ucoz-skills Ad Campaign Landing Skill (agent-native)
One skill = from ad creative → segment landing (hyper-segmentation) + UTM package + push to Yandex Direct / Google Ads via API.
Version 1: agent performs all operations natively — without Python scripts. Publishing via MCP (templates_tool, ftp_tool), API requests via Shell (curl), HTML and JSON generated by the agent.
Rule 0 — Usage accounting
After the skill's main work completes successfully, record usage:
skills_tool(action="register_usage", skill_id="ucoz-ad-campaign-landing-skill")
If the usage API is unavailable or returns an error, show a brief warning and continue the skill's main workflow. Do not ask the user for a UUID, token, or site URL — the site and token come from the active ucoz-mcp connection.
Main scenario (creative → landing)
references/PIPELINE.md §2A — ad exists → headline_1 = H1 → separate URL → page_add.
ad (CSV or text)
→ agent builds segments.json (headline_1 = H1, page_ownurl = transliteration)
→ agent checks message match (COPY.md §Message Match)
→ agent generates HTML per LANDING_PLAYBOOK.md rules
→ LANDING_PLAYBOOK.md (page_add, §0.1)
→ agent builds UTM strings
→ (optional) push to Yandex Direct/Google Ads — only with user permission
→ OUTPUT.md
No creative → Wordstat/WebSearch: references/PIPELINE.md §2B (branch 2B).
Start with user
No full brief → references/INTAKE.md → brief.json → pipeline.
Must clarify with user (if not stated), before API push and large-scale deployment:
| Question | Brief field | Impact |
|---|---|---|
| Test or full Yandex Direct API access? | yandex_direct_access |
sandbox vs prod host |
| One landing or separate URL per ad/segment? | landing_scale, max_segments |
number of page_add, message match |
Branching and checklist: INTAKE.md (§ API access, § Landing scale).
What the agent does natively (no scripts)
| Task | How agent does it |
|---|---|
| CSV → segments.json | Reads CSV in context, builds JSON per templates/segments.manifest.example.json |
| Slug transliteration | Converts headline_1 → latin, 2–4 words, hyphenated |
| Message match check | Rules from COPY.md §Message Match (overlap_ratio in agent's context) |
| Landing HTML generation | Per LANDING_PLAYBOOK.md + COPY.md rules |
| UTM URL building | Per formula from PIPELINE.md §6, URL-encode via Shell if needed |
| Publishing pages on uCoz | LANDING_PLAYBOOK.md → MCP page_add |
| Final report | Per OUTPUT.md template, data from segments.json |
API modes — Yandex Direct and Google Ads
Rule: push only with explicit permission
Before any real API call to Yandex Direct or Google Ads the agent must:
- Show the user the payload (what exactly will be sent).
- Ask for permission: "Shall we submit to [sandbox / prod]?"
- Receive explicit confirmation → execute
curlvia Shell.
Without confirmation — dry-run description only.
Yandex Direct
| Mode | Host | When to use |
|---|---|---|
| Sandbox (test) | api-sandbox.direct.yandex.com |
Debugging, no live token |
| Production | api.direct.yandex.com |
Real campaigns (created as SUSPENDED) |
API structure, payload, authorization: references/DIRECT_API.md
Get OAuth token (manual step for user):
1. Open: https://oauth.yandex.ru/authorize?response_type=code&client_id=<CLIENT_ID>
2. Get code → POST https://oauth.yandex.ru/token → bearer token
3. Save to env: YANDEX_DIRECT_TOKEN=<token>
Google Ads
| Mode | How to activate | When to use |
|---|---|---|
| Dry-run | Agent describes payload, doesn't call API | Debugging |
| Test Account | GOOGLE_ADS_CUSTOMER_ID=<test ID> |
Test account |
| Production | GOOGLE_ADS_CUSTOMER_ID=<live ID> |
Real campaigns (created as PAUSED) |
API structure, payload, authorization: references/GOOGLE_ADS.md
Fetching existing ads from account
Agent calls API directly via Shell (curl). Request and response format: references/ADS.md §Fetch.
Full cycle "account → landing → update URL": references/ADS.md §Sync.
Rules
| Rule | Meaning |
|---|---|
| Yandex Direct: access mode | Ask test / full; host per INTAKE.md; do not push to prod with test application |
| Landing scale | Ask single / multi / match_ads; default multi |
| API push | Only after explicit user permission; show payload before sending |
| Yandex Direct: no token | DIRECT_API.md → OAuth instructions |
| Google: no credentials | GOOGLE_ADS.md → developer token, OAuth |
| Message match | H1 = headline (COPY.md §Message Match) |
| Ad copy | COPY.md — required before package |
| URL / segment | page_ownurl + page_add before delivering link |
| Wordstat | 3–5 seeds per brief (RESEARCH.md); fallback — WebSearch |
| Competitors | research.competitors → competitor_insights (RESEARCH.md §Competitors) |
| API errors | Billing → key → folder → roles |
Documentation
| File | Content |
|---|---|
| PIPELINE.md | Full pipeline: branches 2A/2B, steps 0–8, checklist |
| INTAKE.md | Dialog start, questions, brief.json schema |
| ADS.md | Ads: CSV import, fetch from account, formats, URL sync |
| COPY.md | Ad and landing copy, message match |
| RESEARCH.md | Wordstat API, SERP, competitors, WebSearch fallback |
| DIRECT_API.md | Yandex Direct API (payload, OAuth, sandbox/prod) |
| GOOGLE_ADS.md | Google Ads API (setup, push, errors) |
| LANDING_PLAYBOOK.md | uCoz MCP: page_add, CSS/JS, forms, burger, favicon |
| OUTPUT.md | Final report and page deployment |
| EXAMPLE.md | End-to-end examples (3 cases) |
Prohibitions
- Do not connect separate
ucoz-landing-agent— everything is inLANDING_PLAYBOOK.md. - Do not output
final_urlbefore successfulpage_add. - Do not use one landing for all ads.
- Do not use
page_id=1for all segments without request. - Do not push to prod Yandex Direct when
yandex_direct_access=test. - Do not call Yandex Direct / Google Ads API without explicit user permission — show payload first.
- Do not run Python scripts — agent does everything natively.