Promo for a five-session workshop launching in July.
Content system
harness.
A brand voice, a design system and a content pipeline wired into one system, built by Nadia Gorelikova. Every step is a file, every file has a place in the chain, and the numbers go back into the plan.
source of truth. Five skills read the same brand file before they write a word.
post templates with locked colour pairs and character limits, exported straight to PNG.
reach → profile visit on the live account, then 5.2% profile → link click.
Four layers, one loop
Brand knowledge lives in one place and everything downstream reads from it. Design rules are code, not a PDF. Analytics return to the content plan, so the next month starts from the last one's numbers.
brand-core
Voice, vocabulary, limits, checklists. tov.md and visual.md are what a new brand fills in.
Design tokens
Colour, type, 8pt grid, elevation. One CSS file, imported everywhere.
content-strategy
Rubrics, monthly plan, funnel split.
social · video
Platform posts, hooks, storyboards.
copy-editing
Voice check, AI-pattern cleanup, cuts.
carousel-export
Approved text → templates → PNG.
Social sandbox
Optional human-in-the-loop station. 7 templates, locked colour pairs.
Notion calendar
Each post tagged by funnel stage, format and CTA.
PNG / ZIP export
1080×1350 posts and 9:16 stories, ready to upload.
Social dashboard
Built on the Instagram API. Inside signal: your own reach, ER and reach → visit → link conversion, per post.
Trend agent
Outside signal: Apify collects Instagram and TikTok trends, Gemini turns them into adapted video briefs.
Feedback. Dashboard and trend scores go back into content-strategy.
What reads what
The map above as a contract. Each node declares its input and its output, so any piece can be replaced without touching the rest.
| Node | Reads | Produces |
|---|---|---|
| brand-core | The brand brief, filled once into tov.md | Voice rules, vocabulary, banned patterns |
| content-strategy | brand-core, dashboard and trend data | Rubric map, monthly plan, funnel distribution |
| social / video | brand-core, the plan | Platform-ready copy, video scripts |
| carousel-export | brand-core, approved text | Carousel as PNG |
| copy-editing | brand-core, any draft | Draft cleaned against voice rules and AI clichés |
| Social sandbox (optional) | Design tokens, approved copy | Hand-finished carousels and stories as PNG or ZIP |
| Notion calendar | Plan and finished posts | Publishing schedule with funnel, format and CTA tags |
| Dashboard | Instagram API: account and post insights | Per-post reach, ER and conversion chain |
| Trend agent | Apify trend data, keyword list | Ranked trend report, adapted video briefs |
Inside the system
Brand knowledge as a file tree
The difference between a prompt and a system is where the knowledge sits. Here it sits in one document that every skill opens before generating anything.
- Fill
tov.mdwith examples, not rules. Examples travel further. - Skills never copy brand facts into themselves, they point to
brand-core. - Change the file and all five skills change voice.
Design rules in code, sandbox on demand
The visual system is digitised down to the constraint: tokens, templates, colour pairs, character limits. The carousel-export skill maps approved text onto templates, checks every field against its limit and renders PNG, so most content is produced without opening a tool.
The sandbox is optional. It is a human-in-the-loop station for posts that need a person's eye, such as photo covers and key announcements. It enforces the same rules: approved colour pairs, a live counter against each template's limit, export at the correct size.
- Skills and sandbox read the same templates and limits, so automatic and hand-finished posts look alike.
- Seven templates: hook, problem / solution, stat, list, quote, CTA, photo cover.
- Five colour fields, no free recolouring. Cherry is reserved for CTA.
- Limits are set per template before layout, so text never gets squeezed after the fact.
A plan the skill builds
The content-strategy skill builds the rubric map and the month plan from brand-core, and the plan lands in Notion. I approve the map, the skill does the rest.
- Funnel stage and content type are tags the skill assigns, so balance is visible at a glance.
- Each post ends with one action: save, comment or link in bio.

Signals that feed the plan
Two instruments close the loop. The dashboard is built on the Instagram API. It is the inside signal and reads the account the way a funnel is read: reach, then profile visit, then link click. The trend agent is the outside signal: Apify collects Instagram and TikTok trends, they are scored (261 in one run) and Gemini turns the strongest into video briefs adapted to the brand.
- Conversion is measured per step, so a weak post shows where it loses people.
- Trends get a combined score from engagement and keyword fit.


One request, a finished chain
A single request triggers the whole path. My decisions sit at the checkpoints: the rubric map, the draft voice and, for the share of posts planned as human-in-the-loop (30% at most), the final visual. Everything else runs from the files.
“Plan the month and write the announcement carousel.”
- Rubric map and month plan from the brand file
- Seven-frame carousel in the locked palette
- Entries with funnel tags in the calendar
A month of on-voice content planned, designed and scheduled from one request. The checks that stay with me: positioning, tone and the last look at the visuals marked for human review, 30% of posts at most.
What the system produced
Live account, period 29.06 to 14.07, four posts.
reach → profile visit
profile visit → bio-link click
total ER
Small account, short window, and the funnel steps still convert at a clear rate. Business impact from a larger channel is in the TripleTen case: organic social grown into a $1M+ annual-bookings channel.
Packaged for other brands
For other brands I either produce the visual identity with AI or take the one they already have, and translate it into brand tokens an agent can work with: palette, type, templates and a voice file. The skills are then adapted to the brand, or new ones are written for what makes it unique, so content production continues without me.
- Identity: produced with AI, or the brand's existing one.
- Tokens: rules, templates and a voice file the agent reads, packaged by a wrap-up skill into a zip.
Brand tokens for the agent: a brand-shape layer, four templates, a voice file.


Brand tokens for the agent: a dot cover, a colour code per content type, a voice file.


Fuelly, a protein-first tracker
A calorie tracker I vibe-coded in a week. The design and the product logic are mine. Now in beta and backed by tests.
Need a content system for your brand?
I build content and design systems that keep a brand producing.
