07-context-engineering/caption.txt
new file mode 100644
index 000000000..17a8b3b0b
--- /dev/null
+++ b/content/tiktok/2026-07-07-context-engineering/caption.txt
@@ -0,0 +1,9 @@
+Prompt engineering is already dead.
+
+The people shipping real agents stopped tuning clever prompts months ago. The model is not the variable anymore. What you feed it is: the right files, memory, tools, and limits. Give a smaller model the right context and it beats a genius flying blind.
+
+They call it context engineering. It is the actual job now.
+
+comment the word "context" and I'll send you The Context Stack Checklist (free PDF): task, files, memory, tools, and hard limits, every run.
+
+#ai #aiagents #contextengineering #promptengineering #buildinpublic #automation #operator
\ No newline at end of file
diff --git a/content/tiktok/2026-07-07-context-engineering/resource_brief.md b/content/tiktok/2026-07-07-context-engineering/resource_brief.md
new file mode 100644
index 000000000..109365313
--- /dev/null
+++ b/content/tiktok/2026-07-07-context-engineering/resource_brief.md
@@ -0,0 +1,12 @@
+# rs - Context Engineering - Comment CONTEXT - The Context Stack Checklist.pdf
+4 pages, rs_ document system (ink cover + paper content + ink back).
+
+1. **Ink cover** — title "The Context Stack Checklist" + COMMENT "CONTEXT" pill + "stop polishing
+   prompts. build the context an agent actually needs, every run."
+2. **Paper §01 · the five layers** — TASK (one job, one output format, done-definition) / FILES
+   (the exact docs, not the whole drive) / MEMORY (what happened before that matters now) / TOOLS
+   (only what this task needs, scoped) / LIMITS (what it must never do, hard constraints).
+3. **Paper §02 · assemble it every run** — pull context fresh per task, not once / trim to what's
+   relevant (more is not better) / verify the agent can see each layer before it acts / log what
+   context produced which output / a smaller model with the right stack beats a bigger one blind.
+4. **Ink back** — "The model is not the edge. Context is." + @rafa_flows.
diff --git a/content/tiktok/2026-07-07-context-engineering/script.md b/content/tiktok/2026-07-07-context-engineering/script.md
new file mode 100644
index 000000000..96bbc8481
--- /dev/null
+++ b/content/tiktok/2026-07-07-context-engineering/script.md
@@ -0,0 +1,34 @@
+# rs_ — "Context Engineering" (TikTok) — keyword: context
+
+~33s full-frame rs_ cut. Avatar ≤20% (hook + CTA only); cutaways + chips carry it.
+Trend: "context engineering is the new prompt engineering" (AI Engineer World's Fair, this week).
+
+## Spoken VO (HeyGen — VO only)
+
+Prompt engineering is already dead.
+
+People shipping real agents stopped tuning prompts months ago. The model is not the variable anymore. What you feed it is. The right files. The right memory. The right tools. The right limits. Give a smaller model the right context. It beats a genius flying blind.
+
+They call it context engineering. It is the job now.
+
+So stop polishing the prompt. Build the context stack. Task. Files. Memory. Tools. Hard limits. Every run.
+
+The model is not the edge anymore. Context is.
+
+Comment the word context. I will send the context stack checklist.
+
+## Beat map (avatar ≤20%: hook + CTA)
+
+| # | Beat | Line | Top panel |
+|---|------|------|-----------|
+| 1 | Hook | "prompt engineering is already dead" | avatar + "PROMPT ENGINEERING / is dead." |
+| 2 | Cutaway cx_screens | "shipping real agents stopped tuning prompts" | dark room, engineers at glowing screens |
+| 3 | Chips (what you feed it) | "right files / memory / tools / limits" | FILES / MEMORY / TOOLS / LIMITS |
+| 4 | Cutaway cx_stack | "give a smaller model the right context" | components assembling into a glowing stack |
+| 5 | Name it | "context engineering" | MG stamp "// context engineering" |
+| 6 | Numbered chips (the stack) | "task / files / memory / tools / hard limits" | 01 TASK 02 FILES+MEMORY 03 TOOLS+LIMITS |
+| 7 | Cutaway cx_fuel | "the model is not the edge" | fuel/energy into a dark engine core |
+| 8 | Thesis | "model is not the edge. context is." | two-line, "context" blue |
+| 9 | CTA | "comment the word context" | avatar + end card |
+
+Resource: The Context Stack Checklist (4pp). CTA: comment "context".
diff --git a/content/tiktok/topics_log.md b/content/tiktok/topics_log.md
new file mode 100644
index 000000000..45c06a8ef
--- /dev/null
+++ b/content/tiktok/topics_log.md
@@ -0,0 +1,12 @@
+# rs_ TikTok topics log (@rafa.flows)
+
+| date | topic | keyword | status |
+|------|-------|---------|--------|
+| 2026-07-02 | Fable 5 relaunch / model-agnostic stacks | — | posted |
+| 2026-07-03 | AI certifications (rs-ai-certs) | — | posted |
+| 2026-07-03 | Fable 5 use cases (rs-fable5-usecases) | — | posted |
+| 2026-07-06 | OpenClaw agent access | ACCESS | posted 15:00 UTC |
+| 2026-07-06 | Vibe coding vs production | SHIP | scheduled 18:00 UTC |
+| 2026-07-06 | Your first boring AI agent | AGENT | scheduled 21:00 UTC |
+| 2026-07-06 | AWS $1B forward-deployed engineers | DEPLOY | produced · awaiting review (Drive 7/6/26; avatar-minimized, 5 cutaways) |
+| 2026-07-07 | Context engineering > prompt engineering | CONTEXT | produced · awaiting review (Drive 7/7/26; avatar 19%, 3 cutaways) |