Skip to main content
Live. This area is documented as current, user-reliable behavior.

​
Goal

Give every agent you use, Stackie or your own, the same up-to-date StackShift know-how.

​
Prerequisites

  • A coding agent such as Claude Code, Codex, Cursor or VS Code, or Stackie in the dashboard
  • The StackShift CLI for a local install, or an agent connection for MCP

​
Workflow

1
Stackie already has the whole library; nothing to install.
2
For a coding agent, run stackshift setup agent --client claude-code (or codex, cursor, vscode, gemini-cli, windsurf, opencode) in your project, then restart the agent.
3
For an agent connected over MCP, it reads skills with the skill_read tool: no arguments for the entry skill, then the skill for its task.
4
Ask your agent to read the stackshift-cloud-operations skill first. It sends each task to the right focused skill.

​
What is in the library

  • Platform: deploying apps, fixing failed builds, debugging crashes, env vars and secrets, domains, previews, workers and cron, services and private networking, scaling and cost, databases and recovery, servers, Assets, Mail, mobile releases, Support, WordPress and Ghost, Shopify storefronts, billing and teams, Terraform and CI, agents and MCP.
  • Frameworks: JavaScript (Node, Next.js, Vite and static sites), Laravel, Python (Django, FastAPI, Flask), Go, Ruby on Rails, Rust, Spring Boot and ASP.NET.
  • Moving to StackShift from Heroku, Render, Vercel, Railway or Supabase.
  • Payments and messaging: Paystack, Flutterwave, M-Pesa Daraja, USSD and the WhatsApp Cloud API.
  • AI and engineering: building AI agents, RAG with pgvector, LLM evals, MCP servers, agent safety, running LLM apps in production, GPU inference basics, web testing with Playwright, security review, code review and API design.

​
Always current

Skills ship with StackShift itself. Every command, tool name and API path in them is checked against the platform automatically before each release, so a skill never tells your agent to run something that no longer exists. MCP agents and Stackie always read the version that matches the platform. A local install is a copy: run stackshift setup agent again after updating the CLI. Skills are guidance only. They never give an agent permissions it doesn’t already have; risky actions still need your approval.

​
Your own skills

Owners can add workspace skills in Settings → Stackie & agents → Skills: upload a SKILL.md or a zip, review every file, then approve it. Agents see approved workspace skills as workspace:NAME, marked as written by your workspace, and StackShift’s own skills and rules win where they disagree. Names starting with stackshift- are reserved.

​
Expected result

Your agent follows StackShift’s skills: it checks your real projects first, uses commands that exist, asks before risky changes and verifies the result.

​
Common failures

  • setup agent refuses because a skill folder differs from the bundled one: you or your agent edited it. Move or remove that folder, then run setup again.
  • An agent quotes an old command: it is reading a stale local copy. Update the CLI and run stackshift setup agent again, or let it read skills over MCP.

Connect your coding agent

Connect an external client, choose its access and inspect its work in StackShift.

StackShift AI agents

Stackie and five direct-entry specialists perform evidence-based work through durable runs, tenant-scoped tools, and approvals bound to the exact action and target.