菜单
Observability via loctl (promtool, grafanactl, logs). Use when querying metrics in Mimir, logs in Loki, or browsing Grafana dashboards.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Coordinate full-strength Claude and Codex teammates in detached tmux sessions with external stall monitoring, durable file channels, completion markers, and resumable conversations. Use when work can be divided into independent tasks, when the user asks for teammates or subagents, or when delegating research, implementation, frontend, review, or long-running work across model accounts.
Coordinate full-strength Claude and Codex teammates in detached tmux sessions with external stall monitoring, durable file channels, completion markers, and resumable conversations. Use when work can be divided into independent tasks, when the user asks for teammates or subagents, or when delegating research, implementation, frontend, review, or long-running work across model accounts.
Drive the kautopilot binary as a thin in-session controller: ticket/free-form request → merge-ready PRs across one or more repos, as ONE flat session. The binary owns all state, prompts, and next-step decisions; this skill just runs what `kautopilot next` yields. Use when running /kautopilot.
Drive the kautopilot binary as a thin in-session controller: ticket/free-form request → merge-ready PRs across one or more repos, as ONE flat session. The binary owns all state, prompts, and next-step decisions; this skill just runs what `kautopilot next` yields. Use when running /kautopilot.
Iteratively review-and-resolve the current changes using parallel subagent reviewers across multiple lenses (bugs, quality, completion), looping until every reviewer is satisfied. Use when running /floop, or when asked to deeply review and harden a diff until all critiques are addressed.
Give the user a fast, ADHD/dyslexia-friendly recap of the current work or conversation — the outcome first, then what (if anything) you need from them. Use when running /summary, or when asked to "summarize", "recap", "where are we", "catch me up", or "what do you need from me".
| name | liftoff-observability |
| description | Observability via loctl (promtool, grafanactl, logs). Use when querying metrics in Mimir, logs in Loki, or browsing Grafana dashboards. |
https://grafana-dash.qa.vungle.io (GRAFANA_TOKEN_QA), ops → https://grafana-prod.vungle.io (GRAFANA_TOKEN_OPS)http://loki.qa.vungle.io, ops → http://loki-prod.ops.vungle.io:30669http://mimir.qa.vungle.io/prometheus, ops → http://mimir.ops.vungle.io/prometheusThe authoritative tenant list is loctl <env> list tenants. The tables below add purpose annotations.
loctl qa)| Tenant | Description |
|---|---|
lokiTenantQA1 | QA Loki tenant |
loctl ops)| Tenant | Description |
|---|---|
dmx-logs | Primary tenant -- most Kubernetes pod logs and application logs across all clusters |
kafka | Kafka broker logs (GC, controller, etc.) |
gr-logs | Growth & revenue services (account, analyser, feedserver, keycloak, etc.) |
sdk-logs | Mobile SDK logs |
redis | Redis (redis-prod) logs |
rocksdb | RocksDB / KVRocks logs (ddlookup, hbp, loid, winazon, sit) |
albatross | Albatross service logs |
consul-server-internal | Consul server internal logs |
creative-dump-logs | Creative dump / ad manager logs |
data-all | Data platform logs |
haproxy | HAProxy load balancer logs |
hyperouter | Hyperouter / Loki source journal |
tailscale-router | Tailscale router and vault-agent logs |
valkey | Valkey (Redis-compatible) logs |
resources is allowed at the top level, and only resources <list|get|pull|serve|validate>; every other grafanactl command (push, delete, edit, config, ...) is blocked client-sidepush, test, tsdb blocked--help flag anywhere in the args bypasses these checksRead-oriented access to Grafana dashboards and resources. Requires service account tokens: GRAFANA_TOKEN_QA for qa, GRAFANA_TOKEN_OPS for ops.
loctl qa grafanactl resources get dashboards
loctl ops grafanactl resources get dashboards
loctl ops grafanactl resources get dashboards/<uid>
Server URL auto-injected. Use instant queries (no time args needed). Range queries require RFC3339 timestamps.
loctl qa promtool query instant 'up{cluster="eks-qa-us-east-1b"}'
loctl ops promtool query instant 'up{cluster="eks-ops-us-east-1c"}'
Fast native Loki HTTP client. No LOKI_ORG_ID env var needed — tenant is a positional argument. Replaces logcli for all log queries.
loctl ops logs dmx-logs labels
loctl ops logs dmx-logs labels service_name
loctl ops logs dmx-logs query '{service_name="<service>"}' --limit=10
loctl ops logs dmx-logs query '{service_name="<service>"} |= "error"' --limit=10
loctl ops logs dmx-logs query '{service_name="<service>"} |= "error"' --since=1h
loctl ops logs dmx-logs instant 'count_over_time({service_name="<service>"}[5m])'
loctl ops logs dmx-logs series '{namespace="<namespace>"}'
Flags: --limit=N (default 30), --since=DURATION, --from=TIME, --to=TIME, --direction=forward|backward, -o default|raw|jsonl, -q (quiet), -z TZ
--limit (start with 10)|= "error" (exact match, fast) before |~ "err.*" (regex, slow)--since or --from/--to for time ranges{cluster=~".+"} are very slow — use specific label values