# agent-skills A public, agent-agnostic collection of [Claude Code](https://docs.claude.com/en/docs/claude-code) / [agentskills.io](https://agentskills.io/specification)-compliant skills. Each skill lives in its own self-contained folder under `skills/` and versions independently. ## Installing a skill Pick whichever fits your workflow — all three install from this same repo, no separate distribution channel. **skills.sh CLI:** ```bash npx skills add CervantesVive/agent-skills --skill ``` **Claude Code plugin marketplace:** ``` /plugin marketplace add CervantesVive/agent-skills /plugin install @agent-skills ``` **Manual copy:** ```bash cp -r skills/ /path/to/your/project/skills/ ``` ## Skills | Skill | Category | Description | Version | | --- | --- | --- | --- | | [expose-private-service](skills/infra/expose-private-service) | infra | Expose a private-network-only homelab service over HTTPS through an existing Traefik reverse proxy and Cloudflare DNS, using a trusted wildcard cert. Use when adding a new self-hosted service (Docker Compose or otherwise) that should be reachable only over a private network (Tailscale, WireGuard, plain LAN, a cloud VPC's private subnet, etc.), not the public internet. | 1.1.1 | | [llm-gateway-incident-triage](skills/infra/llm-gateway-incident-triage) | infra | Diagnostic playbook for triaging a self-hosted LLM gateway (LiteLLM or any Postgres-backed OpenAI-compatible proxy) running via Docker Compose on a homelab host. Use when a client reports connection failures, timeouts, "provider unreachable", "model server not responding", an unexpected fallback trigger, or an auth/key error against the gateway, and you need the actual root cause rather than a guess. | 1.0.0 | | [repo-first-service-staging](skills/infra/repo-first-service-staging) | infra | Use when deploying or updating a homelab self-hosted service through a git-tracked staging repo first, then previewing and applying config into runtime paths like /opt, /etc/systemd/system, /etc/traefik/dynamic, and persistent service directories. | 1.2.0 | ## Contributing / maintaining - Add a new skill: create `skills//SKILL.md` (optionally grouped one level deep as `skills///SKILL.md`) with `name`, `description`, and `metadata.version: 1.0.0` frontmatter (plus any `scripts/` it needs), then run `bun run refresh` to pick it up in this table and in `.claude-plugin/marketplace.json`. - Bump a skill's version after changing it: `bun run bump `. - `bun run refresh` is safe to run anytime — it never changes version numbers, only adds/removes skills and syncs descriptions. - If a skill has supporting files (`scripts/`, `references/`, `templates/`, `assets/`), mention each one's relative path somewhere in `SKILL.md`'s body — installers like Hermes only bring along files that are explicitly referenced there.