菜单
Pull and inspect Aevatar mainnet/prod Kubernetes logs with the scoped kubeconfig at ~/Code/aelf-shared-k8s-prod.yaml. Use when debugging Aevatar online logs, Lark bot replies, NyxID relay callbacks, AgentRun/ConversationGAgent turns, Ornn skill publish behavior, approval hallucinations, webhook delivery, or console-backend stdout in prod.
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。
Use when product meaning is unclear or drifting across UI, copy, API contracts, data models, tests, docs, or user expectations. Trigger for questions like "why is this here?", "this feels unlike n8n/Linear/GitHub", "this field name is misleading", "optional control is too prominent", "backend concept leaked into UI", "should this be a trigger/input/prompt/setting?", "make the product semantics self-evolve", or any task that should update implementation plus tests/docs so the corrected product semantics persist across future changes.
Unattended three-phase refactor loop (analyze → implement → verify) driven by codex CLI in isolated git worktrees. Use when user wants fully autonomous parallel refactoring against CLAUDE.md violations, with /loop dynamic wakeups and per-cluster worktree merges.
Orchestrate a fully automated multi-agent frontend refactoring workflow to audit frontend code, detect backend contract impact, fix, review, browser-QA, and PR frontend architecture and design issues against CLAUDE.md and docs/canon frontend rules.
Inject NyxID auth session into the test browser's localStorage via refresh token. Solves automated browser testing login for apps using OAuth2/PKCE with localStorage-based session storage.
Write and edit Aevatar workflow YAML definitions. Covers canonical schema, closed_world_mode, formal roles config (provider/model/limits/event modules/routes/connectors), all 26 primitives, branching, retry/error policies, and validation constraints.
Unattended sequential implementation loop driven by codex CLI + Claude subagent reviewer. Walks every issue under a GitHub milestone in order; each issue gets implement (codex) → stacked PR → review (subagent) → fix (codex if needed) until reviewer passes; then advances to the next issue. Use when user wants the full milestone landed as a stacked PR train with /loop dynamic wakeups.
| name | aevatar-prod-logs |
| description | Pull and inspect Aevatar mainnet/prod Kubernetes logs with the scoped kubeconfig at ~/Code/aelf-shared-k8s-prod.yaml. Use when debugging Aevatar online logs, Lark bot replies, NyxID relay callbacks, AgentRun/ConversationGAgent turns, Ornn skill publish behavior, approval hallucinations, webhook delivery, or console-backend stdout in prod. |
Use this skill when the user asks to check Aevatar online/prod/mainnet logs. The kubeconfig is scoped and already exists at:
export KUBECONFIG=/Users/zhaoyiqi/Code/aelf-shared-k8s-prod.yaml
Do not print kubeconfig contents, tokens, or bearer headers. Do not modify ~/.kube/config. Default to read-only commands: get, describe, logs, and auth can-i. Do not exec, delete, apply, or restart prod workloads unless the user explicitly asks.
The token is namespace-scoped. Avoid cluster-wide discovery such as kubectl get pods --all-namespaces.
aismart-app-mainnet: Aevatar mainnet apps. aevatar-console-backend lives here.aismart-mainnet: shared mainnet infra.aismart-testnet: testnet apps.aismart-shared: shared namespace.Current mainnet backend target:
kubectl -n aismart-app-mainnet get pods -l app=aevatar-console-backend
kubectl -n aismart-app-mainnet logs -l app=aevatar-console-backend --tail=-1 --since=2h --timestamps --all-containers=true
Important: when using kubectl logs -l ..., include --tail=-1; otherwise kubectl may return only the last few lines per pod and hide the incident window.
For Lark/NyxID relay and AgentRun turns:
kubectl -n aismart-app-mainnet logs -l app=aevatar-console-backend --tail=-1 --since=2h --timestamps --all-containers=true \
| rg -i 'lark|nyxid-relay|channel-relay|ConversationGAgent|AgentRun|NeedsLlmReply|ReplyProduced|ReplyDispatched|LlmReplyDelivered'
For Ornn publish and approval checks:
kubectl -n aismart-app-mainnet logs -l app=aevatar-console-backend --tail=-1 --since=4h --timestamps --all-containers=true \
| rg -i 'ornn|ornn-api|proxy/s/ornn|api/v1/skills|skill-format|approval|RemoteToolApproval|ToolReceipt|tool_receipts|approval_required'
For one known run, actor, message, or correlation id:
kubectl -n aismart-app-mainnet logs -l app=aevatar-console-backend --tail=-1 --since=4h --timestamps --all-containers=true \
| rg '<run-id-or-message-id-or-actor-id-or-correlation-id>'
For previous container logs after a restart, resolve the pod first; --previous is pod-specific:
POD=$(kubectl -n aismart-app-mainnet get pod -l app=aevatar-console-backend -o jsonpath='{.items[0].metadata.name}')
kubectl -n aismart-app-mainnet logs "$POD" --previous --tail=500 --timestamps
Kubernetes timestamps are UTC. Convert to the user's local time when matching screenshots or chat times. If logs show the AgentRun replied but no matching outbound Ornn/NyxID HTTP request, say that plainly and distinguish it from "the tool succeeded." Do not infer approval or publish completion from LLM prose; rely on HTTP request logs and typed tool receipts.