Agent-Skills.md

Agent Skills: Grafana Logs & Metrics Skill

>

UncategorizedID: liuchengxu/dotfiles/grafana-logs

Repository

liuchengxu/dotfiles
liuchengxu
21391

Install this agent skill to your local

pnpm dlx add-skill https://github.com/liuchengxu/dotfiles/tree/HEAD/claude/skills/grafana-logs

Skill Files

Browse the full folder contents for grafana-logs.

Download Skill

Loading file tree…

claude/skills/grafana-logs/SKILL.md

Skill Metadata

Name
grafana-logs
Description
>

Grafana Logs & Metrics Skill

Datasource UIDs

Loki

| Environment | UID | Trigger keywords | |---|---|---| | Testnet | ae9b8e3z0u4u8b | testnet, TBV testnet | | Devnet | ddwaz0qqzgvswf | devnet, dev | | Mainnet | cdvc8k8n8737kd | mainnet, prod, production | | Tooling | cdvc8n7808feob | tooling, tools |

Prometheus

| Environment | UID | |---|---| | Testnet | ae9b890fmf9j4a | | Devnet | cdvc8q8szvev4d | | Mainnet | fdvc8owxsx3i8b | | Tooling | prometheus |

Infrastructure Types

TV / Hetzner infra (default)

  • Base selector: {infrastructure="hetzner"}
  • Add host label to filter by specific node
  • Discover available hosts via list_loki_label_values with labelName: "host"
  • Triggered when user mentions: testnet, devnet, vault provider, VP, VK, UC, or specific host names

Ubernets / Kubernetes infra

  • Labels: namespace, pod, container, node, job
  • Base selector: {namespace="...", container="..."}
  • Triggered when user mentions: ubernets, kubernetes, k8s, pods, namespace
  • Same Loki datasources, different label set

Noise Filter (Testnet/Devnet Default)

!= `eth_event_poller` != `getblockcount` != `getblockhash` != `getblock` != `rustls` != `reqwest` != `request for pegin` != `(will retry) `
  • Always apply unless user explicitly asks for unfiltered logs
  • If investigating a subsystem that would be excluded (e.g., reqwest issues), remove that specific exclusion

Query Workflow

  1. Identify environment from context -> resolve datasource UID
  2. Identify infra type (Hetzner vs Kubernetes) -> choose label set
  3. Discover hosts via list_loki_label_values if host not specified
  4. Default time range: 1 hour. Convert user-specified times to RFC3339.
  5. Build LogQL: base selector + user filters (|= "txid") + noise exclusions
  6. Execute via query_loki_logs with limit 50, direction backward
  7. Present results highlighting errors, extracting pegin_txid/claim_txid for drill-down

Common Query Templates

  • By pegin txid: {infrastructure="hetzner", host="..."} |= "<TXID>" <noise_filters>
  • Errors only: append |= "ERROR"
  • By daemon component: append |= "vault_provider" / |= "vault_keeper"
  • Pattern detection: use query_loki_patterns with base selector only (no noise filters)

Prometheus Metrics

All vaultd metrics are prefixed vaultd_* (defined in crates/vaultd/src/metrics.rs). Discover dynamically via list_prometheus_metric_names with filter vaultd_.