Most agent integrations stall on brittle adapters and undocumented automation steps; this repo treats each capability as a self-contained "Skill" the agent can load on demand. The practical payoff is predictable agent behaviour: instead of hand-assembling prompts and glue code, you drop a skill folder (SKILL.md + optional assets) into the agent and let the agent follow a tested workflow.
What Sets It Apart
- Opinionated skill contract: each skill is a self-contained folder anchored by SKILL.md (YAML frontmatter + instructions). That contract lets agents decide when to activate a skill and keeps behaviour deterministic across agents.
- Multi-agent compatibility: tested paths and install tooling target Claude Code (marketplace), Cursor, Codex CLI and other agent runtimes; five install methods (npx skills CLI, marketplace, pinned release .zip, manual copy, git submodule) cover CI, air-gapped, and local-hack workflows.
- Practical, task-focused skills: notable skills include web-video-presentation (record-ready 16:9 stage, 23 themes, pluggable TTS), gpt-image-2 (three runtime modes: local Garden, host-native delegation, advisor-only), kb-retriever (progressive, source-aware retrieval for Markdown/PDF/Excel) and beautiful-article (editorial-first article assembly with hard collaboration checkpoints).
- Reproducibility and release hygiene: each skill publishes pinned .zip artifacts with checksums for CI/production use, and the README exposes the current pinned links so installs can be deterministic.
Who It's For and Trade-offs
Great fit if you build or integrate AI coding agents and need reusable, reviewable automation units: product teams wiring Claude Code or Cursor plugins, researchers prototyping agent workflows, and engineers automating media production or document retrieval pipelines. It reduces ad-hoc prompt sprawl by encoding decisions into skill manifests.
Look elsewhere if you need a single standalone application (this repo is a skills library, not a hosted SaaS) or if your environment cannot run or host an agent runtime — many skills expect an agent orchestration layer. The collection also assumes familiarity with Node tooling and common agent install paths; it's not a zero-setup consumer product.
Where It Fits
Use this as the canonical skills registry when you want predictable agent behaviour and reproducible installs. Combine skills for larger workflows (for example: kb-retriever → beautiful-article → web-video-presentation) and pin skill zips in CI for stability. The repo emphasizes editorial checkpoints and source-aware retrieval over fully autonomous execution, so human-in-the-loop review is part of the intended workflow.
