Agent Skills from Hexclave for coding agents (Claude Code, Cursor, Codex, OpenCode, Gemini CLI and anything else that reads the Agent Skills spec).
Skills are grouped by category under skills/<category>/<skill-name>/, each with its own SKILL.md and supporting files (for example skills/deployments/ for Hexclave Deployments). This is the catalog layout the skills CLI discovers, and the same multi-skill style as vercel-labs/agent-skills and anthropics/skills.
With the skills CLI (skills.sh):
# see what's in the repo
npx skills add hexclave/skills --list
# install everything
npx skills add hexclave/skills
# install one skill
npx skills add hexclave/skills --skill hexclave-otel-clickhouseAdd -g to install for your user instead of the current project, and -a claude-code (or cursor, codex, ...) to pick an agent.
Claude Code plugin marketplace:
/plugin marketplace add hexclave/skills
/plugin install hexclave-deployments@hexclave-skills
| Skill | What it does |
|---|---|
hexclave-otel-clickhouse |
Self-hosted observability on Hexclave Deployments that replaces New Relic. Deploys ClickHouse, an OpenTelemetry Collector gateway (OTLP/HTTP with bearer-token auth) and an rrweb session-replay ingest service with a built-in player, then verifies them end to end. |
hexclave-admin-dashboard |
A private admin dashboard on Hexclave Deployments: Hexclave sign-in limited to a project permission, a Bun + React service that forwards to your private backend, the one-time trusted-domain and admin-account setup, and how to read and steer an AI agent's conversations from it. |
Three services in one Hexclave deployment group:
clickhouse: ClickHouse server, 8GB, always on, volume-backed, private onlyotel-gateway: OTel Collector contrib, OTLP/HTTP in, bearer tokens, disk-backed queue, public HTTPSreplay-ingest: Bun service for rrweb chunks, session list and player, public HTTPS
It ships the deploy file, Dockerfiles, ClickHouse schema and retention scripts, a Node OTel bootstrap, a masked rrweb recorder, and a deploy script with dry-run, check, apply, verify, logs and pause/resume modes. It also covers running the stack: a private dashboard with Hexclave Auth and a ClickHouse query gate, a site-analytics variant behind one public nginx, admin through hexclave exec (domains, secrets, logs, single-service deploys), and a tested path off PostHog Cloud. Measured throughput, compression and sizing are in references/costs.md.
You need a Hexclave project on a paid plan plus HEXCLAVE_PROJECT_ID and HEXCLAVE_SECRET_SERVER_KEY. Deploying creates billable resources.
One public serverless service added to your app's deployment group. It checks the Hexclave session and a project permission on every request and forwards allowed calls to /internal/dashboard/* on your private backend with a bearer token, so it never holds database credentials. It ships a typechecked Bun + React + Tailwind starter, the config and deploy snippets, a backend route skeleton, a script that sets the trusted domain and creates admin accounts with the permission, and a script that signs a browser into the development project for local testing.
You need a Hexclave project with Authentication, RBAC and Deploy, an existing private backend service, and hexclave login for the one-time setup.
pip install skills-ref
agentskills validate skills/deployments/hexclave-otel-clickhouse
agentskills validate skills/deployments/hexclave-admin-dashboardAdd a folder under skills/<category>/ (for example skills/deployments/) with a SKILL.md that follows the spec (name must match the folder name), add it to the table above, to skills.sh.json and to .claude-plugin/marketplace.json, and run the validator.