一键导入
go-orca-offload
Offload expensive agent work to a go-orca cluster via the mcp-orca MCP bridge (workflows and ad-hoc persona runs).
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。
菜单
Offload expensive agent work to a go-orca cluster via the mcp-orca MCP bridge (workflows and ad-hoc persona runs).
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 this skill when implementing code for any module in this repository.
How to configure, connect to, and use MCP (Model Context Protocol) servers in go-orca workflows.
Use when implementing or reviewing Next.js App Router applications in go-orca workflows.
Model routing and structured-output rules for go-orca workflow personas (Director, Project Manager, Architect).
Techniques for writing clear, compelling technical blog posts and documentation.
Checklists and patterns for systematic QA review of code, APIs, and infrastructure changes.
基于 SOC 职业分类
| name | go-orca-offload |
| description | Offload expensive agent work to a go-orca cluster via the mcp-orca MCP bridge (workflows and ad-hoc persona runs). |
Use when a local agent (Cursor, VS Code, Claude Desktop) should delegate implementation, QA, or full pipelines to a go-orca cluster.
Point your MCP client at the cluster bridge:
{
"mcpServers": {
"go-orca": {
"url": "http://mcp-orca:3000/mcp",
"headers": {
"Authorization": "Bearer <zitadel-or-authentik-access-token-or-pat>"
}
}
}
}
On the bridge pod, set GOORCA_MCP_API_KEY to the same Bearer token (proxied to go-orca-api), or rely on MCP client headers when your client forwards them.
Set GOORCA_TENANT_ID and GOORCA_SCOPE_ID on the bridge pod, or pass tenant_id / scope_id per tool call.
server:
oidc:
userinfo_url: "https://auth.example.com/oidc/v1/userinfo" # Zitadel
required: true
Authentik: https://<host>/application/o/<slug>/userinfo/
| Tool | Use when |
|---|---|
orca_workflow_create | Full pipeline: PM → Architect → Pod → QA → Finalizer |
orca_workflow_status / orca_workflow_events | Poll progress |
orca_persona_run | One persona, one shot (no full workflow) |
orca_task_run | Single Pod or QA task (convenience wrapper) |
Read MCP prompts: offload-workflow, offload-persona, orca-capabilities.
Pass template_id on workflow create, e.g. software-default, content-default, ops-default. Templates pin required_personas, models, and pod specialties.
full — full tool registry on Podmcp_agent — invoke_mcp_agent only (auxiliary / small model; good for MCP toolchain calls)none — no toolsHomelab may run ollama-gpu (primary) and ollama-cpu (auxiliary). PM and MCP-agent work can route to CPU when GPU is saturated. See workflow-orchestration for persona model rules.
POST /api/v1/workflows — full workflowPOST /api/v1/persona-runs — ad-hoc personaGET /api/v1/workflow-templates — template catalogRequires X-Tenant-ID, X-Scope-ID, and Authorization: Bearer when server.oidc.required is true (validated via userinfo).