菜单
Control the Agent Ops Center dashboard — open/show it, restart the bridge, reset the agent tiles, or check status. Use when the user asks to open/show the agent dashboard, start or stop the visualizer, or clear the agents.
默认使用会先检查来源的 Prompt；你也可以切换为直接命令，或下载本地副本。
决定是否安装前，请先阅读 SKILL.md，以及 SkillsMP 当前展示的配套文件。
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。
直接命令不会经过审查 Prompt；运行前请先检查来源。
npx skills add https://github.com/weellio/gander --skill agent-ops命令会保持在同一行。复制前请横向滚动并检查完整内容。
想先保存到本地？可下载 SkillsMP 当前能够提供的文件。
Autonomously execute a multi-step task — plan it, spawn sub-agents per task, and run to completion without asking for confirmation between steps. Use when the user says "build/do X autonomously", "keep going until done", or "don't ask me, just finish it".
Use when the user wants to create a new Claude Code agent, skill, or slash command — "build me an agent that…", "make a skill that…", "create a command for…". Generates correct YAML front matter + body following best practices and writes the file to the right .claude location (one project, several, or global). Pairs with the Gander dashboard's New-component builder.
Use when the user wants to cut Claude Code token cost or asks "why is this session so expensive / what's eating my tokens / trim my CLAUDE.md / reduce context". Audits the always-on context a project re-sends every turn (CLAUDE.md, memory, MCP servers, skills, agents), ranks it by per-turn token cost, and proposes concrete trims — including a leaner CLAUDE.md rewrite. Pairs with the Gander dashboard's per-session cache-hit / context-fill gauges.
基于 SOC 职业分类
正在显示 SKILL.md
| name | agent-ops |
| description | Control the Agent Ops Center dashboard — open/show it, restart the bridge, reset the agent tiles, or check status. Use when the user asks to open/show the agent dashboard, start or stop the visualizer, or clear the agents. |
Agent Ops Center is a live dashboard that visualizes this session's Claude Code agents and sub-agents as animated avatar tiles. A small local bridge server holds the agent registry and serves the dashboard at http://localhost:3131/. Claude Code hooks push real tool activity to it automatically, so normally it "just works" once the plugin is installed — this skill is for manual control.
The bridge + dashboard live under the plugin root (${CLAUDE_PLUGIN_ROOT} in hook/MCP contexts). When running commands from this skill, use the plugin's bridge/ and dashboard/ folders.
Run the launcher. It starts the bridge if it isn't already running and opens the dashboard once:
node "<plugin-root>/bridge/launch.js"
If the browser tab doesn't open, navigate to http://localhost:3131/ manually.
A quick request to the state endpoint tells you if the bridge is up and how many agents are tracked:
Invoke-RestMethod http://localhost:3131/api/statecurl -s http://localhost:3131/api/stateClears all agents back to just the root Orchestrator:
Invoke-RestMethod -Method Post http://localhost:3131/api/resetcurl -s -X POST http://localhost:3131/api/resetStop the Node process serving port 3131:
Get-NetTCPConnection -LocalPort 3131 | Select-Object -Expand OwningProcess | ForEach-Object { Stop-Process -Id $_ -Force }lsof -ti:3131 | xargs killYou can also pipe a headless run straight into the bridge:
claude -p "task" --output-format stream-json --verbose | node "<plugin-root>/bridge/server.js" --stdin
3131; override with the AOC_PORT environment variable (and update the hook URLs in hooks/hooks.json to match).dashboard/assets/avatars/ and list them in dashboard/assets/avatars/manifest.json.