Reusable, verification-oriented skills for AI agents. Each skill is a focused, dense workflow — written for the model, not as a tutorial — and can bundle references, decision guides, or deterministic helper scripts.
They follow the open SKILL.md format and work with Claude Code,
Codex, Cursor, Gemini CLI, GitHub Copilot, VS Code, opencode, and other compatible agents.
Skills are grouped into categories under skills/. Add a new one in the fitting
category (or a new category); the tooling discovers any skills/**/SKILL.md.
A connected set for running Meta ads across clean and grey verticals: four domain layers plus one orchestrator that decides between them.
| Skill | Purpose |
|---|---|
| meta-ads | Clean-marketing reference: plan, launch, audit, diagnose, and optimize Meta ads — objectives/ODAX, budgets & bidding, targeting, pixel/CAPI, policy, and a canonical Marketing API error catalog. |
| fb-grey-ops | Infrastructure & survival layer for grey verticals: antidetect/proxy/session discipline, agency accounts, token/session death, API mass-launch, portfolio forensics, and per-vertical playbooks (nutra, iGaming, crypto, news). |
| tracker-ops | Affiliate tracker operations: Keitaro & Binom APIs, postback/S2S, metric discipline (payout event, timezones, CPL), cohort nowcasting, the Meta↔tracker mapping & backend-optimization contract, and the daily spend-sync. |
| measurement-experimentation-ops | Decide whether a result is real before acting: testing-mode selection (causal / screening / infrastructure), validity traps (SRM, peeking, contamination, lag), and Meta's measurement tools — A/B Test, ad_study API, Conversion Lift, GeoLift, Robyn/MMM. |
| senior-buyer-ops | Senior-buyer / team-lead operating layer: the day-1 operating contract, portfolio allocation, kill/watch/scale + marginal-scaling ladder, creative-intelligence pipeline, and end-to-end funnel QA — orchestrates the adapters above. |
| Skill | Purpose |
|---|---|
| responsive-adapter | Adapts existing web interfaces from 320px phones to 2560px+ displays while preserving the original visual system, then verifies across a device matrix. |
| design-stack-picker | Selects compatible fonts, icons, components, imagery, motion, and design primitives with a reuse-first approach. |
| Skill | Purpose |
|---|---|
| secure-coding | Secure defaults and review guidance for JavaScript, Node.js, HTML, CSS, APIs, authentication, databases, uploads, GraphQL, and AI-assisted code. |
| js-obfuscation | JavaScript protection, obfuscation, anti-automation, and anti-debugging techniques for authorized testing and software protection. |
| Skill | Purpose |
|---|---|
| deep-research | Traceable multi-source research with primary-source prioritization, verification, confidence labels, and explicit knowledge gaps. |
Clone the collection:
git clone https://github.com/sergeyizmailov/Claude-Skills.gitInstall one skill (copy the skill directory itself, not its category):
mkdir -p ~/.claude/skills
cp -R Claude-Skills/skills/media-buying/meta-ads ~/.claude/skills/
cp -R Claude-Skills/skills/security/secure-coding ~/.claude/skills/Or keep the categories and install everything:
cp -R Claude-Skills/skills/* ~/.claude/skills/Agents discover skills recursively, so category subfolders are fine. For Codex, copy into
~/.codex/skills/ instead; other agents may use a different skills directory.
skills/<category>/<name>/
├── SKILL.md # required — frontmatter (name, description) + the workflow
├── agents/ # optional — agent-facing metadata
├── references/ # optional — on-demand docs, loaded only when needed
└── scripts/ # optional — deterministic helpers
Only name and description load for discovery; the body and supporting files load when the
skill is selected. The media-buying skills also cross-reference each other by name (e.g.
senior-buyer-ops routes into meta-ads / fb-grey-ops / tracker-ops /
measurement-experimentation-ops).
Improvements and new skills are welcome. See CONTRIBUTING.md for the format, quality bar, and local checks.