AgentSkillsCN

Tundra Dome

Tundra Dome

SKILL.md

yaml-language-server: $schema=https://json.schemastore.org/yaml

name: tundra-dome description: Operate the Tundra Dome stack (Kafka/Airflow/Datadog) locally and in KIND; use when deploying/validating Tundra Dome, Kafka bridges, lane routing, or TD observability.

tundra-dome

Standardize bead/PR/issue flow across Gas Town → Tundra Dome.

KIND deploy on vibecode VM

  • •Manifests live at infra/tundra-dome/ (tundra-dome.clean.yaml, datadog.json).
  • •Bootstrap:
    bash
    kubectl create ns tundra-dome && kubectl create ns datadog
kubectl -n tundra-dome create secret generic tundra-dome-secrets \
  --from-literal=DD_API_KEY=$DD_API_KEY --dry-run=client -o yaml | kubectl apply -f -
kubectl -n datadog create secret generic tundra-dome-secrets \
  --from-literal=DD_API_KEY=$DD_API_KEY --dry-run=client -o yaml | kubectl apply -f -
kubectl apply -f infra/tundra-dome/tundra-dome.clean.yaml
kubectl get pods -n tundra-dome -w
  • •Expected pods: airflow-api-server, airflow-scheduler, airflow-dag-processor, kafka, td-event-emitter, tundra-observer, datadog-agent.

When to use

  • •You need to sync GitHub/Gitea issues → beads
  • •You need lane routing (critical/standard/experimental)
  • •You need KPI snapshots + Datadog dashboard/metrics validation
  • •You need Kafka/launchd health + topic hygiene
  • •You need quick Tundra Dome rollout checks across hosts

Prereqs (env + tools)

  • •bd, gt, kafka-topics, kafka-consumer-groups
  • •launchctl, kubectl, kind, docker (if k8s)
  • •datadog-agent CLI for check status (optional)
  • •GitHub dispatcher: GITHUB_OWNER, GITHUB_REPO, GITHUB_TOKEN

Quick actions (pick what matches the request)

  1. •Issue → bead sync (GitHub/Gitea)
  • •Dry-run first:
    • •Set GITHUB_DRY_RUN=true
    • •Run daemon/kafka-dsm/github-issue-dispatcher.sh
  • •Live:
    • •Set GITHUB_DRY_RUN=false
    • •Run daemon/kafka-dsm/github-issue-dispatcher.sh
  1. •Lane routing (labels → lanes)
  • •Map labels:
    • •critical → tundra-lane-critical-beads
    • •standard → tundra-lane-standard-beads
    • •experimental → tundra-lane-experimental-beads
  • •Update:
    • •airflow/dags/tundra_gitea_cicd.py
    • •daemon/kafka-dsm/github-issue-dispatcher.js
  1. •KPI snapshot + DD checks
  • •Snapshot:
    • •python3 daemon/kpi_snapshot.py > logs/kpi_snapshot.json
  • •Verify core counters:
    • •tundra.td_event_emitter.processed
    • •tundra.td_event_emitter.heartbeat
    • •tundra.observer.processed
    • •gastown.kafka_consumer.processed
  1. •Kafka health + topic hygiene
  • •TD helpers (preferred):
    • •td kafka summary (works; outputs lane counters)
    • •td kafka topics (works; lists tundra-* topics)
    • •td kafka status (currently exits 1 in local rig; use topics/summary instead)
  • •List topics:
    • •kafka-topics --bootstrap-server localhost:9092 --list | sort
  • •Delete dot topics:
    • •kafka-topics --bootstrap-server localhost:9092 --delete --topic <topic.with.dots>
  • •Verify consumer groups:
    • •kafka-consumer-groups --bootstrap-server localhost:9092 --list
    • •kafka-consumer-groups --bootstrap-server localhost:9092 --describe --group <group>
  1. •Launchd service sweep
  • •Restart a service:
    • •launchctl kickstart -k gui/$(id -u)/<label>
  • •Inspect logs:
    • •tail -n 200 logs/<service>.err.log
  1. •Superdome sling (Tundra)
  • •Use td-sling to fan out lane + audit + work-intake + in-progress topics:
    • •daemon/kafka-dsm/td-sling.sh sling <bead-id> <target> --lane standard --message "..."

Default topic set (add if missing)

  • •Commands: tundra-mayor-commands, tundra-deacon-commands, tundra-polecat-commands
  • •Comms: tundra-nudges, tundra-whispers, tundra-mail-outbox, tundra-mail-inbox
  • •Beads: tundra-beads-created, tundra-beads-in-progress, tundra-beads-completed, tundra-beads-escalated, tundra-beads-failed
  • •Lanes: tundra-lane-critical-beads, tundra-lane-standard-beads, tundra-lane-experimental-beads
  1. •OpenCode + OpenRouter Free Models (Dynamic Selection)
  • •Auto-select best model (tests availability first):
    • •td opencode run "Explain this error" → tests models, picks first working
    • •td opencode spawn polecat-1 → spawns with dynamic model selection
  • •Manual model testing:
    • •td opencode test-models
  • •List all free models:
    • •td opencode models
  • •Direct with specific model:
    • •opencode-auto → dynamic selection wrapper
    • •gt sling <bead> --agent opencode-auto → uses dynamic selection

OpenCode Setup (v1.1.48+)

  1. •

    Install:

    bash
    curl -fsSL https://opencode.ai/install | bash
# Binary: ~/.opencode/bin/opencode
  2. •

    Config (~/.config/opencode/opencode.json):

    json
    {
  "$schema": "https://opencode.ai/config.json",
  "provider": {
    "openrouter": {
      "options": {
        "apiKey": "sk-or-v1-..."
      }
    }
  }
}
  3. •

    Working free models (with tool use):

    • •meta-llama/llama-3.3-70b-instruct:free - Best for coding
    • •google/gemma-3-27b-it:free - Good general
    • •tngtech/tng-r1t-chimera:free - Reasoning
  4. •

    Models WITHOUT tool use (don't work with OpenCode):

    • •liquid/lfm-*:free
    • •allenai/molmo-*:free
  5. •

    Free tier limits:

    • •50 req/day (or 1000/day with $10+ credits)
    • •Models change availability throughout day - ALWAYS test first

Notes

  • •Prefer hyphenated topic names (no dots).
  • •Keep td-event-emitter as the health-ping owner.
  • •Test free models before spawning polecats: td opencode test-models