
Claude Skills by AgentShekel
github.com/AgentShekel[PROTOCOL] Tier-aware per-engagement acceptance methodology shared by all agency managers (marketing / dev / design). Defines per-tier (S/M/L) protocols: S has no manager phase, M adds Opus adversary, L adds cross-family consilium. Managers are JUDGES between producer and adversary, not re-runners. Reference-only — loaded by `*-manager` agents via skills frontmatter, not triggered by keywords.
[PROTOCOL] Single intake point for any agency task (marketing / dev / design). Classifies the domain, captures acceptance criteria, hands off to the matching domain lead. Does not plan, dispatch, or execute work. Use when: "мне надо агенси задачу", "агенси задача", "задача для агенси", "agency task", "agency work", "route agency".
[METHODOLOGY] Multi-format banner design (social, ads, web hero, print) with multiple art direction options and AI-generated visual elements. Preloaded by marketing-banner-designer agent.
[TOOL] Direct MCP integration with Codex CLI as agency tool. Codex (GPT-5-class) is available as MCP server, exposing tools `mcp__codex__codex` (start session) and `mcp__codex__codex-reply` (continue). Use Codex for: image generation, UI mockups, vision/multimodal review, cross-family second opinion. Reference- only — loaded by visual designers (design-visual-designer, marketing-banner- designer, design-ui-designer, design-presentation-designer) and adversary scripts via skill frontmatter.
[METHODOLOGY] AI-generated visual assets pipeline — logo generation (55 styles, Gemini AI), corporate identity program (50 deliverables, CIP mockups), icon design (15 styles, SVG, Gemini 3.1 Pro), social photos (HTML→screenshot, multi-platform). Preloaded by design-visual-designer agent.
[METHODOLOGY] Decompose a design engagement plan into atomic task files (engagement/tasks/*.md). Each task = one specialist + one deliverable + measurable done-when. Enables targeted re-dispatch on iter-2 REJECT instead of re-running the whole brand/UI phase. Pure reference — invoked by design-lead at Phase 2.5 (after plan.md, before dispatch). Mandatory for size: L. Recommended for size: M with ≥2 specialists. Skipped on size: S (single deliverable).
[METHODOLOGY] AI-First development methodology (spec-driven pipeline, project structure, skills/agents ecosystem, quality gates). Preloaded by dev-lead and dev-* lead agents.
[PROTOCOL] Minimal contract every dispatched specialist (engineer, designer, copywriter, analyst, etc.) follows when running inside an agency engagement. Carve-out from engagement-protocol — the 6-bullet specialist subset only, without lifecycle / cross-domain / archival detail. Preloaded by specialist agents via skills frontmatter. Pure reference — no triggers.
[PROTOCOL] Canonical contract for agency engagement lifecycle: artefact schemas, state files, iteration budget, escalation rules, cross-domain handoff, observability. Single source of truth referenced by agency-intake, leads, managers, directors, and pipeline skills. Pure reference — no triggers, not invoked by the secretary as a router target.
[METHODOLOGY] Feature delivery orchestration (agent waves, review cycles max 3 rounds, commit-per-wave). Preloaded by dev-engineering-lead agent.
[METHODOLOGY] Decompose a marketing engagement plan into atomic task files (engagement/tasks/*.md). Each task = one specialist + one deliverable + measurable done-when. Enables targeted re-dispatch on iter-2 REJECT instead of re-running the whole phase. Pure reference — invoked by marketing-lead at Phase 2.5 (after plan.md, before dispatch). Mandatory for size: L. Recommended for size: M with ≥2 specialists. Skipped on size: S (single specialist, single deliverable).
[METHODOLOGY] Git-backed task management that survives context compaction (tasks as .md files in work/tasks/ with dependencies and statuses, auto-save context for post-compact recovery). Preloaded by dev-engineering-lead agent.
[PROTOCOL] Generic system-optimization (skill-evolution) methodology shared by the three domain director-optimizers (dev / design / marketing). The SkillOpt loop adapted to a creative/dev agency: event-driven reflect → bounded edit → gate → slow-update over the agent/skill corpus, with a cross-family proposer and judge-only directors. Reference- only — loaded by `*-director` agents via skills frontmatter, not triggered by keywords.
[METHODOLOGY] Tech-spec decomposition into atomic task files with parallel creation and validation. Preloaded by dev-tech-architect agent.
[METHODOLOGY] Tech-spec creation (architecture, decisions, testing strategy, implementation plan) from approved user-spec. Preloaded by dev-tech-architect agent.
[METHODOLOGY] UI/UX design intelligence database (50+ styles, 161 color palettes, 57 font pairings, 161 product types, 99 UX guidelines, 25 chart types across 10 stacks — React, Next.js, Vue, Svelte, SwiftUI, React Native, Flutter, Tailwind, shadcn/ui, HTML/CSS). Preloaded by design-ux-designer agent.
[PROTOCOL] Cross-cutting validation chains. Defines which validators to run, in what order, with which triggers, across marketing / dev / design domains. Preloaded by lead agents (marketing-lead, dev-lead, design-lead). Pure reference — no triggers, not invoked by secretary.
[TOOL] Unified Yandex analytics hub configuration (shared tokens, cross-service workflows across Webmaster, Metrika, Wordstat, Direct, Search, seo-auditing, semantic-drift). Preloaded by marketing-web-analyst and marketing-traffic-lead agents.