SkillVault — Curated Skill Packs for AI Coding Agents
The community has written hundreds of agent skills; most projects need about five of them at any given moment. SkillVault is the curation: 25 vetted third-party skills, staged by project lifecycle.
Overview
SkillVault is a curated starter-pack of 25 third-party Claude Code skills — 289 files of community work, hand-picked and organized into five project-lifecycle stages (design, security, debug, backend, process) so a new project copies in exactly what its current stage needs instead of everything. An agent-readable index maps each skill to trigger keywords and use cases. The curation is the work here, and the repo says so: none of the skill content is mine; the selection, staging, security-vetting and indexing are.
The Problem
Agent-skill discovery is a firehose: awesome-lists with hundreds of entries, no quality signal, no sense of when in a project's life each skill matters. Installing everything bloats the agent's context and the attack surface; installing nothing wastes the community's best work. The missing layer is opinionated curation with a lifecycle map — and a security gate in front of the door.
Questions Addressed
- 01
Which community skills actually earn a place in a working stack, stage by stage?
- 02
How does a curated set stay consumable by agents, not just humans?
Methodology
Select & Stage by Lifecycle
25 skills organized into five stages — 11 design (UI auditing, design tokens, Tailwind systems), 2 security, 2 debugging, 4 backend, 6 process (TDD, code review, verification loops) — so a project at the design stage copies the design pack, not the world. Everything passes the same scan-before-install gate as the rest of the toolchain.
Index for Agents
An AGENTS.md index gives every skill trigger keywords and a one-line use case, making the vault navigable by the agents that consume it. Several live projects — including this portfolio site — run skill sets sourced from it.
Key Results
Key Findings
Curation is a real contribution in an ecosystem drowning in options — the value is what's excluded and when what's included applies.
Explicit non-authorship is the honest frame: organizing other people's best work is useful precisely because it says so.
Conclusion
SkillVault is a shelf, not a product — young, single-purpose, and already consumed by the other projects on this page. Together with the Skills Lab it splits the problem cleanly: the lab develops what I author; the vault stages the best of what everyone else did.