Back Office · office.temerarii.xyz
One asset, all the way in — composition, the wireframe + storyboard, the output format stack, and the template, all read from the SAME content-index record. The expected output matches what /media surfaces for this post.
post longform-W44-Sunkind longformweek W44date 2026-11-01campaign longform-youtubepillar multimediabeat asset videoduration 251.9sground whitescenes 10

Checklist the per-video bar — engine/sim

100.0/100
plain languagevo coverageno dead airuniquenesscaption fitcompletenesscleanliness
quantitative quality · weights learn from your reviews (engine.sim.memory review longform-W44-Sun good|bad)
✓ all static checks pass — one-focal/scene · tier-by-beat · one-track caption · colorway · cast+shape correct · no banned/fabricated. (audio + visual tiers verify on the rendered finals — Phase 2)

Composition comp · template family · expected output

composition LongFormChaptersfamily / template LongFormChapters
9:16 Reelpending1:1 Squarepending16:9 Widepending9:16 4Kpending1:1 4Kpending16:9 4KpendingGIF (SMS)pending
render pending — silent master not yet on disk
expected output: 0/7 rendered — same matrix the /media preview surfaces for this asset.

Composition layer × scene 10 scenes · 251.9s · comp_id + rendered still + tier + the script

#Layer (comp_id · still · tier)BeatTimecodeMotionLogoAudioVO / on-screen / caption
1s1
matches intent
shared field
signature-3d
open0–30.9sspatial-parallaxicon·color♪ bed_in
This week the calendar says one thing: build presence in public. Most people treat social media like a slot machine. They post, they pray, they refresh. We are going to do the opposite. We are going to show you the actual machine that runs social for The Big T-M, the whole strategy, out in the open. A presence is not luck. It is a set of plain moves you can name, and once you can name them, a model can run them with you watching.
on-screen: We build in public on purpose
expected on screen: white ground · octa hero in the shared Signal Field · Lumen leads · node-graph · spatial-parallax · icon·color logo · caption bottom-left
spec (the prompt): comp_id shared Signal Fieldvisual node-graphshape octaground whitetreatment colormotion spatial-parallaxpower summoninstrument summon→We build in public on purpose
2s2
matches intent
NumberedList
template
teach30.9–57.9skinetic-buildicon·color♪ node_lock
Here is the first move. The whole social plan starts in one place: a single source file on disk. A plain text file, JSON, that lists every week, every day, every post. We do not keep the plan in someone's head or in twelve browser tabs. We keep it in one file the machine can read. When you change the file, every post downstream changes. One source of truth, not forty copies drifting apart.
on-screen: One file feeds every post
expected on screen: white ground · a NumberedList panel over a dimmed Signal Field · Lumen leads · node-graph · kinetic-build · icon·color logo · caption bottom-left
spec (the prompt): comp_id NumberedListvisual node-graphshape octaground whitetreatment colormotion kinetic-buildpower morphinstrument morph+laser→One file feeds every postcurate items, nodes
3s3
matches intent
ChecklistCard
template
teach57.9–85.2skinetic-buildicon·color♪ node_lock
The second move is the brain. We open the terminal and run Claude Code, a coding agent that holds the whole project at once. We point it at that source file. It reads the calendar the same way a person would, except it never loses its place. You are not clicking through an app. You are telling one agent what the week is about, and it already knows the plan because it just read it.
on-screen: Claude Code reads the plan
expected on screen: white ground · a ChecklistCard panel over a dimmed Signal Field · Lumen leads · node-graph · kinetic-build · icon·color logo · caption bottom-left
spec (the prompt): comp_id ChecklistCardvisual node-graphshape octaground whitetreatment colormotion kinetic-buildpower morphinstrument morph+laser→Claude Code reads the plancurate items, nodes
4s4
matches intent
ListCard
template
teach85.2–108.6skinetic-buildicon·color♪ node_lock
Third move. A model that can only talk is useless for social. So we wire it to real tools using MCP servers, small adapters that let the agent call an app directly. One adapter for posting, one for listening, one for the calendar. The model stops being a chatbot and becomes a hand that can reach into your accounts and actually do the work.
on-screen: MCP wires the model to tools
expected on screen: white ground · a ListCard panel over a dimmed Signal Field · Lumen leads · node-graph · kinetic-build · icon·color logo · caption bottom-left
spec (the prompt): comp_id ListCardvisual node-graphshape octaground whitetreatment colormotion kinetic-buildpower morphinstrument morph+laser→MCP wires the model to toolscurate items, nodes
5s5
matches intent
LegacyCard
template
teach108.6–132.7skinetic-buildicon·color♪ node_lock
Fourth move. You do not need to be everywhere. We list the platforms where the audience actually is and ignore the rest. For us that is a short list, not nine. Tell the agent which platforms matter, and it shapes each post to fit. A long post for one, a short one for another, vertical video for a third. Same idea, different clothes, picked on purpose.
on-screen: Pick the platforms that fit
expected on screen: white ground · a LegacyCard panel over a dimmed Signal Field · Lumen leads · node-graph · kinetic-build · icon·color logo · caption bottom-left
spec (the prompt): comp_id LegacyCardvisual node-graphshape octaground whitetreatment colormotion kinetic-buildpower morphinstrument morph+laser→Pick the platforms that fitcurate nodes
6s6
matches intent
CodeWindow
template
teach132.7–157.2skinetic-buildicon·color♪ node_lock
Fifth move. We take one idea from the plan and ask the model to write it many ways, one per platform, in our voice. We give it the brand rails as plain rules: short sentences, no empty hype words, teach the real method. The model drafts. We cut. What used to be a full day of writing becomes a review of drafts that already sound like us.
on-screen: One idea becomes many posts
expected on screen: white ground · a CodeWindow panel over a dimmed Signal Field · Lumen leads · node-graph · kinetic-build · icon·color logo · caption bottom-left
spec (the prompt): comp_id CodeWindowvisual node-graphshape octaground whitetreatment colormotion kinetic-buildpower morphinstrument morph+laser→One idea becomes many postscurate codeLines, nodes, windowTitle
7s7
matches intent
StepFlow
template
teach157.2–181.0skinetic-buildicon·color♪ node_lock
Sixth move. Posting blind is guessing. So we listen. We wire the agent into a search tool and into Reddit through an MCP, and pull what people are actually saying about our corner of the work. The model reads the chatter and hands back the real questions people ask. Those questions become next week's posts. The audience writes our outline for us, without knowing it.
on-screen: Listening feeds tomorrow's posts
expected on screen: white ground · a StepFlow panel over a dimmed Signal Field · Lumen leads · node-graph · kinetic-build · icon·color logo · caption bottom-left
spec (the prompt): comp_id StepFlowvisual node-graphshape octaground whitetreatment colormotion kinetic-buildpower morphinstrument morph+laser→Listening feeds tomorrow's postscurate nodes, steps
8s8
matches intent
NodeGraphCard
template
teach181.0–203.3skinetic-buildicon·color♪ node_lock
Seventh move. Once the posts exist, we do not paste them in by hand at midnight. We hand the agent a posting tool and the dates from the source file, and it queues every post to go out on its own. The whole week lines up from the same file we started with. You approve once. The machine handles the clock.
on-screen: The calendar schedules itself
expected on screen: white ground · a NodeGraphCard panel over a dimmed Signal Field · Lumen leads · node-graph · kinetic-build · icon·color logo · caption bottom-left
spec (the prompt): comp_id NodeGraphCardvisual node-graphshape octaground whitetreatment colormotion kinetic-buildpower morphinstrument morph+laser→The calendar schedules itselfcurate hub, nodes
9s9
matches intent
RankList
template
proof203.3–228.5sreceipts-counticon·color♪ node_lock
Here is the honest proof. The video you are watching right now came out of that same source file. The plan, the script, the schedule, all of it traces back to one file and one agent in a terminal. We did not stage a demo. You can see the whole thing running, live, at office dot temerarii dot xyz. If it works on us, it can work on you.
on-screen: This video came from that file
expected on screen: white ground · a RankList panel over a dimmed Signal Field · Lumen leads · receipts · receipts-count · icon·color logo · caption bottom-left
spec (the prompt): comp_id RankListvisual receiptsshape octaground whitetreatment colormotion receipts-countpower receiptsinstrument spotlight→This video came from that filecurate rows, statsLabels
10s10
matches intent
shared field
signature-3d
resolve228.5–251.9scoalescenceicon·color♪ bed_out
So that is the takeaway. Building presence in public is not a personality trait. It is a few plain moves: one source file, one agent, real tools through MCP, listen, write, schedule. Name each move, then let the model run the boring parts while you keep the taste. Go watch it run at office dot temerarii dot xyz, then go name your first move.
on-screen: Name the move, then automate it
expected on screen: white ground · octa hero in the shared Signal Field · Lumen leads · coalescence · coalescence · icon·color logo · caption bottom-left
spec (the prompt): comp_id shared Signal Fieldvisual coalescenceshape octaground whitetreatment colormotion coalescencepower coalescenceinstrument coalescence→Name the move, then automate it

Format stack 1 aspects · same scenes[], re-cropped

16:9
1920×1080
X/Twitter · YouTube · LinkedIn video

Channels 2 destinations

YouTubeBlog

Social captions supplemental published copy · per channel (comp_id level)

youtubeHow We Run Social Media in Public With One File and an AI Agent Most people treat social media like a slot machine: post, pray, refresh. We do the opposite, and in this video we walk you through the whole machine that runs social for our studio, out in the open. The core idea is that a presence is not luck. It is a set of plain moves you can name, and once you can name them, a model can run the boring parts while you keep the taste. What we cover, step by step: - One source file. The whole plan lives in a single JSON file on disk that lists every week, every day, every post. Change the file and every post downstream changes. One source of truth instead of forty copies drifting apart. - The brain. We open the terminal and run Claude Code, a coding agent that reads that file and holds the whole project at once. It never loses its place in the calendar. - MCP. A model that can only talk is useless for social, so we wire it to real tools with MCP servers, small adapters that let the agent post, listen, and read the calendar directly. - Pick the platforms that fit. You do not need to be everywhere. List where your audience actually is, ignore the rest, and let the agent shape each post to fit. - One idea, many posts. We take one idea and ask the model to write it many ways in our voice, using the brand rules as plain instructions. We draft, then we cut. - Listening. We wire the agent into search and into Reddit through an MCP and pull what people actually say about our corner of the work. Their questions become next week's posts. - The calendar schedules itself. We hand the agent a posting tool and the dates from the source file, and it queues the whole week. You approve once. The honest proof: this video came out of that same source file. The plan, the script, and the schedule all trace back to one file and one agent in a terminal. You can watch the whole thing running at office.temerarii.xyz. The takeaway: name each move, then let the model run the parts that do not need a human. Keywords: social media automation, AI agent, MCP, Claude Code, content workflow.

Cross-links every lens is a view on this one record