Skills Reference
The Obol plugin contains the skills below. Claude picks the right one from your request. You can also call one directly with /obol:<skill-name>.
| Skill | Use it to | Runs against |
|---|---|---|
create-cluster-invitation | Create a DV cluster invitation and coordinate the DKG | Docker (charon image) or the Obol SDK |
test-a-dv-cluster | Run charon alpha test suites against a node or cluster | A running Charon, or Docker |
dvpod | Deploy, upgrade, back up, and troubleshoot a DV on Kubernetes | kubectl + helm |
dvpod-monitoring | Query the metrics and logs of a deployed DVpod (read-only) | kubectl + helm |
obol-monitoring | Diagnose clusters, nodes and the fleet in Obol's hosted Grafana: missed duties, offline peers, slot timelines, the biggest clusters' easy wins | Python 3, Grafana API token |
run-obol-stack | Install, operate, and sell services from the Obol Stack, and run your part of a DV in it | Docker |
distributed-validators | Look up clusters, validator effectiveness, exits, operators and who runs a peer in the Obol API (read-only) | The bundled Obol MCP server, or curl |
dv-steward | Let the Obol Stack's Hermes agent watch the Stack's own DV node and alert you (read-only) | An Obol Stack agent |
create-cluster-invitation​
This skill walks a cluster creator through making the cluster-definition.json that other operators accept, and then through the DKG ceremony that produces the key shares and cluster-lock.json. It works for solo operators, groups of friends, squads, and institutional or Lido Simple DVT clusters. It also covers the choice between inviting operators by Ethereum address (they accept on the Launchpad) and inviting them by ENR (they accept on the CLI), and hands the result to a dv-pod by Launchpad invite or by the definition's config hash. Before anything is created, it checks that every operator runs the same Charon minor version, which the DKG requires. It explains what it means to use an OVM or splits contract as the withdrawal address.
The skill uses the charon Docker image by default. It turns to the Obol SDK only when you need to integrate programmatically. It pins the Charon version and keeps key material out of git repositories and cloud-synced folders. Before any deposit, it walks you through checking the cluster-lock, the withdrawal and fee recipient addresses, and the deposit data.
Needs: Docker. Have the operator addresses or ENRs, the withdrawal and fee recipient addresses, and the target network ready.
Try:
Help me create a 4-operator DV cluster on Hoodi with my friends, inviting them by Ethereum address.
Related docs: Create a DV With a Group, Create a DV Alone.
test-a-dv-cluster​
This skill runs the charon alpha test suites (infra, peers, beacon, validator, mev, all) and picks the right suite for where you are. A solo operator waiting on cluster-mates starts with infra, while the full suite is the final check before activation. Where it can, it runs the tests inside your running Charon container with --publish, so the signed results are tied to your node's real ENR. It reads Poor and Fail results back to you along with Charon's own suggestions.
Needs: a running Charon (Docker Compose or Kubernetes), or Docker for a one-off infra run.
Try:
Run the Charon test suites against my node and tell me if anything needs fixing before activation.
Related docs: Test a Cluster.
dvpod​
This skill deploys and manages a DVpod, which runs a distributed validator on Kubernetes using the dv-pod Helm chart. It handles the following actions:
deploy: create a new cluster or join an existing onestatusandlogstroubleshootupgradeenr: retrieve your node's ENRbackupandrecoverdestroy
It also sets up monitoring, either by remote-writing metrics to Obol, through a ServiceMonitor, or by shipping logs to Loki. It pins chart versions, sets the network at install, and reads your node's ENR and DKG progress from the chart's ConfigMaps. In the Obol Stack it installs with obol helm from the chart's Obol Stack values file, after the local node has synced. It backs up charon-data from the validator client container, since from chart 0.23 the Charon container can't see the key shares. It always asks before deleting a release, a secret, or a persistent volume, before upgrading a release whose validators are active, and before restoring keys during recover. It never reads private keys out of Kubernetes secrets.
Needs: a Kubernetes cluster, kubectl, helm, and a reachable beacon node endpoint.
Try:
Deploy a DVpod on my Kubernetes cluster to join an existing Obol cluster on mainnet.
dvpod-monitoring​
This skill investigates a running DVpod. It checks health, Charon errors, peer connectivity, duty performance, and beacon node behaviour. It queries the bundled Prometheus, the Obol Stack's Prometheus, Charon's /metrics endpoint, kubectl logs, or Loki, depending on what is set up. It is strictly read-only and passes any change over to dvpod.
Needs: kubectl and helm access to a deployed DVpod.
Try:
Give me a health snapshot of my DVpod and explain any Charon errors from the last hour.
obol-monitoring​
This skill diagnoses clusters through Obol's hosted Grafana, using Prometheus metrics and Loki logs. It maps failure reasons to concrete fixes, and it can:
- run a first-pass triage of a cluster. This classes every peer as healthy, not sending metrics, degraded or offline, from what the other peers see, so a node that has been down for weeks stands out even though it sends nothing. It also gives the cluster's remaining fault tolerance, and which node reported each failure.
- rebuild a slot's consensus timeline from the logs to explain one failed duty.
- give a fleet overview: client and Charon version spread, the largest clusters by active stake, and the worst performers.
It names nodes by their operator nickname. For a peer that has been offline too long to have one in the metrics, it looks the operator up in the Obol API.
Needs: Python 3.6 or later and OBOL_GRAFANA_API_TOKEN (see Configuration). Your cluster must push metrics and logs to Obol. The skill doesn't work with a self-hosted Grafana.
Try:
Triage my Obol cluster "<cluster name>" on mainnet and explain why attestations are failing.
Do a performance check of our biggest clusters by active stake and list the easy wins.
run-obol-stack​
This skill helps you install, boot, and operate the Obol Stack. It covers prerequisites, obol stack up, model setup, syncing networks, and Cloudflare tunnels. To run your part of a DV in the Stack, it syncs a local Ethereum node, waits for obol network status to report synced, and then hands over to dvpod, create-cluster-invitation and test-a-dv-cluster. It then helps you turn your agent into a paid service with x402 payments and ERC-8004 registration. Once your agent is live in the Stack's dashboard, the agents inside the Stack take over with their own built-in skills.
Needs: Docker. Ollama, an Anthropic API key, or an OpenAI API key for model inference.
Try:
Install the Obol Stack on this machine and help me sell my first paid agent service.
Related docs: Obol Stack Quickstart, Build a Profitable Obol Stack.
distributed-validators​
This skill reads public data about DV clusters from the Obol API: cluster configuration, validator effectiveness and beacon chain states, operator details, exit progress, and who runs each Charon peer. It looks clusters up by lock hash, config hash, name or operator address, and it can't create clusters or submit exits. In Claude Code it uses the bundled Obol MCP server, falling back to curl.
It works as a Claude Code skill, and you can also add it to an Obol Stack agent with obol agent skills add distributed-validators (Obol Stack v0.16 and later).
Needs: no API key. uv for the bundled MCP server, otherwise curl.
Try:
How are the validators in my cluster with lock hash 0x... performing this week?
dv-steward​
This skill is for the Hermes agent inside the Obol Stack, not for Claude Code. It watches the Stack's own Ethereum node and dv-pod releases: sync, peers, Charon readiness, cluster peers, duties, DKG status, restarts and disk. It alerts you only when something needs you, for example on Telegram, on a schedule you set with obol hermes cron. It's strictly read-only: it never restarts anything or reads keys, and gives you the command to run instead.
Needs: Obol Stack v0.16 or later, a dv-pod installed with the chart's Obol Stack values (so Prometheus scrapes it and Hermes can read its status), and obol agent skills add dv-steward.
Related docs: Run Your Part of a DV.