Yandex Direct PPC Skill
The first open-source AI-agent skill for Yandex Direct (API v5), letting you audit, manage, and optimize Russian-market PPC campaigns through natural language.
Credentials stored in ~/.secrets with chmod 600 and explicit tool declarations; but Bash/WebFetch are allowed with no explicit user-confirmation or rollback mechanism for external write operations (create, delete, bid changes), and the scripts themselves are not visible to audit data flow.
SKILL.md orchestration logic is self-consistent and command structure clear, but the referenced scripts/*.sh and references/*.md are not present in reviewed files; no tests and no evidence of failure-feedback design on abnormal input.
Triggers are explicit (Russian/English), audience (Russian-market PPC) clear, quality gates disclosed; however core function depends entirely on the Yandex API/OAuth whose reachability from mainland China is doubtful, and non-fit boundaries are not declared.
Well-layered docs (on-demand references), bilingual README, quick start, clear MIT license; deducted for absence of versioning/changelog and unclear maintenance ownership and update path.
Claims 55 audit checks, a scoring system and benchmarks — structurally complete with marginal value, but only a self-reported one-evening origin and a single-account anecdote (stepback.ru); no representative output samples verify direct usability.
Only author claims and one anecdotal case (D→C+); no tests, no CI evidence, no third-party corroboration; key claims are not independently reproducible.
- Core function depends entirely on the Yandex API; reachability from mainland China is unverified — test endpoint connectivity before use.
- The referenced scripts and reference files were not part of the reviewed evidence; actual behavior and safety are unverified and this static review is low confidence.
- Destructive API operations (create/delete/bid changes) lack explicit confirmation and rollback design; trial read-only first on production accounts.
- No versioning or changelog; maintenance ownership and update path are unclear.
What does this skill do, and when should you use it?
This open-source Agent Skill by Nick Serebrov & Kobe 🐍 provides full-cycle management of Yandex Direct search and YAN (RSY) advertising via API v5. It exposes 8 commands covering account audits (55 checks, weighted 0-100 scoring with A-F grades), campaign CRUD, keyword and bid management, performance reporting, budget analysis, and optimization recommendations. The skill centers on a SKILL.md with on-demand reference files (audit checklist, scoring system, Russian market benchmarks, bidding strategies, compliance rules, image specs) and shell-based API wrapper scripts. The authors describe it as the first open-source skill for Yandex Direct, created after finding that claude-ads covers Google/Meta/LinkedIn but not Yandex.
Calls the Yandex Direct API v5 (https://api.direct.yandex.com//v5/, Bearer OAuth auth, credentials in ~/.secrets/yandex-direct.) through the scripts/yd-api.sh wrapper. Concretely: /direct audit fetches all campaigns, ad groups, keywords, ads, and sitelinks, then runs 50+ checks to produce a 0-100 score, A-F grade, findings, and a prioritized action plan; /direct campaigns lists campaigns with statuses and key metrics; /direct create builds campaigns via campaigns.add, adgroups.add, keywords.add, and ads.add (TextAd for search, TextImageAd for YAN); /direct keywords adds, suspends, resumes, re-bids, or deletes keywords; /direct report queries the Reports service for TSV data (campaign/adgroup/keyword/search_query presets, multiple date ranges); /direct optimize analyzes 30 days of data for zero-conversion keywords and low-CTR ads using the 3x Kill Rule; /direct budget computes run rate and projected monthly spend; /direct negative manages campaign-level and shared negative keyword sets.
- A PPC marketer running Yandex Direct campaigns in the Russian market who wants a full account health check and prioritized fix list from one prompt
- An agency account manager who must review multiple Direct accounts quickly to spot wasted spend and low-quality ads
- A user of claude-ads (Google/Meta/LinkedIn ads) who needs to cover the Yandex Direct channel their current skill doesn't support
- A beginner advertiser who wants their account scored against built-in Russian CTR/CPC/CVR benchmarks (as the authors did, improving their own account from grade D to C+)
- An optimizer who needs recurring spend reports, budget forecasting, and alerts for under- or overspending campaigns
What are this skill's strengths and limitations?
- Fills a genuine gap: the first open-source AI-agent skill for Yandex Direct, a platform that claude-ads and similar skills don't cover
- Concrete, executable audit methodology: 55 numbered checks (YD01-YD55), weighted scoring across 6 categories, and built-in Russian market CTR/CPC/CVR benchmarks
- Covers the full campaign lifecycle, from creating campaigns via real API calls to negatives and budget forecasting
- Encodes industry-quality guardrails: 3x Kill Rule, no broad match without auto-strategy, no strategy changes in the first 7 days, mandatory UTM parameters
- MIT licensed, free to fork and modify
- Requires Yandex Direct API access approval, which can be a barrier to quick onboarding
- Yandex Direct / Russian market only — limited value for teams advertising elsewhere
- Reference benchmarks are static (e.g., search CTR 5-8%, CPC 30-150₽); actual figures vary heavily by industry
- No repository topics, no visible test suite, or independent validation; reliability rests on the authors' own account (a single D→C+ audit anecdote)
- Installation is documented for OpenClaw and Claude Code only; other agent frameworks are unverified
How do you install this skill?
1) Apply for Yandex Direct API access at direct.yandex.ru → API settings; 2) Get an OAuth token via https://oauth.yandex.ru/authorize?response_type=token&client_id=YOUR_CLIENT_ID; 3) Write credentials to ~/.secrets/yandex-direct. (client_id, client_secret, oauth_token) and run chmod 600 on the file; 4) Copy the yandex-direct folder to your agent's skills directory: for OpenClaw run cp -r yandex-direct ~/clawd/skills/, for Claude Code run cp -r yandex-direct ~/.claude/skills/. No other installation methods are documented.
How do you use this skill?
After installing, issue a natural-language command such as "Run a Yandex Direct audit". SKILL.md defines eight subcommands — /direct audit, campaigns, create, keywords, report, optimize, budget, negative — with trigger phrases including "Yandex Direct", "Яндекс Директ", "директ", and "РСЯ". API credentials must be configured beforehand; without them, all API-dependent commands cannot run.